University of Southern California Dissertations and Theses

Dynamic radio resource management for 2G and 3G wireless systems

(USC Thesis Other)

Dynamic radio resource management for 2G and 3G wireless systems

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content DYNAMIC RADIO RESOURCE MANAGEMENT FOR 2G AND 3G WIRELESS SYSTEMS Copyright 2002 by Huan Chen A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements of the Degree DOCTOR OF PHILOSOPHY (ELECTRICAL ENGINEERING) December 2002 Huan Chen Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3093745 Copyright 2002 by Chen, Huan All rights reserved. ® UMI UMI Microform 3093745 Copyright 2003 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90089-1695 This dissertation, written try Huan Chen under the direction o f h dissertation committee, and approved by all its members, has been presented to and accepted by the Director o f Graduate and Professional Programs, in partial fulfillment o f the requirements fo r the degree of DOCTOR OF PHILOSOPHY fry /* Director Date December 18. 2002 Dissertation Committee Chair Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D edication Dedicated with love to my family and my friends. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A cknow ledgm ents I would like to express my deepest gratitude to my advisor, Prof. C.-C. Jay Kuo. He brings me into the exciting field of wireless networks. His constant instruction and encouragement make it possible for me to have the quality study environment and learning experience. Not only do I obtain abundant technological knowledge, but I also learn a lot from his endless energy and solid attitude of doing research. The guidance under Prof. Kuo in these years helps me not only have the ability to solve challenging problems, but also have faith and confidence in ourselves. I am grateful to Prof. Daniel C. Lee and Prof. Roger Zimmermann for finding valuable time and efforts to serve on both my qualifying examination and dissertation committee. Their rich research knowledge and experience significantly improve the quality of the thesis. Also, I would like to thank Prof. Alexander A. Sawchuk and Prof. Shrikanth S. Narayanan for serving my qualifying examination and for their constructive comments and suggestions. Special thanks are extended to Dr. Sunil Kumar. As a mentor and good friend, I work with him closely. His expertise and valuable suggestion in this field deserves a lot of credit. I thank all my colleagues in Prof. Kuo’s research group. Working iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. with them broadens my scope of research and makes my staying at the University of Southern California valuable and interesting. Finally, there is nothing enough for me to show my appreciation to my family. W ithout their warm supporting, I could never fulfill my dream here. It is their continuous support that makes me achieve one of my most important milestones. Thank you all very much. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C ontents D edication ii A cknow ledgm ents iii List of Tables ix List of Figures x A bstract xiii 1 Introduction 1 1.1 Significance of the Research ! ..................................................................... 1 1.2 Issues on Resource Management in Multimedia Wireless Systems . . . 5 1.2.1 QoS for Multimedia A p p lic atio n s................................................. 5 1.2.2 Issues in Cellular S y s te m s .............................................................. 8 1.3 Contribution of the R esearch........................................................................ 10 1.4 Outline of D issertatio n.................................................................................. 13 2 Background 14 2.1 Evolution of Cellular C om m unication........................................................ 14 2.1.1 The First Generation (1G) S y stem s............................................. 14 2.1.2 The Second Generation (2G) System s.......................................... 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.1.3 The Third Generation (3G) Systems .......................................... 17 2.2 Challenges of Radio Resource M anagem ent.............................................. 18 2.2.1 Handoff and M obility........................................................................ 19 2.2.2 Channel Assignment and R eservation........................................... 21 2.3 QoS Guarantees and Preferential T re a tm e n t........................................... 22 2.3.1 QoS G uarantees.................................................................................. 22 2.3.2 Preferential T re a tm e n t..................................................................... 24 3 D ynam ic R esource M anagem ent for 2G T D M A /F D M A System s 27 3.1 Introduction..................................................................................................... 27 3.2 Fixed Guard Channel Scheme Extension with Multiple Thresholds . . 29 3.2.1 Service M o d e l..................................................................................... 30 3.2.2 Proposed Multiple Thresholds GC Scheme................................... 31 3.2.3 Analytical M o d e l.............................................................................. 32 3.2.4 Simulation R e s u lts ........................................................................... 36 3.2.4.1 The effect on system utilization and handoff blocking rate with priority h a n d o ff.............................................. 38 3.2.4.2 The effect on system utilization and handoff block ing rate with priority handoff and differentiated QoS classes.................................................................................. 39 3.2.4.3 Comparison of the handoff blocking rate between high and low priority service classes............................. 40 3.3 Dynamic Guard Channel S c h e m e ............................................................. 41 3.3.1 Mobile Simulation S y s te m ............................................................. 42 3.3.1.1 Message loops in the mobile simulation system . . . 42 3.3.1.2 Use of SNR and distance inform ation.......................... 45 3.3.2 Service Model and Traffic Profile ................................................ 47 vi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.2.1 Service model with multiple QoS c la s s e s ..................... 47 3.3.3 Proposed Dynamic GC S c h e m e .................................................... 49 3.3.3.1 Proposed resource reservation estim ation..................... 49 3.3.3.2 Proposed call admission control (CAC) algorithm . 53 3.3.4 Simulation Results-Scenario I ....................................................... 55 3.3.4.1 System and service model param eters........................... 55 3.3.4.2 Performance comparison under traffic with multi level priority but without rate adaptive character istics (without mobility differentiation)....................... 58 3.3.4.3 Performance comparison under traffic with muti-level priority classes and rate adaptive characteristics ( without mobility differentiation) ................................. 59 ■3.3.4.4 The performance comparison using time-aware weight ing factor in the presence of mobility differentiation . 60 ■ 3.3.5 Simulation Results-Scenario I I ........................................................ 62 3.3.5.1 System and service model param eters........................... 62 3.3.5.2 Performance comparison between proposed dynamic scheme and fixed GC s c h e m e ....................................... 63 3.3.5.3 Performance comparison for rate adaptive and non rate adaptive s y ste m ........................................................ 63 3.3.5.4 Performance comparison among different priority classes in proposed service model using dynamic scheme . . 64 3.3.6 Conclusion and Future W ork........................................................... 66 4 Radio R esource M anagem ent in Code D ivision M ultiple A ccess (C D M A ) System 67 4.1 Introduction ...................................................................................................... 67 4.2 Priority Handoff with Guard C h a n n e ls ..................................................... 71 4.3 Capacity and Load Estimation in CDMA S y stem s................................. 75 vii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.4 Preferential Treatment with I G M .............................................................. 79 4.4.1 Service M o d el...................................................................................... 79 4.4.2 Interference Guard Margin (IG M ).................................................. 81 4.4.3 Dynamic Resource-Reservation Estimation (R R E ).................... 83 4.4.4 Call Admission Control (CAC) Algorithm ................................ 86 4.5 Simulation R e s u lts ........................................................................................ 90 4.5.1 System Model and Link C haracteristics....................................... 90 4.5.2 Non-rate Adaptive T ra ffic .............................................................. 91 4.5.3 Rate Adaptive Traffic....................................................................... 92 4.5.4 System U tiliz a tio n ............................................................................ 99 4.6 Conclusion and future w o r k .......................................................................... 100 5 Optim al Radio R esource M anagem ent w ith M arkov D ecision Pro cess 102 ■ 5 .1 Introduction....................................................................................................... 102 5.2 System Capacity in Interference-limited S ystem s....................................... 109 5.3 MDP and Optimal CAC P o lic y ................................................................... 112 5.3.1 Markov Decision Process M o d e l...................................................... 113 5.3.2 Uniformization T e c h n iq u e ................................................................ 116 5.4 Solution via Linear P rogram m ing ................................................................118 5.5 2-D MDP Optimal CAC Policy in Homogeneous Handoff Systems . . 121 5.5.1 System M o d el.......................................................................................122 5.5.2 Modified Linear Program m ing......................................................... 125 5.5.3 Numerical Results for CAC Policy T r e n d ..................................... 126 5.5.3.1 Light Traffic Scenario ....................................................... 126 5.5.3.2 Heavy Traffic S cen ario....................................................... 128 viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.5.3.3 Discussion on Relationship between CAC Policies and Traffic Conditions................................................................. 130 5.5.4 Performance Com parison....................................................................134 5.5.4.1 Traffic P a ra m e te rs ...............................................................134 5.6 3-D MDP and Optimal CAC P o licy ............................................................. 138 5.6.1 System M o d el........................................................................................138 5.6.2 Solution via Linear P rogram m ing.................................................... 142 5.6.3 Numerical Results .............................................................................. 144 5.6.3.1 Traffic P a ra m e te rs .................................................... ... . 144 5.6.3.2 OPNET Im plem entation.....................................................145 5.6.3.3 Performance Comparison ................................................. 147 5.7 Complexity of the Markov Decision Process using Linear Program ming A p p ro a c h ................................................................................................. 152 5.7.1 Complexity Issue for Linear P ro g ram m in g .....................................153 5.7.2 Derivation of the Problem Size ......................................................... 157 5.7.2.1 The 2D C a s e ....................................................................... 157 5.7.2.2 The 3D C a s e ....................................................................... 159 5.7.2.3 Complexity Discussion for Higher Dimension................ 161 5.8 Conclusion and Future W ork ............................................................................161 6 Conclusion and Future Work 163 6.1 Summary of the Research ...............................................................................163 6.2 Future W ork.........................................................................................................166 6.2.1 Embedded System Design for Resource Management .................167 R eference List 170 ix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. L ist of T ables 1.1 Characteristics of multimedia applications................................................. 6 3.1 Traffic Test S e ts ................................................................................................ 38 3.2 Message in New Call Request Message L o o p ............................................ 43 3.3 Message in Update Handoff Candidate Message L o o p ........................... 44 3.4 Message in Handoff Request Message L o o p ............................................. 45 3.5 Message in Call Terminate Message L o o p .................................................. 46 ■ 3 .6 Parameters used in call admission control alg o rith m ................................ 55 ■ 5 .1 Complexity for Solving L P ...............................................................................157 5.2 Total number states and decision variables in a 2-D MDP model . . . 158 5.3 Total number states and decision variables in a 3-D MDP model . . . 161 x Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. L ist of F igures 2.1 The wireless system evolution from 1G to 3G............................................ 17 3.1 Illustration of multiple thresholds applied to different priority classes. 32 •3.2 The decision rule for call admission control........................................... 33 3.3 The queueing model for the multiple-threshold GC scheme.............. 34 3.4 (a) System utilization without (test set 1) and with (test set 2) priority handoff mechanism, (b) Handoff blocking rate without (test set 1) and with (test set 2) priority handoff mechanism............................................. 39 3.5 (a) System utilization without (test set 3) and with (test set 4) priority handoff and QoS classes, (b) Handoff blocking rate without (test set 3) and with (test set 4) priority handoff and QoS classes....................... 40 3.6 Handoff blocking r a t e s ................................................................................. 41 3.7 The new call request message loop in the simulation system.............. 43 3.8 The update handoff candidate message loop in the simulation system. 44 3.9 The handoff request message loop in the simulation system............... 45 3.10 The call terminate message loop in the simulation system................ 46 3.11 The non-linear weighting curve for resource reservation estimation. . . 50 3.12 Illustration of set S ......................................................................................... 52 3.13 The mobile simulation system with seven cells, 140 mobile termi nals (MT), 7 base stations (BS) and one main traffic switching office (MTSO)............................................................................................................... 56 3.14 Performance comparison under traffic with multi-level priority but without rate adaptive characteristics (without mobility differentia tion) for different schemes under (a) light traffic loading (A = 3) and (b)heavy traffic loading (A = 10)................................................................... 59 xi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.15 Performance comparison under traffic with muti-level priority classes arid rate adaptive characteristics ( without mobility differentiation for different schemes with rate adaptive abilities for all mobiles, under (a) light traffic loading (A = 3) and (b)heavy traffic loading (A = 10) . . . 60 3.16 The performance comparison by using and not using time-aware weight ing factor, in the presence of mobility differentiation with fixed-rate traffic class for all mobiles,under (a) light traffic loading (A = 3) and (b)lieavy traffic loading (A = 1 0 )................................................................. 61 •3.17 The performance comparison by using and not using time-aware weight ing factor, in the presence of mobility differentiation with rate adap tive traffic class for all mobiles,under (a) light traffic loading (A = 3) and (b)heavy traffic loading (A — 1 0 ) ....................................................... 62 3.18 System performance comparison between proposed dynamic scheme and fixed GC scheme in (a) rate adaptive and (b) non-rate adaptive system................................................................................................................. 64 3.19 System performance comparison among different priority classes in the (a) rate adaptive and (b) non-rate adaptive system.............................. 65 •3.20 (a)System reward for rate adaptive and non-rate adaptive system, (b) System utilization for rate adaptive and non-rate adaptive system under traffic condition A = 10, I — 2 0 ........................................................ 65 4.1 The state diagram of a M /M /m /m queue with a total of m channels and g guard channels. .................................................................................. 72 4.2 The load curve and the load estimation....................................................... 82 4.3 The load increment Ap, for different source rates R t and different media activities iy............................................................................................. 82 4.4 Set S (j) in resource-reservation estimation: (a) 11(f) > II(j) and (b)A(j)* € A(i).................................................................................................. 87 4.5 The proposed call admission control algorithm.......................................... 88 4.6 Performance comparison for non-rate adaptive users under light to moderate traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph.................................. 93 4.7 Performance comparison for non-rate adaptive users under moderate to heavy traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph.................................. 94 xii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.8 Performance comparison for rate adaptive users under light to moder ate traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph................................................... 96 4.9 Performance comparison for rate adaptive users under moderate to heavy traffic load: (a) the cost function J and (b) the new call block ing rate Pn and the handoff dropping rate Ph............................................ 97 4.10 Performance comparison between rate adaptive (RA) and non-rate adaptive (Non-RA) schemes for the cost function J ................................. 98 4.11 Comparison of system utilization w.r.t. rate-adaptive and non-rate- adaptive traffic under the heavy load with A = 5........................................ 101 5.1 Optimal CAC policies in the light traffic scenario.......................................127 5.2 Optimal CAC policies in the heavy traffic scenario.....................................129 5.3 Optimal CAC policies for (a) (± h = 4, (b) fih = 3, (c) j±h = 2, (d) Hh — 1, with the total system capacity equal to C — 10 unit bandwidth, \ i = A /,. = 4 and /j,n — 1.....................................................................................131 5.4 Optimal CAC policy under heavy traffic with different new call service rates /rn: (a) fin — 4 and (b) fin — 3, where the total system capacity C = 10, An = Xh — 4 and fih — 1 are fixed.).................................................133 5.5 Implementation steps for the MDP-LP CAC policy................................... 134 5.6 The MDP CAC policy for the 2D Case.........................................................136 5.7 The performace comparison by weighted reward (a) MDP-LP, GC(0), GC(1), GC(2) and GC(3) schemes, (b) a close look at the performance among MDP-LP, GC(1) and GC(2) schemes................................................ 137 5.8 (a) The new call blocking probabilities for various schemes, and (b) handoff dropping probabilities for various schemes......................................138 5.9 Illustration of the transition probability P (y |x , a ).....................................142 5.10 The OPNET simulation system......................................................................145 ■ 5 .1 1 The OPNET simulated processes................................................................... 148 5.12 The MDP CAC policy.......................................................................................150 5.13 The average reward function........................................................................... 151 xiii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5/14 The blocking probabilities for (a) the complete sharing scheme, (b) the GC(1,3) guard channel scheme and (c) the MDP stationary policy. 151 5.15 The computation of the total number of states Nx in a 2-D MDP model. 158 5.16 The total number of states lY x in a 3-D MDP m odel................................160 6.1 I/O of embedded CAC systems for (a) 2-D and (b) 3-D cases................. 169 xiv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A bstract The design of efficient radio resource management (RRM) schemes, including priority handoff and call admission control, for 2G and 3G cellular networks is studied in this research. The RRM module is responsible for efficient utilization of the system resource. It is also critical to the Quality of Service (QoS) provision to mobile users with various QoS requirements. In this research, we have addressed the problem of efficient resource management in order to provide Quality of Service (QoS) and preferential treatment to traffic with different QoS requirements. The first part of this research develops an efficient dynamic resource management for channel-based TDMA/FDMA systems. The second part of the research examines the dynamic call admission control system. The proposed scheme is suitable for interference- based CDMA systems via the use of interference guard margin (IGM). The QoS performance of proposed RRM schemes in terms of a set of system defined objective functions is evaluated with respect to heterogeneous traffic scenarios. The third part of this research focuses on the development of the optimal stationary call admission control (CAC) scheme based on the Markov decision process (MDP) model and the linear programming (LP) technique under homogeneous as well as hybrid handoff xv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. processes. The proposed scheme is suitable for both channel-based systems as well as interference-based systems. xvi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C hapter 1 Introduction 1.1 Significance of the R esearch W ith the advance of radio and network technologies, the mobile system has evolved from preliminary communication devices to a ubiquitous personal communication service (PCS) system that provides users new channel capacity and features, such as data, voice and multimedia applications and services with flexible quality of service (QoS) requirements at any time, anywhere, and even in any format [1, 2], The PCS system development has come to an era of the convergence of information and telecommunication technologies. During the evolution of wireless systems in the past, the development of wireless systems has been struggling with different fundamental key problems. The system was pushed a leap forward whenever each key problem was solved. Zander [2 ] pointed out three key bottlenecks in the past. 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • Path loss problem In early days, the receivers of the communication system are passive devices, which cannot compensate the signal power lost in the transmission path. The consequence of such a design was that all the power needed in the receiver should be generated in the transmitter. Thus, the path loss problem greatly hinders the development of a wireless system due to the use of the passive receiver. The advent of the electron tube amplifier (ETA) solved this problem. As the technology evolved, ETA has been replaced by integrated circuits (IC) nowadays. A mobile receiver equipped with amplifier IC can greatly compen sate the signal power loss in the transmission path. • Therm al noise The problem of the existence of thermal noise lies in the fact that whenever the signal is amplified, the noise is amplified as well. This problem has been well solved due to the breakthrough of digital communication theory pioneered by Shannon [3 ] and further refined by numerous researchers. In past decades, the large scale integrated circuits (LSIC) and the digital signal processing (DSP) technologies have made it possible to push the performance of wireless systems to close to Shannon’ s theoretical limits. The main technology components include channel coding schemes, advanced compression and communication schemes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • Lim ited spectrum When the population of wireless users increases, multiple concurrent users struggle against the inevitable interference due to resource sharing. To re duce the interference from other users, some administrative solution should be involved. The International Telecommunication Union (ITU) [4], founded in 1865 and operating as a United Nations agency in 1947, is responsible for the development of international treaties, regulations and standards in the area of telecommunications. Spectrum utilization can be more efficient through reg ulation. However, only administrative regulation is not enough. To increase spectrum efficiency technically, the cellular concept was originated at the Bell Lab in 1947 [5 ] and the first cellular system was started in 1983. A cellular system uses many small radio coverage areas called cells to serve hundreds of mobile users. Such a cellular concept allows a frequency to be re-used by mobile users in different cells that are far apart. The rationale behind the cellular concept is that djacent cells should use different frequencies to avoid interference while widely separated cells can reuse the same frequencies [6]. The cellular concept has removed the bottleneck of the limited spectrum and greatly increases spectrum efficiency. However, the cellular architecture bring another challenge; that is, how to provide mobile users with continuous con nection when they move from one cell to another (i.e. handoff). The handoff issue is a crucial problem especially in the provision of QoS guarantees. 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In addition to the handoff issue aforementioned, the success of the Internet tech nology leads to the shift of a new service paradigm. Due to the great success and the enormous impact of IP networks, Internet access (such as sending and receiving e-mails) and web browsing have become the ruling paradigm for next generation wireless systems. The service paradigm has shifted from the conventional voice ser vice to seamlessly integrated high quality multimedia transmission over broadband networks. The multimedia content may include data, voice, audio, image, video and so 0 1 1. Furthermore, new wireless services and applications are enabled by mobile Internet appliances such as the next generation cellular phone, the personal digital assistant (PDA) and portable computers (or devices). The accessibility to core net works from a mobile terminal is easier and more diversified. The number of mobile users and the demand on multimedia information by mobile users increase rapidly. As a result, there is an urgent need to design a proper system that supports het erogeneous multimedia services and seamlessly access to the desired resource via wireless connections. One key issue in providing multimedia services over a mobile wireless network is the quality of service (QoS) support in the presence of changing network connectivity, which is the consequence of the cellular design. It is im portant to provide seamless service transition when a mobile user moves from one cell to another, i.e. when the handover process occurs. The solution to this challenge is the proper management Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of scarce resources to satisfy both the service provider and mobile users. Sometimes, these two goals are conflicting to each other so that some trade-off should be made. In this research, we address the problem of providing QoS to multimedia ap plications, and propose some solutions via the design of dynamic radio resource management (RRM) schemes. Both the resource reservation estimation (RRE) and the call admission control (CAC) are examined under a unified RRM framework to support multimedia traffics with various QoS requirements such as priority and rate adaptivity. Our research work covers RRM design for the channel-based wireless system, such as the Time Division Multiple Access (TDMA) and the Frequency Di vision Multiple Access (FDMA) systems, as well as the interference-based wireless system, such as the Code Division Multiple Access (CDMA) system. In the next section, we briefly discuss issues that are critical to the RRM design to support multimedia over a wireless cellular system. 1.2 Issues on Resource M anagem ent in M ulti- m edia W ireless System s 1.2.1 QoS for M u ltim ed ia A p p lication s According to the delay constraint, multimedia applications can be classified into two types, i.e. real-time and non-real-time applications [7]. Real-time applications tolerate little delay and/or delay variation. Most interactive applications, such as 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. video conferencing and telephone calls, have a stringent time constraint. In contrast, non-real-time applications can tolerate a relatively larger amount of delay and/or delay jitter. For example, many conventional data or message applications, such as Telnet, FTP, e-mail applications, belong to this type. These applications are also called elastic since they are able to stretch gracefully when delay increases (note that they still benefit from a shorter amount of delay). The delay requirements of elastic applications vary from interactive applications like Telnet, to more asynchronous ones like email, with interactive bulk transfers like FTP in the middle. In Table 1.1, we show characteristics of several important multimedia applications. Table 1.1: Characteristics of multimedia applications. Multimedia Applications Realtime Interactive Loss tolerance Video Conference Yes Yes Tolerable Distance Learning Yes No Tolerable Streaming Audio/Video Yes No Tolerable Multiuser Gaming Yes Possibly Intolerable for control Em ail/FTP No No Intolerable Online Banking No No Intolerable In the presence of multiple QoS requirements for different multimedia traffic, the key problem in the design of a multimedia wireless system is to balance the two opposing objectives of the system operator (the service provider) and mobile users (customers). The former wants to achieve high system utilization so that more users can be accommodated by the system and more revenue can thus be obtained while the later wants to receive better quality of service (QoS). Consequently, a user 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. would like to request the system to reserve enough resources for him/her to use. In a wireless telephony system, user’ s QoS requirements can be expressed objectively in terms of probabilistic metrics such as the probability of call blocking. Squeezing more users into the system inevitably leads to more resource competing and heavier interference in the system. This results in poor QoS guarantee to users. Striking a proper balance between system utilization and user’ s QoS satisfaction is the focus of our research. To meet the large bandwidth requirement of multimedia traffic, it is important to utilize the system resource efficiently and provide preferential treatment according to mobile user’s traffic profile when the system is congested. The Radio Resource Management (RRM) module in the cellular network system is responsible for the management of air interface resources. RRM is needed to offer efficient system utilization and guarantee a certain QoS level to different users according to their traffic profiles. The call admission control (CAC) mechanism is one of the most important components of RRM that affects the resource utilization efficiency and QoS guarantees provided to users. The radio resource reservation estimation (RRE) mechanism helps CAC to decide how much resource is needed to be reserved in order to provide QoS guarantees to mobile users. The RRE module residing in each base station dynamically estimates the amount of resource to be reserved by referencing traffic conditions in neighboring cells periodically or upon the call request arrival depending on the design of the system. 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.2.2 Issu es in C ellular S ystem s The cellular network has several distinct characteristics as detailed below. • Cellular m obility and handoff A cellular network employs a central switching office, i.e. the Mobile Switching Office (MTSO) to interconnect small radio coverage areas into a larger system [6], In order to maintain the service of a call, it is important to properly manage the resource along the path of mobile terminals (MT), especially when they move from one cell to another. Since each base station (BS) typically serves a limited number of users, when a BS has reached its capacity (i.e. the maximum number of radio channels), additional customers cannot access the cellular system through that BS. Therefore, an existing service may be disconnected during handoff due to the lack of the necessary resource in consecutive cells. To provide QoS guarantees to mobile users, a certain resource reservation scheme has to be implemented for potential handoff calls. This is the focus of the resource reservation module. • Lim ited resource and preferential treatm ent W ith the increasing number of users in a wireless communication system, the load on the frequency spectrum becomes heavier. Thus, it is important to manage the resource effectively. The resource under our consideration is ei ther the number of channels (i.e. time slots or frequency narrow-bands) in 2G FDMA and TDMA systems or the maximum tolerable interference margin 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in 3G CDMA systems. When a system is congested, the preferential treat ment should be applied to optimize the objective function of the system. This practice is in line with the interests of the system operator as well as mobile users. • Channel im pairm ents Classical communication considers the point-to-point level connection that can be disturbed by various propagation conditions. A service is blocked during the handover process if there is no resource available along the path. The ”break- then-make” handoff process adopted by the FDMA/TDMA system suffers a great deal from channel impairments. In contrast, the soft handoff process in the CDMA system mitigates such a situation since it allows each MT to connect to two or more base stations at the same time. 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.3 Contribution of the Research Contributions of this research include the following. • E xtended GC schemes Mathematical modeling of the conventional preferential treatment scheme for handoff calls, known as the guard channel (GC) scheme [8], with a single type application (i.e. voice) has been extended to cover multimedia contents with multiple priority classes [9]. This extension is given in Chapter 3. • Com prehensive service m odel A comprehensive service model for wireless multimedia applications is also presented in Chapter 3. The service model includes multiple priority classes, multiple service rates as well as rate adaptive services. The performance is measured in terms of call blocking rates. • D ynam ic radio resource m anagem ent Based on the proposed service model, a dynamic radio resource management (RRM) scheme comprising adaptive call admission control (CAC) and asso ciated resource reservation estimation (RRE) schemes has been developed in Chapter 3 by considering the QoS requirements for each service class. Ex tensive simulation results are conducted by using the OPNET simulation tool [10]. 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ® Dynam ic IGM schem e for C DM A system The Guard Channel (GC) resource reservation scheme used in the 2G TDMA/FDMA system has been modified to be a novel scheme, called the Interference Guard Margin (IGM) [11], which is suitable for 3G CDMA sys tems. • 2-D im ension Optim al Call A dm ission Control Schem e Trend Anal ysis The CAC policy has been studied based on two types of calls (i.e. new and handoff calls) for a homogeneous handoff (with a single type handoff) wireless system in Section 5.5. In contrast with previous work, we focus on the relation ship between the optimal policy and traffic parameters. The Markov Decision Process (MDP) is used for system modelling while the linear programming (LP) technique is adopted to find the optimal call admission policy. Then, we describe the relationship between the optimal policy and traffic parameters for the 2-dimension case. Here, the term 2-dimension indicates that two types of calls are used to charac terize the system state for a homogeneous handoff system, which include new calls and one type of handoff (hard or soft handoff). The optimal policy de rived from the MDP modelling and the LP technique are discussed thoroughly for different traffic conditions. 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • 3-D im ension Optim al Call Adm ission Control Schem e Perform ance Comparison The 2-dimensional framework has been further extended to the 3-dimension case, where we consider the call admission control for the hybrid handoff (hard/soft handoff) process. Here, by the term of 3-dimension, we mean that three types of calls are used to characterize the system state for a hybrid hand off system. They are new calls, soft handoff calls and hard handoff calls. The proposed scheme can be applied to both channel-limit and interference-limit system via the use of effective capacity to estimate capacity in the CDMA sys tem. Numerical results are conducted using OPNET and compared with other CAC schemes such as the guard channel scheme and the complete sharing scheme. • O ptim al Call A dm ission Control Schem e C om plexity In Section 5.7, we compare the MDP complexities in 2-dimension and 3- dimension cases under different system capacities. This serves as a complexity metric for implementation. Finally, the embedded system design of the devel oped algorithm is discussed as possible future work. 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.4 O utline of D issertation This rest of this thesis proposal is organized as follows. Chapter 2 provides RRM- related background information on the evolution of cellular networks, technical chal lenges and their QoS provision. Previous research work in this area is also reviewed. In Chapter 3, mathematical models for resource reservation schemes based on the GC concept [ 8 ] are reviewed and extended to provide the foundation of circuit- switched connection level QoS analysis. A comprehensive service model is used to study the effect on QoS metrics in terms of call blocking rates and the associ ated objective function. The service model includes not only multiple bandwidth requirements for multimedia traffic but also the priority classes and flexible rate capability. The performance of the proposed dynamic RRM is compared with that of those schemes without RRM. In Chapter 4, the channel-limited TDMA/FDMA paradigm is shifted to interference-limited CDMA environment. A dynamic resource management scheme based on Interference Guard Margin (IGM) is used to support connection-level QoS of multimedia applications in a CDMA system. In Chapter 5, an optimal call admission control scheme using the Markov Decision Process mod elling and the linear programming technique for homogeneous and hybrid handoff processes are discussed. Finally, Chapter 6 gives some concluding remarks on the current work and presents future research plans. 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C hapter 2 Background 2.1 Evolution of Cellular Com m unication Wireless communications have experienced an enormous amount of growth during the last two decades. In this section, we will review the evolution of cellular systems and highlight improvements from one generation to the other. 2.1.1 T he F irst G en eration (1G ) S ystem s The first-generation (1G) wireless communication systems that used analog trans mission for speech services were introduced in early 1980s. There were several types of analog cellular systems including the Nordic Mobile Telephone (NMT) [Nordic countries: Sweden, Nor way, Finland and Denmark, 1981], the Advanced Mo bile Phone Service (AMPS) [U.S., 1983], the Total Access Communications System (TACS) [UK,1984], etc. All 1 G cellular systems used the frequency division multiple access (FDMA) method to achieve spectrum sharing among multiple users. To allow 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. simultaneous transmission and reception, the BS transmits along one set of radio channels, called forward channels, and receives along another set of channels, which are reverse channels from the MT. Since the number of channels is limited by the allocated frequency spectrum of a system, a cellular system adopts the frequency reuse strategy to tackle this problem to increase the number of radio channels. As cellular systems get advanced, directional antennas are applied to sector a cell, so the interference can be minimized. 2.1.2 T he Second G en eration (2G ) S ystem s To meet the need of the increasing capacity of the cellular system and to estab lish compatibility with the evolution of wired networks towards digital systems, the second-generation (2G) wireless cellular systems based on digital transmission tech niques were introduced in early 1990s. Digital cellular systems fall into three basic types of cellular technologies: frequency division multiple access (FDMA), time di vision multiple access (TDMA) and code division multiple access (CDMA). The IS-136 system is sometimes referred to as Digital AMPS (DAMPS), it is the new generation of the TDMA system beyond the IS-54 dual-mode analog-digital system used in north American. IS-136 was led by the Universal Wireless Communica tions Consortium (UWCC) and Committee T1 (Tl) sponsored by the Alliance for Telecommunications Industry Solutions and accredited by the American National 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Standards Institute. The Global System for Mobile Communications (GSM) devel oped in Europe is a mixed type of TDMA/FDMA system. The development of GSM was led by the GSM association and the European Telecommunications Standards Institute (ETSI). CDMA systems differ from FDMA and TDMA systems through the use of coded radio channels. IS-95 (cdmaOne) is an example of the CDMA technology, which was originated by the CDMA Development Group (CDG) and Telecommunications Industry Association (TIA). The 2G digital system has many advantages over the 1G system in terms of capacity, quality, flexibility, security, and system complexity. W ith the demand of new innovative services in general, and wide-band multimedia services in particular, the currently deployed 2G wireless systems have further evolved towards the third- generation (3G) systems to offer more advanced service features. Some systems that extend the existing 2G system are called 2.5G systems. The main feature of 2.5G systems is the data packet service enhancement. They are developed to bridge the 2G and the 3G systems. One example of the 2.5G system is the General Packet Radio Services (GPRS) that can provide higher data-rate packet-switching services up to 115Kbps. This effort leads to Enhanced Data Rates for GSM Evolution (EDGE), which can support services with a data rate up to 384 Kbps. The other example is the step-by-step synchronous approach of CDMA-2000 evolving from current IS- 95 networks, led by the 3GPP2 [12] effort in North America. W ith this approach, the maximum transmission speed of 2G is 64Kbps in IS-95A/B specification while 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the maximum transmission speed of 2.5G is up to 384 Kbps and the maximum transmission speed of 3G is up to 2Mbps [3GP00, ITUOO, RicOO]. 2.1.3 The T hird G en eration (3G ) S ystem s The International Telecommunications Union (ITU) has standardized the 3G sys tem under the name of IMT-2000 (International Mobile Telecommunications-2000). Under the IMT-2000 framework, the third generation (3G) air interfaces include CDMA or TDMA technologies will be developed. IMT-2000 will provide packet and circuit switched services with on-air data rates between 384 Kbps (for the wide-area coverage), and 2 Mbps (for indoor coverage) [aP98]. An illustration of the wireless system evolution from 1 G to 3G is shown in Fig. 2.1. GSM Association W-CDMA iiwisi UWCC UPJiC-133 ( CDG cdmnOne i- ^ C y . cdma2000 10 Knps 6 ) -1 t-i K»:>s 384K-2 Mbps Figure 2.1: The wireless system evolution from 1G to 3G. 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Key features of IMT-2000 include the following[14, 15]: • bit rates up to 2 Mbps (high bit rate transmission); • a variable bit rate to offer bandwidth on demand (rate adaptivity); • high quality requirements from 1 0 % frame error rate to 1 0 ” 6 bit error rate (a wide range of QoS requirement); • high spectrum efficiency (via radio resource management and other mecha nisms) ; • multiplexing of services with different quality requirements; • support of packet-based transmission. IMT-2000 provides a framework for worldwide wireless access by integrating a diverse system consisting of both terrestrial and satellite networks. It also exploits the potential synergy between digital mobile telecommunication technologies and fixed wireless access (FWA) Systems [4]. 2.2 C hallenges of Radio R esource M anagem ent The radio transmission technology and the computer network technology are two core components in a cellular Communication system. From the perspective of call connection, a mobile terminal (MT) will experience some discontinuity when it moves from one cell to another. This is called the handoff process. Unlike the 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. wirecl-line telephone system, a cellular system has to consider the impairment of air-interface and the change of connection due to the mobility of MT. A traditional cellular system has two basic functions: locating and tracking the MT. It should choose an available base station (BS) to continue the connection. Quite a few BS are connected to a centralized main traffic switching office (MTSO), which is also sometimes referred as Mobile Switching Center (MSC), via wired lines. Many tasks are performed in MTSO such as handover control and call admission control. In this research, we adopt the centralized system network architecture. With more and more QoS features and multimedia support in the next generation cellular system, the radio resource management (RRM) becomes an crucial task to make the system efficient and to make users satisfied according to their priority. RRM is the process of developing decisions and taking actions to optimize the system utilization. Technical challenges in RRM are addressed below. 2.2.1 H andoff and M obility Handoff is an inevitable process in a cellular communication system. The main reason is that, to achieve a higher system capacity, the territory is divided into several regions and cells so that the frequency bandwidth can be reused in far-apart cells without interfering each other. Thus, each BS can provide services only to a limited number of MT. As an MT is not attached to a fixed infrastructure, it can 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. move from one cell to the other. The cellular system needs to track and locate MT accordingly. The handoff process and the mobility of MT make the RRM task very chal lenging. With the rapid increase in the size of the wireless mobile community and its demands for high-speed multimedia communications, the allocated spectrum re source appears very limited. Efficient spectrum (or radio resource) management is of great importance to the assignment of channels and the transm itter power to the radio access port and the terminal under the interference constraint. The handoff process has three phases. 1. H andoff detection In the first phase, the system has to decide when and where to handoff. There are three handoff detection strategies. The schemes, whereby MT controls the handoff, are called the mobile-controlled handoff (MCHO). Most cellular systems, such as GSM, rely on a strategy called the mobile-assisted network- controlled (MAHO) handoff scheme. In MCHO, MT periodically takes link quality measurements, and reports signal levels and quality to MTSO, the handoff decision will be made if the measurement is below acceptable threshold. For the third strategy, the total network carries out the handoff process. 2. Handoff resource assignm ent This topic is within the scope of our research. It decides how much resource needed to allocated or reserved for MT in order to achieve the QoS objectives. 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Here, the resource in the TDMA/FDMA system is referred to the available number of channels. In the CDMA system, they correspond to the available interference margin that a system can tolerate. These two basic characteristics have been reflected on the design of CAC and the associated resource reserva tion schemes. In the mobile-assistant handoff system, the CAC mechanism is implemented in MTSO. 3. Handoff execution This last phase is to execute the handoff operation. The main challenge is to reliably transfer underlying data during the process. 2.2.2 C hannel A ssign m en t and R eservation Upon a new call arrival or a handoff attempt, there are many ways of allocating the resource. The goal is to maximize system efficiency while meeting the QoS requirement of users. Some channel assignment techniques allow channel sharing among several cells. Such schemes have a common resource sharing pool that can be allocated to users upon request. Such schemes are referred as Dynamic Chan nel Assignment (DCA) [16, 17] schemes. With DCA, channels are not permanently assigned to cells. These schemes can borrow channels from neighboring cells if nec essary. However, once a channel is borrowed from a neighboring cell, all other cells that are within the co-channel reuse distance are prohibited from using the channel. Therefore, its performance could be worse in the heavy traffic scenario. The other 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. channel assignment schemes allocate a fixed amount of resource to each cell. They are referred to as static channel assignment (SCA) schemes. Most practical wireless systems belong to the SCA category since the complexity of DCA is much higher than that of SCA. Our work falls into the SCA category due to its simplicity. Our proposed schemes adopt preferential treatment among priority classes by selecting proper resource reservation schemes. Furthermore, our schemes can be further di vided into fixed and dynamic schemes depending on the detailed implementation of the preferential treatment policy. 2.3 QoS G u aran tee s an d P re fe re n tia l T re a tm e n t 2.3.1 QoS G uarantees QoS provision for multimedia traffic has been extensively studied for wired net works. Due to user’ s mobility and channel impairments, cellular systems provide a more hostile communication environment in terms of a lower bandwidth, a higher latency and a higher burst error rate in comparison with communication over wired networks. Consequently, providing multimedia services with QoS guarantee in such an environment presents more challenges. Furthermore, to achieve location and plantform transparency (i.e. QoS guarantee at any time wherever a mobile host moves) is again a difficult task. Unlike the wired network system, it is not sufficient to provide an end-to-end path with adequate resources in a wireless cellular system 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. due to the handoff phenomenon. However, with proper resource reservation, one might pre-assign resources so that they are dedicated to a specific connection for a flow, thus guaranteeing the bandwidth, delay, and error rates contracted by the ap plication. This effort means that, in order to provide QoS guarantee to some specific MT, it is necessary to reserve the resource in the neighboring cell in advance. QoS metrics are often measured in terms of the new call blocking rate and the handoff dropping rate. A new call is blocked when the resource is unavailable to sup port this call. Sometimes, even though there is some resource, the new call request is still rejected due to the overall system performance consideration as illustrated by a few examples in the following chapters. The call dropping probability is the probability of an on-going call being forced to terminate before its completion. This usually happens when a mobile user moves to a new cell during the call’s lifetime and the new cell does not have enough resources to support it. For a mobile user, drop ping an on-going call is generally more unbearable than blocking a new call request. Therefore, minimizing the call dropping probability is one major objective in the wireless system design. On the other hand, the goal of a network service provider is to maximize the revenue by improving network resource utilization, which is however tied with minimizing the call blocking probability. The solution involves radio re source allocation, call admission and resource reservation for handoff calls. A system planner should consider all these factors in defining the objectives function, which is a weighted sum of several cost functions. 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.3.2 P referen tial T reatm ent Nobody wants to wait in lines or to be dropped during a conversation. However, due to a limited amount of resources, a system may not be able to provide services to meet requirements of all users at the same time. Then, preferential treatment has to be implemented. Users who want to pay more or are considered more important can be assigned a higher priority to receive a better QoS guarantee. One difficulty here is how to choose the proper weighting among priorities. This is rather a subjective issue. Heuristic and empirical methods are usually given in the literature depending the objectives. There are so far two common strategies adopted to achieve preferential treatment in differentiating handoff and new calls. They are the handoff queue (HQ) [18] and the guard ehannel(GC) [ 8 , 19] schemes. HQ-based methods follow the following principle. When the resource becomes available, one of the calls in the handoff queue is served. If there is no available resource, call requests are queued until the resource becomes available again. The HQ scheme needs a large buffer and a sophisticated scheduling mechanism to provide required QoS to real-time multimedia traffic so as to ensure that queued data do not expire. The basic idea of GC-based admission control schemes is to reserve resources known as guard channels a priori in each cell and to give preferential treatm ent to high-priority and/or handoff calls. As a result, the scheme offers mobile users a better connectivity than users requesting a new 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. call. In such a system, call requests of a lower priority are rejected if the available resource is less than a certain threshold. Detailed implementations of the GC scheme may differ in the number of guard channels reserved by a base station. Hong and Rappaport [ 8 ] used a fixed GC scheme to treat new calls and handoff calls differently, by reserving the same amount of resource for the handoff calls in the entire period of the simulation cycle. In this work, only one traffic class was considered. In [20], Rapport and Purzynski extended the work to multiple service classes. They analyzed the performance based on a proposed mathematical model with the assumption of stationary traffic. Epstein and Schwartz [21] considered a mixed traffic with calls of narrow and wide-band. Our work in Chapter 3 extends the single threshold in the fixed GC scheme to multiple thresholds to deal with multimedia traffic with different priorities. All schemes discussed above are static since such GC schemes cannot adapt to the quick variation of traffic patterns. Many dynamic GC schemes have also been discussed in the literature to improve system efficiency while providing the QoS guarantees to high-priority calls. These dynamic schemes adaptively reserve the amount of resources needed for high-priority calls and, therefore, accept more lower priority calls as compared to the fixed scheme. Naghshineh and Schwartz [22] proposed an analytical model to estimate the resource requirements for handoff calls. In their model, all connection requests have an iden tical traffic profile and the traffic is under stationary conditions. Ramanathan et 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. al. [23] proposed a dynamic resource allocation scheme by estimating the maximal expected resources needed for handoff calls. In [24], Acampora et al. applied a linear weighting scheme (LWS) as a part of their admission control algorithm that uses the average number of ongoing calls in all neighboring cells within the region of awareness to determine the admission. Sutivong and Peha [25] adopted a hybrid scheme based on the weighted sum of ongoing calls in the originating cell and in other cells to determine the call admission policy. 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3 D ynam ic R esource M anagem ent for 2G T D M A /F D M A System s 3.1 Introduction Wireless communication technologies are witnessing rapid growth, lately. The third generation wireless communication systems can support multimedia traffic at a tar get transmission rate of up to 2Mbps for static mobile users and 384kbps for high mobility users. In a cell-based wireless system, geographic region is divided into small sized cells that employ reuse frequency and have increased system capacity and reduced power requirements. Unlike wired networks, communication entities in wireless networks change their connectivity via handoff when they move from one cell to another. The use of micro or pico-sized cells makes the role of handoff proce dures very important in maintaining the service continuity and QoS guarantees to the multimedia applications. 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Due to limited bandwidth resources in wireless multimedia systems, efficient call admission control (CAC) and resource reservation (RR) schemes are necessary to maintain the desired QoS. CAC schemes enable the system to provide QoS to new incoming as well as existing calls. RR scheme, such as GC, is used to reserve the resources for certain high priority calls. On the other hand, network is required to take the advantage of resource sharing among traffic in order to achieve better channel utilization. Obtaining a right balance between the two opposing criteria is a big challenge. A rule of thumb is that the more information we obtain about the on-going calls in the system, the easier it is to estimate the amount of resources to be reserved from sharing pool, in order to provide the QoS guarantees to priority calls, while ensuring high overall system efficiency. This work addresses the issue of how to provide seamless handoffs to mobile users, under the constraint of limited resources, in a multimedia wireless network. We adopt the concept of the guard channel (GC) scheme, which gives preferential treatm ent to the high priority calls, including handoff calls, by reserving a fixed num ber of channels exclusively for them. However, such schemes decreases the handoff dropping rate at the cost of increasing the blocking rate for other users due to poor channel utilization. To deal with this challenge, we introduce a dynamic resource reservation module to efficiently estimate the resources needed to be reserved for high priority calls, by using the SNR and distance information of a mobile in the neighboring cells. 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The rest of this chapter is organized as follows. In Section 3.2, we present an extension of the conventional fixed GC scheme with multiple thresholds to accom modate multiple priority classes traffic in multimedia wireless networks. By a fixed GC scheme, we mean that the number of guard channels in a cell is fixed during the entire operation process. In order to achieve a higher degree of system utilization, we develop dynamic GC schemes in later sections. In Section 3.3.1, we describe a framework that characterizes interaction among communication elements, and find criteria for good CAC and RR designs to improve the QoS measure. A service model and the corresponding traffic profiles are detailed in Section 4.4.1, followed by CAC and RR designs in Section 3.3.3 along with the discussion of the rationale behind them. Sections 3.3.4 and 3.3.5 provide some simulation results. Finally, concluding remarks and future research work are given in Section 4.6. 3.2 Fixed Guard Channel Schem e E xtension w ith M ultiple Thresholds A wireless multimedia system cannot always meet different QoS requirements of mobile users, due to resource constrains. Therefore, the system requires rules to decide who will receive the services according to predefined cost function(s), to avoid unwanted call blocking and handoff dropping while maximizing channel utilization. Usually, handoff calls are assigned higher priority over new calls. 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. How to seamlessly transfer resources between cells during handoff is an important issue. Sophisticated resource reservation and call admission schemes should be in tegrated with the handoff mechanism to provide more flexibility to all mobile users and better QoS guarantees for premium users. Many different admission control strategies have been discussed in the literature to provide priorities to high-priority call and handoff requests, without unnecessary blocking of new connection requests. We first extended the simplest guard channel (GC) [ 8 ] schemes to have multiple thresholds in this sections. 3.2.1 Service M od el The service model and the associated traffic profiles describe the characteristics of the traffic. Three attributes are included in the traffic profile of call i: (1) the bandwidth Consumption (2) the handoff or the new call indicator denoted by P h n, and (3) the priority class, Ifi. In a mobile communication system with a maximum capacity of N channels, the ith, (i < N ) user’ s traffic profile can be represented by T P {§ i, P h n, Iff}. The communication entity and the resource reservation control mechanism take necessary action according to the information in the traffic profile. The bandwidth consumption is different for different media types in typical applications. Currently, we include both audio and video in our media, and their values are assigned accordingly. Parameter P hn is a binary number to indicate whether the call is a handoff or a new call request. The priority class attribute, 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. denoted by II,allo w s users to request one of the two priority service classes, i.e. the high and the low priorities. Users can request a connection of one of the two multimedia traffic types with one of the two different priority attributes (high or low). The indicator P hn is automatically assigned by the wireless system according to each call connection status. While making a decision on admitting a low priority call, the network has to reserve the resources for the expected incoming traffic of a higher priority within the period of its residential time. The residential time means the period between the time the call arrives at a cell (or a new call born in this cell) to the time when the call completes connection in this cell (or it handoffs to other cell). 3.2.2 P rop osed M u ltip le T h resh old s G C Schem e Let us first assign the priority order as follows: handoff call with high priority class (highest priority), new call with high priority class, handoff call with low priority class and new call with low priority class (lowest priority). The concept of the proposed multiple threshold GC scheme is illustrated in Fig. 3.1. In Fig. 3.1, the number of reserved channels, i.e. guard channels, is deter mined by the thresholds set for different priority classes. For each call request, the call admission control (CAC) mechanism calculate available resource (Not allocated resource in Fig. 3.1) according to their priority level. 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 1 1 8 Reserved resource I I Not allocated resource Allocated resource TH_1 TH 2 TH_3 TH_4 r - --- --------------* \ m High Priority High Priority Low Priority Low Priority Handoff call New call Handoff call New call Figure 3.1: Illustration of multiple thresholds applied to different priority classes. The call admission control decision rule is illustrated in Fig. 3.2. The N ^ y indicate the number of current busy channels. The associated GC thresholds are set to TH1=0, TH2=?ni, TH 3 = m 2 and TH4=m 3 for high to low priority classes, respectively. 3.2.3 A n a ly tic a l M od el We can model multiple GC schemes by a M /M /C /C queueing system as in Fig. 3.3. In this model, we consider a traffic scenario with differential treatments among pri ority classes. Four priority classes, high priority handoff call, high priority new call, low priority handoff call, and low priority new call are considered. The bandwidth consumption for each call is considered equally. In order to provide preferential 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C a ll A rrival No| Yes Yes High priority handoff call REJECT CSl! No No, Yes Yes High priority new call N basy < M l y Nof Yes Yes Low priority > handoff call s J V busy < m No, Yes Yes Low; priority new call ] $ [ b usy ' ■ K 'ifS t' 3 Figure 3.2: The decision rule for call admission control, treatm ent to higher priority call, the associated GC thresholds are set to TH1=0, TH 2 = m i, TH 3=m 2 and TH4=m 3, respectively. The value of mi to m 3 are in num ber of channels. We assume the Poisson distribution for total call arrivals (include all priority classes) with an average rate of A and exponentially distributed call holding time Tc with a mean E[TC ] = 1 f[i. We also assume th at the cell residence time is Tr for a mobile user which is exponentially distributed with a mean E[Tr\ = l/rj. The average channel occupation time becomes lj{ji+rj) [26]. We let the call arrival rates, average call holding time, and average cell residence time, for each priority class be 33 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. denoted as a,- • A , l / ( a • p) and l/ ( a • rj) respectively. In which j € {1,2,3,4} for priority classes from high to low, respectively, where (a\ + a 2 + ■ a 3 + 014) = 1 . The call blocking rate for each priority class as given by equations from (3.4) to (3.7). X X ( a i + a 2 + a 3 ) l ( a i + a 2 ) A , a xX r \ r x r x c c y T i t • • Q s M (m l K y K y K y v / K y K 7 (q+p) 2(q+p) m2(n+p) mlft+n) C(n+p) Figure 3.3: The queueing model for the multiple-threshold GC scheme. Then, we can derive the steady-state probability Pj for j channels to be busy as follows. XP 0 = (/j, + rf)Pi A To + 2 • ( j - i + t j ) P 2 = X P i + ( q + r / ) P i XPm.3 - 1 + ( ™ 3 + 1 ) ( M + ri)Pm3+l = ( “ l + “ 2 + < * 3 )XP m 3 + ( m 3 ) ( / i + r))Pm 3 (a x + a 2 + a 3)A P m2_x + ( m 2 + l ) ( p + ^ ) P m 2+x — ( a x + a 2)A P m 2 + (m2)(/r + ??)-Pm2 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (ax + a 2)\Pmi~i + (mi + l)(/r + rj)Pmi+x — axXPmi + (mx)(ji + f])Pm2 ■ aiAPmi_! + (mi + l)(/i + rj)Pmi+i = aiXPmi + (mx)(fi + ij)P„ aiX P c-i - C(/i + r?)Pi The above equation set can be summarized below. (3.1) P- • ■ Po K t* + v (ai+a2+«3): l m 3 (ai+Q2)J m 2-(«i +0 2+0 3)7 7 1 2 m 3 f \ \i p 7! '•/i-H j/ ® if 0 < j < m 3 if m 3 < j < m 2 if m 2 < y < mi (ai)J m l-(ai+a2)m l , ■ (oi+az+as)^, 2 , , P 3 . (_A_\.7 . P if 77I1 < 7 < C 1! v / - < • + ? ? ' — — (3.2) where P o - E m3 .1. . ( \ \7 1 \p m 2 ( a i+ a 2 + « 3 ) J m3 _ / A o ' . J=0 j! 7713+I j! \ 1 y ^ m i ( a i + a 2 ) :,~ m'2 ' ( a i + a 2 + « 3 ) m 2 " m 3 . / A \ j \ 2~ij=m 2 + I j! ' V/z+p ( a i P ~ m l • ( « ! + a2 )ml •(<*! + « 2 + « 3 ) m 2 ~"m 3 . f A \ j 2 -^ j= m 1 + 1 j! ‘ yfi+jjJ - 1 (3.3) Based on (3.2) and (3.3), we can derive the call blocking rate for each priority class as given by equations from (3.4) to (3.7). The blocking rate for the high 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. priority handoff call, the high priority new call, the low priority handoff call and the low priority new call are denoted by PC 1 , PC 2 , PC 3 and PCi, respectively. PC 1 = Pc Pc2 - E pj j~mz Pcs = E Pi j=n% 2 a . = E C j=mi 3.2.4 S im u lation R esu lts Two most popular forms of network simulation are the analytical modelling sim ulation (AMS) and the discrete event simulation (DES). AMS characterizes the network as a set of equations by exploring queuing theory. An over-simplified an alytical model often leads to unrealistic simulation results. DES characterizes the network as a set of finite state machines (FSM) according to control techniques. The nature of random processes and dynamic evolution of FSM make it possible to simulate more complicated and realistic network control mechanisms. While DES has many advantages, it requires far greater processing time. Fortunately, with the advance in the computing power, DES has become more promising and attractive than before. We used the OPtimized Network Engineering Tool (OPNET) in our simulation studies [1 0 ]. OPNET follows the DES concept as the mean to analyze 36 (3.4) (3.5) (3.6) (3.7) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the system performance and behavior. OPNET can be described as a set of deci sion support tools and finite state machines. It has been designed to support the modeling and simulation of a large range of communication systems from a single LAN to a global satellite network. It is equipped with a range of tools, which allow developers to specify models in great detail, to identify elements of the model of interest, to execute the simulation and to analyze generated output data. A seven-cell model is adopted in the simulation, and a maximum of 12 channels per cell is allowed for mobile users. This assumption is valid under the very small cell size scenario. System contains 92 mobile users, with a call generating rate of 10 calls/hour per mobile user and an average call length of 20 minutes. At the center of each cell is a base station, through which all mobile users transmit. All base stations are connected to a single device called MTSO (Mobile Telephone Switching Office). At any instance, each mobile user is controlled by a specific base station and logically belongs to that cell. When a mobile user leaves a cell, its base station notices the decrease in its signal strength. The base station then asks all surrounding base stations how much power they are getting from it. The current base station will transfer ownership to the cell getting the strongest signal. The mobile is then informed of its new base station and assigned a new channel. We compare the traffic under different scenarios as shown in Table 1. Our goal is to investigate effects of two mechanisms, ”priority handoff’ and ”differentiated QoS service class” on system utilization and handoff call blocking rates. Test sets (1) 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and (2) consist of audio signals without any differentiated QoS classes. The system utilization and handoff blocking rate in two test sets are compared to illustrate the effect of the use of priority handoff mechanism in test set (2). In test sets (3) and (4), we include audio and video, and each media type has an equal number of users. In test set (4), audio and video are allocated to the high and low priority classes, respectively. The simulation is run for 60 minutes for each test set. Table 3.1: Traffic Test Sets Test Set 1 Test Set 2 Test Set 3 Test Set 4 Traffic 1 0 0 % Audio 100% Audio 50% Audio 50% Video 50% Audio 50% Video Bandwidth (unit) 1 (Audio) 1 (Audio) 1 (Audio) 2 (Video) 1 (Audio) 2 (Video) Preferential treatment NO New v.s. Handoff NO NO New v.s. Handoff QoS Priority classes NO NO NO 2 classes 3.2.4.1 The effect on system utilization and handoff blocking rate w ith priority handoff System utilization for test sets (1) and (2 ) is shown in Fig. 3.4(a). In test set (2), the system treats handoff calls with a higher priority than new calls, whereas no such priority scheme is used in test set (1 ) (i.e. the handoff as well as new calls are treated with the same priority). The system utilization is a little lower in test set (2 ) than in test set (1 ) since the system reserves some resources for higher priority 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 . 9 ......... © U 0-8.... ’! 0.7 ......... 0.6 0:4 ffi....1 0.2 0.1 — ■ 0m 10m 20m 30m 40m 50m 60m 0.9 0.8 05 0.4 0.3 0.0 0 m 1 0m 30m 40m 50m 60m Simulation Time (min) Simulation Time (min) (a) (b) Figure 3.4: (a) System utilization without (test set 1) and with (test set 2) priority handoff mechanism, (b) Handoff blocking rate without (test set 1) and with (test set 2 ) priority handoff mechanism. handoff calls. However, as observed in Fig. 3.4(b), the handoff blocking rate for test set (2 ) is greatly reduced compared with that for test set (1 ). 3.2.4.2 The effect on system utilization and handoff blocking rate w ith priority handoff and differentiated QoS classes System utilization and handoff blocking rates are compared for test sets (3) and (4) in Figs. 3.5(a) and 3.5(b), respectively. Both test sets (3) and (4) consist of audio and video data with an equal number of users. The priority handoff mechanism as well as the QoS service class are applied to data in test set (4), while none of them is used for test set (3). The system utilization is lower in test set (4) than that in 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. \ C O t S p ‘ g ■ S c c n L > j ! I rr w * ■ r (3) ! * / i (4) U - ..... 1 Om 10m 20m 30m 40m 50m 60m Simulation Time a J - 0m 10m 20m . 30m 40m 50m 60m Simulation Time (b) Figure 3.5: (a) System utilization without (test set 3) and with (test set 4) priority handoff and QoS classes, (b) Handoff blocking rate without (test set 3) and with (test set 4) priority handoff and QoS classes. test set (3) because the system reserves resources not only for handoff calls but also for the higher priority service class. However, as shown in Fig. 3.5(b), the handoff blocking rate for test set (4) is greatly reduced compared with that for test set (3). 3.2.4.3 Com parison of the handoff blocking rate betw een high and low priority service classes Fig. 3.6 shows the handoff blocking rate comparison between the premium and the assured service QoS classes. We see that the handoff blocking rate is much lower for handoff calls in the premium service class. This is due to the fact that more resources have been reserved for handoff calls belonging to the premium class as compared to those belonging to the assured service class. As shown in Figs. 3.5(b) and 3.6, our 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0.6 r - I ------------ j " ..............| .............— j- "5 0 .2 - - m ta o .3 - - o 0.0 0.1 Higii priority class Gm 10m 20m jQm 40m 50m 60m Simulation Time (min) Figure 3.6: Handoff blocking rates scheme can provide a lower blocking rate compared with the system without the use of any priority handoff mechanism and differentiated QoS class. Moreover, our scheme provides a lower blocking rate to the premium user. The dynamic GC scheme achieves a better performance by exploiting the informa tion of message loops. Message loops are inter-networking communications among wireless system components, i.e. Mobile terminals (MT), Base stations (BS) and Main telephone switching office (MTSO). In the following sections, we will intro duce mechanisms of these message loops in a mobile simulation system, and then propose a dynamic scheme. 3.3 D ynam ic G uard Channel Schem e 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.1 M obile Sim ulation S ystem 3.3.1.1 M essage loops in the mobile sim ulation system A mobile network consists of three key communication elements: the mobile terminal (MT), the base station (BS) and the main telephone switching office (MTSO). Their interactions are described in Fig.3.7 to Fig. 3.10 by message loops. • NCQ-ML: The New Call Request Message Loop (see Fig. 3.7) is responsible for request ing a new call. It consists of four messages: RequesLNew, Forward_Req_New, Resp-Req and ForwardLResp-Req. Actions will be taken by each communication element depending on the message they receive. Each mobile terminal sends a new call request to its BS, according to its traffic profile at the startup. Finally, mobile terminal (MT) will either receive an accept or reject acknowledge. MT starts sending data when it receives an accept acknowledge and terminates the connection request if it receives a reject acknowledge. BS forwards a new call request to MTSO. Call admission control module in the MTSO will then decide whether to admit the request according to the traffic profile of that call request, its current cell capacity as well as predefined cost function. • UHC-M L: The Updaie Handoff' Candidate Message Loop (see Fig. 3.8) updates the hand off candidates and consists of CalLUpdate and Forward^CalLUpdate messages. 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mobile Terminal Base Station MTSO Request_New Forward Req New Forward_Resp_Req Resp_Req Figure 3.7: The new call request message loop in the simulation system. Table 3.2: Message in New Call Request Message Loop Message From — > To Description RequestJsfew MT -> BS New call request send to BS Forward.Req.New BS MTSO Forward the new call request Resp.Req MTSO -> BS MTSO’ s decision feedback Foruiard.Resp.Req BS -> MT MTSO’ s decision feedback via BS to MT This message loop begins with an active MT, whose signal is detected by neighboring BSs. Each BS receives signals not only from its associated MTs but also from MTs in the neighboring cells within its power range. If received SNR of an MT in a neighboring cell is greater than the threshold, BS will send a report message to MTSO to register itself as the handoff candidate of that neighboring MT. Mobile terminals can know where to handoff by referencing this handoff candidate registration table in the MTSO when their signal fades. • HR-ML: The Handoff Request Message Loop (see Fig. 3.9) consists of CalLUpdate, Handoff.Request, Resp.Hoff'.Request and Forward.Resp.Hoff.Req messages to 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mobile Terminal Call_Update Forward_CalI_Update MTSO Base Station Figure 3.8: The update handoff candidate message loop in the simulation system. Table 3.3: Message in Update Handoff Candidate Message Loop Message From — > To Description CalL Update MT -♦ BS MT send out signals to allow its current BS and neighboring BS within its radio scope to update handoff candidates Forwards CalL Update BS -► MTSO If the signal of CalLUpdate is stronger than ” reporting” threshold, such BS will send out message to register itself as handoff candidate request a handoff. The BS makes a handoff request to MTSO when it finds that its associated MT has signal degraded down below the threshold. MTSO then checks the HCR table to find out a good candidate cell and admits the handoff request. The decision is sent back to the new as well as old base station, which is forwarded to MT. • C T-M L: The Coll Terminate Message Loop (see Fig.3.10) includes CalLTerm and For- ward-CalLTerm messages to terminate calls. 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mobile Terminal Call_Update Handoff_Request Forwaxd_Resp_Hoff_Req Base Station MTSO Figure 3.9: The handoff request message loop in the simulation system. Table 3.4: Message in Handoff Request Message Loop Message From To Description CalLUpdate MT -* BS MT send out signals to allow its current BS and neighboring BS within its radio scope to update handoff candidates Handoff.Request BS— > MTSO If the signal of CalLUpdate is weaker than acceptable threshold, such BS will send out message to request handoff Resp.Hoff.Req MTSO -► BS MTSO’ s feedback for handoff request Forward.Resp.Hoff.Req BS -> MT MTSO’ s feedback for handoff request 3.3.1.2 U se of SN R and distance inform ation As mentioned earlier, a handoff candidate registration (HCR) table is maintained in the MTSO to record the handoff candidate for each MT. This is how a mobile terminal knows where to handoff when its signal fades. We utilize the SNR and the associated distance information [27] of the mobile radio propagation to identify the 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mobile Terminal Call_Tenninate Forward_Call_Term MTSO Base Station Figure 3.10: The call terminate message loop in the simulation system. Table 3.5: Message in Call Terminate Message Loop Message From — y To Description CalL Terminate MT -♦ BS MT receive reject response from MTSO feedback message or upon call termination itself Forward-CalL Term BS -► MTSO Call termination message is forwarding to MTSO weighting of resources to be set aside for high-priority calls. The relation between the SNR and the associated distance are given below. Pr = Pt - 157.7 - 38.41og(d) + 201og(h1) + 101og(/i2) + Gt + Gm, (3.8) S N R = Pn - P r (3.9) where Pt and Pr are the transm itted and received power (in decibels), respectively, Pn is the environment noise (in dB) received by BS, and Gt and Gm are antenna gains in dB for transm itter and mobile, respectively. The antenna heights of hi (BS) and h2 (MT) are in feet. Suggested values are given in Table 4.1 in [27], for hl=100ft, h2=10ft. Finally, d is the distance, measured in miles, for a given received power. 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. According to above formulae, the SNR and the distance values can be exchanged with each other. 3.3.2 S ervice M od el and T raffic Profile 3.3.2.1 Service m odel w ith m ultiple QoS classes We consider a more comprehensive service model for dynamic scheme here than that used in Section 3.2.1, with the following service attributes: ® ^min) • The maximum and minimum bandwidth requirements describe the bandwidth consumption of the traffic. If traffic is non-rate adaptive, the value of is equivalent to On the other hand, if traffic is rate adaptive, the value of should set to be less than that of $ > max- ® I r a t e • The rate adaptivity indicator describes whether a connection is flexible in its bandwidth requirements. If a connection is rate adaptive, it can be serviced in a degraded mode when congested, and it thus has high probability to receive service, either in full or degraded rate. The value of Irate is give by 1 if call is rate adaptive Irate = { (3.10) 0 if call is non-rate adaptive 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • n: The priority with a high value is assigned to connections that are willing to pay more. They are likely to receive better QoS guarantees in terms of better chance to receive service and in better quality mode. Similarly, system will gain higher rewards if it services such priority calls. The reward function is defined in Eq. (eq:rewardFunetion). There is another traffic attribute, i.e. mobility, denoted as M . It describes the speed property of a mobile terminal. Different mobility traffic will have different weighting factor of the estimated bandwidth needed to be reserved. This information is taken into consideration in the process of resource reservation estimation. In a mobile communication system with a maximum capacity of N channels, the z-th, (i < N ) user’ s traffic profile, T P , can be represented as TPi = {($max,i, IT, Mi}. (3.11) While making a decision on admitting a low priority call, the network has to reserve resources for expected incoming higher-priority traffics in the neighboring cells within the range of a BS’s awareness. The traffic profile of each active call carries its own connection description for the use of requesting a connection as well as the priority referencing by other calls. 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.3 P rop osed D yn am ic G C Schem e 3.3.3.1 Proposed resource reservation estim ation As mentioned in Section 3.3.1, BS will register itself as a handoff candidate, when it receives signals from a specific MT whose SNR is higher than a threshold. Tradi tionally, this handoff candidate registration (HCR) table is used to inform the MT about where to handoff when its signal fades. This table also provides very useful information to estimate future handoff calls for a given cell. Our proposed dynamic resource reservation estimator is based on a non-linear weighting sum as described below, which is different from Linear Weighted Sum (LWS) scheme described in Section 2.3.2. • When an incoming call j requests for resources in the cell jtarget_ceii, a non linear weighting curve as shown in Fig. 3.11 is used to estimate the threshold of resources to be reserved in this cell for serving higher-priority active calls from set S. The following two equations define the weighting factor with and without traffic mobility differentiation, respectively. W t = drh/di if di > dxh 1 if di < dxh (3.12) 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. m = where W{ is the weighting factor and di is the distance obtained by (4.13) and (3.9), and Txh/Ti if Ti > TTh (3.13) 1 if Ti < Txh where W) is the weighting factor when mobility is considered in the service model, and T) and Txh are time related parameters. Weighting factor (Wi) 1 Distance (di) or Time(Ti) Figure 3.11: The non-linear weighting curve for resource reservation estimation. • < F rea as defined in (4.11) represents the resources needed to be reserved for the use of priority calls in our proposed CAC algorithm. Let us first define a set S (j) for call j to be considered for the estimation of resource reservation. The set S (j) consists of all neighboring active calls that satisfy the following two criteria. First, the handoff candidate cell of call i in HCR table is the same as the target cell of call j. Second, the priority of call i is higher than that of incoming call j. Note that, there is one more hidden assumption, i.e. the current cell of neighboring call i is not equal to the target cell of incoming call j. We define the following operations for call i (similar for call j) n (i) : Priority of call i (3.14) 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A(i) : The handoff candidate cell of call i (3.15) A * « The target cell of call i (3.16) (the cell with maximum SNR among the handoff candidate cells) (3.17) Thus, set S (j) can be represented as S (j) = ( * | n ( i ) > n ( j ) , A ( 0 = A * ( ; ) } - (3.18) Fig. 3.12 illustrates an example of set S. In this example, we consider base station B S q as the target cell, on which call admission control algorithm and resource reservation estimation scheme are applied. Assume that at an in stance, there is an incoming call, call j, who just broke the connection with the home base station, B S 2, and making a handoff call request to target cell of B S 0. Apparently, incoming call j knows that B S 0 is its target cell because call j has got the handoff direction command from MTSO. MTSO has scanned the handoff candidate registration table associated to the call j and find out that BSo can receive the strongest signal strength from call j. Upon the call request of j, who has low priority class attribute, resource reservation and call admission control mechanism will do appropriate resource reservation for the potential high priority calls in the neighboring cells. In our case, we exam the 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. operations in Eq. (3.15 and (3.15) on current neighboring active call ii to i5 one by one to decide which elements are considered in set S. B Sj6 BS_2 Incoming call/, B S O BS_3 BS_S BS_4 ® Terminal with high priority class i_n 4 " Call request association ® Terminal with low priority class i_n ■ • ■ — * • Handoff candidate association ^ Base Station (BS) Actual connection association Figure 3.12: Illustration of set S - O p e ra tio n I: n(i2 ), n(i3 ), II(i5) > II(j). Among active calls, i.e. ii to i*,, in the neighboring cells, only * 2 ^ 3 and i5 whose priority greater than incoming handoff call request j. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. - Operation II: A*(j) = B S 0 ; A(i{), A(i3), A(i4), A(i5) = B S 0- Check the handoff candidate registration (HCR) table for each neighbor ing call, we found calls i\, z3, z4 and i$ having B S q on their lists. — Set S: S e tS = {i3, i5}. According the definition, set S includes neighboring calls who satisfy both Operations I and II. • Two types of mobility - high and low - are considered in our simulation. We assign 1 unit and 2 unit speeds for low and high mobility traffic, respectively. The time related parameters, {Ti, Tth}, in Eq. (3.13) can be represented as {di, d,Th} and {di/2, d^h} for low and high mobility traffic, respectively. This implies that a high speed MT is more likely to handoff into current cell even when it is farther from a low speed MT. $res{j) - (3.19) ies • The HCR table helps BS in finding out the most likely potential handoff calls, which would help increase the resource reservation efficiency. 3.3.3.2 Proposed call adm ission control (C A C ) algorithm Our proposed CAC algorithm for new call request is illustrated as follows: 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If Call request is non-rate adaptive If ( N b u S y + $ m a x r e s ) Admit call request with ban d w id th ^ Else Reject call request End Else / * It is a rate adaptive call * / i f {N ^y + ^ m a x) < (c — ^Ves) Admit call request w ith$m ax Else If (Nbusy + 1/2-$ m a x ) < ( c - $ _ ) Admit call request with 1/2 • $ m ox Else If (N busy + $ m in) < ( c - Admit call request w ith$m jn Else Reject call request End End The notations are summarized as in Table 3.6. 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3.6: Parameters used in call admission control algorithm Parameter Legend, C Total number of channels -N busy Number of busy channels * ^res Estimated resource needed to be reserved Ym ax Maximum number of channels requested ®min Minimum number of channels requested Finally, we define our reward function as P ^ 1 ^ adm it(Q * ^h^nip) ' ^ L ,n(f)j (3.20) accept call i where § admit (i) is the bandwidth usage admitted in the process of CAC algorithm test for call i. u>u(i) is the reward weighting factor for priority call i. Similarly, tU h ,n(i) is the reward weighting factor for handoff or new call requests. 3.3.4 S im u lation R esu lts-S cen ario I 3.3.4.1 System and service m odel param eters Simulations are conducted by using OPtimized Network Engineering Tool (OPNET) [10], which is a discrete event simulator. We have implemented our service model and CAC algorithm by using OPNET and compared the traffic under differ ent scenarios. Our goal is to investigate the QoS measures in terms of system reward 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. function defined in Eq.(3.20). We compare our proposed dynamic CAC scheme with the performance of fixed guard channel scheme. BS Figure 3.13: The mobile simulation system with seven cells, 140 mobile terminals (MT), 7 base stations (BS) and one main traffic switching office (MTSO). In our simulation, we used a simple network topology with 7 cells as shown in Fig. (3.13), which covers a region in a non-overlapped fashion. Each cell has its own base station and cell capacity of 50 unit resources. A central control node MTSO is connected with each base station via a wired link. There are a number of mobile users with their own traffic profile in each cell, which can move across two or more cells, according to their predetermined trajectories. Along its trajectory, a mobile user can originate connection requests randomly at its call generation rate. We assume a Poisson generation rate of connection requests, i.e. the inter-arrival time between two consecutive requests from a mobile user is exponentially distributed, 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and the connection duration is exponentially distributed. They are controlled by the following two parameters: • A: the mean request arrival rate, in number of connections per hour • I: the mean duration of each flow in minutes. Parameter I is assigned the value of 20 minutes for each call connection. Increas ing the value of A results in the increment of network traffic load. We performed experiments for light traffic (A = 3 and I = 20) and heavy traffic (A = 10 and I = 20) conditions. We also apply a comprehensive service model, which includes properties of a call connection in its traffic profile: • Qmax, $min ' ■ the maximum and minimum bandwidth requirement of a con nection. Therefore, the bandwidth usage, $, is limited in the range of ( # m aa;, ). In our simulation # G { 1,2,4 }. • Irate : the rate adaptivity indicator. Irate € { 1, 0}. • II : the priority of a connection. II € { 1,2,4, 8 } with equal probability in all simulations. • M : the mobility of a connection, p € { H IG H , L O W } . 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.4.2 Perform ance com parison under traffic w ith m ulti-level priority but w ithout rate adaptive characteristics (w ithout m obility differentiation) To evaluate the performance of the proposed scheme, we used the QoS performance in terms of system reward function, defined in Eq. ( 3.20). The goal of this experi ment is to evaluate the performance of proposed scheme in the presence traffic with multi-level priority classes under the traffic environment where all calls have no rate adaptive capability and mobility differentiation. Common traffic parameters used are < f > = 4 > max, Irate= 0, II € { 1,2,4,8 }, / l l —LOW . The following four schemes are compared: • SI: proposed dynamic CAC and RRE scheme. • S2: 0% fixed guard channel scheme. • S3: 10% fixed guard channel scheme. • S4: 20% fixed guard channel scheme. Fig. 3.14 shows that our proposed dynamic scheme (SI) has best QoS perfor mance in receiving the maximum reward among all the four schemes in light as well as heavy traffic load conditions. 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7,000 6,000 5,000 4 U U l) .000 2,000 1,000 Simulation Time (min) 10.000 (SI) 1 | 6.000 & J j 4,000 2,000} 20m 0 m 40m 60m (a) Simulation Time (min) (b) Figure 3.14: Performance comparison under traffic with multi-level priority but without rate adaptive characteristics (without mobility differentiation) for different schemes under (a) light traffic loading (A = 3) and (b)heavy traffic loading (A = 10). 3.3.4.3 Perform ance comparison under traffic w ith m uti-level priority classes and rate adaptive characteristics ( w ithout m obility differentiation The goal of this experiment is to evaluate the performance of proposed scheme in the presence of traffic with multi-level priority classes under the traffic environment where all calls are rate adaptive. Common traffic parameters are 0 = { 0 mm ,0m a:c}, Irate=1, ft € { 1,2,4,8 }, M —LOW . The same four schemes are compared as in previous experiment. Fig.3.15 shows that our proposed dynamic scheme (SI) has best QoS performance in receiving the maximum reward amongst the four schemes in light as well as heavy traffic load conditions. 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Si) 6 ,0 0 0 "2 5.0001 4,000 X 3.000 2,000 f 1,000 G m 20m 40m 60m 9,000 (SI) 1,000 7.000 1 3 ‘ c3 6.000 i £ ! 5.000} "3 4.000} X > 1 -S 3.000 I O I 2.0001 (S4) 1,000 0 m 20m 40m 60m Simulation Time (min) (a) Simulation Time (min) (b) Figure 3,15: Performance comparison under traffic with muti-level priority classes and rate adaptive characteristics ( without mobility differentiation for different schemes with rate adaptive abilities for all mobiles, under (a) light traffic loading (A — 3) and (b)heavy traffic loading (A = 10) 3.3.4.4 T h e p erfo rm a n ce com parison using tim e-aw are w eighting factor in th e p resen ce of m obility d ifferen tiatio n To investigate the advantages of the use of time-aware weighting factor described in eq. (3.13), The following two schemes, S chem e A and Schem e B are compared. Common traffic parameters used are II € { 1,2,4,8 }, M — { L O W (75%),H IG H (25%) }. The bandwidth usage param eter used are 4 > = and } in the presence of fixed-rate and adaptive rate traffic classes, respectively. • Schem e A: The network support mobility and priority differentiation. 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • Schem e B: The network uses different priority classes, but does not support mobility differentiation. Figs. 3.16 and 3.17 show the simulation results for both schemes with fixed-rate and adaptive rate traffic classes (under light and heavy traffic load), respectively. Results show that dynamic Scheme A performs better due to the use of time-aware weighting factor in the resource reservation estimation(under light as well as heavy traffic load conditions). 6,000 -<A) IB ) 5,000 40m 60m 0 m 15,000 •(A) *<B) 1 10.000 < D P 4 T s £ > O 3 5.000 • Q m 20m 40m 60m Simulation Time (min) Simulation Time (min) (a) (b) Figure 3.16: The performance comparison by using and not using time-aware weight ing factor, in the presence of mobility differentiation with fixed-rate traffic class for all mobiles,under (a) light traffic loading (A = 3) and (b)heavy traffic loading (A = 10) 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Om 20m 40m 60m Om 20m 40m 60m Simulation Time (min) Simulation Time (min) (a) (b) Figure 3.17: The performance comparison by using and not using time-aware weight ing factor, in the presence of mobility differentiation with rate adaptive traffic class for all mobiles,under (a) light traffic loading (A = 3) and (b)heavy traffic loading (A = 10) 3.3.5 S im u lation R esu lts-S cen ario II 3.3.5.1 System and service m odel param eters A seven cells system is used in our simulation. Mobile terminals move in the system according to a certain trajectory and calls are generated in each MT, following the Poisson distribution. The call holding time is an exponential distribution. Cell capacity is 60 unit bandwidth for each cell and the maximum bandwidth requirement for each call supporting multimedia is 6 unit bandwidth. Traffic is classified into four priority classes, Priorityy to P riority4) where P riorityi has highest priority level and class Priority^ has lowest priority. We have assumed that each priority class has equal number of calls in the system. If calls are rate adaptive, they can be serviced 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in degraded quality mode at any discrete rate within the range of ($mox, If calls are not rate adaptive, they can only be serviced at a full rate. The average call holding time (I) is 20 minutes. Three mobility types (Low, Moderate and High speed) are equally distributed in the system. 3.3.5.2 Perform ance com parison betw een proposed dynam ic schem e and fixed GC schem e The performances of proposed dynamic CAC scheme and fixed GC schemes (0% GC, 5% GC) are compared for average call holding time I = 20 (min), and average call generating rate A — 10 (calls/ min/ mobile). Results for rate adaptive and non rate adaptive cases are illustrated in Figs. 3.18(a) and (b), respectively. The result shows that our proposed dynamic scheme out-performs 0% and 5% GC schemes in terms of global system reward R, in both rate adaptive as well as non-rate adaptive system. 3.3.5.3 Perform ance com parison for rate adaptive and non-rate adaptive system The QoS metrics in terms of handoff dropping rate (Phandoff) for each priority class are compared using proposed dynamic CAC scheme for I = 20 (min), and A — 10 (calls/ min/mobile). Results for rate adaptive and non-rate adaptive cases are shown in Fig. 3.19 (a) and (b), respectively. The result shows that higher priority class will receive lower handoff dropping rate due to the use of resource reservation. 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.5.4 Perform ance comparison among different priority classes in proposed service m odel using dynam ic schem e The performances between rate adaptive system and non-rate adaptive system are compared under different call generating rate A (— 3,6 and 10). QoS metrics in terms of reward, R, and system utilization are illustrated in Figs. 3.20(a) and 3.20(b), respectively. The results show that in rate adaptive system, both system reward and utilization will increase because calls are allowed to be serviced in a degraded mode when system is congested. 15,000. ( 1) R A - D y n a m ic ( 2) R A - G C 0% ( 3) R A - G C 5% « c - 1 0 * g 0 2 u G C E £ O m 40m 50m Sim ulation tim e (min) (a) 6.000 3,000 (1) Non RA-Dynamic (2) NonRA-GC 0% (3) NonRA-GC 5% 2.000 1 0 m Simulation time (min) (b) Figure 3.18: System performance comparison between proposed dynamic scheme and fixed GC scheme in (a) rate adaptive and (b) non-rate adaptive system. 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (1) Priority_1 (2) Priority_2 (3) Priorlty^S (4) Priority_4 u ) a s a Q. 63m 20m 50m Simulation time (min) (a) 4 > s Ui C a a 2 ■o o T 3 £ W X !(2) ...if. (1) Prlority_1 (2) Priority_2 (3) Priorlty_3 (4) Priorfty_4 !(1) 1 0 « 2 t a 3 0 » f t SOm Simulation time (min) (b) Figure 3.19: System performance comparison among different priority classes in the (a) rate adaptive and (b) non-rate adaptive system. d ecs RA Non-RA D A Non-RA 4 s b / 0 y 1U 1 Om 10m 20m 30m 40m SOm 60m X(caSSs/mSn/mobISe) Simulation time (min) (a) (b) Figure 3.20: (a)System reward for rate adaptive and non-rate adaptive system, (b) System utilization for rate adaptive and non-rate adaptive system under traffic condition A = 10, I = 20 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.6 C on clusion and Future W ork In this paper, we present a dynamic CAC and associated RR schemes based on the concept of guard channel to adapt the resource access priority by the use of SNR and distance information of the potential higher-priority calls in the neighboring cells, which are likely to handoff. Under light as well as heavy traffic conditions, our proposed CAC scheme out performs the fixed GC schemes. The cases for different traffic profiles of mobile terminals under various traffic conditions are also discussed. We have considered a comprehensive service model, which includes not only mo bile terminals’ bandwidth requirements but also their different levels of priority, rate adaptivity as well as their mobility. Our RR scheme provides more accurate esti mation of the potential higher-priority calls arrival, and thus increases the system reward while providing QoS guarantees to higher-priority calls. Higher system re ward implies that our proposed scheme can get a good balance between resource sharing and resource reservation to achieve the opposing goals of accommodating more calls, while providing QoS guarantees for high-priority classes connections. 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C hapter 4 R adio R esource M anagem ent in C ode D ivision M ultiple A ccess (C D M A ) System 4.1 Introduction The second generation (2G) wireless systems, such as GSM and IS-95, have enabled voice traffic to go wireless, but their capabilities to handle other services, such as data, images and video, are still very limited. W ith the increasing demand for mul timedia services and the stimulus of the International Mobile Telephony 2000(IMT- 2000) standard [28, 29], industry and academia are actively working on efficient methods for providing multimedia services over wireless channels. The third gener ation (3G) wireless systems target at broadband wireless multimedia services. The wide-band CDMA (W-CDMA) technology has emerged as the main air interface for 3G wireless systems, which promises to provide a transmission rate from 144Kbps to 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2Mbps, enabling multimedia services as those provided by broadband wired networks [2]. To meet the large bandwidth requirement of multimedia traffic, it is important to utilize system resources efficiently, and provide preferential treatment according to mobile user’ s traffic profile when the system is congested. The Radio Resource Man^ agement (RRM) module in the cellular network system is responsible for efficient utilizing of air interface resources and guarantee a certain QoS level to different users according to their traffic profiles. The call admission control (CAC) mech anism is one of the most important components of RRM. It affects the resource management efficiency and QoS guarantees provided to users. Besides, the radio resource-reservation estimation (RRE) mechanism helps CAC to decide the amount of resource to be reserved in order to provide QoS guarantees to mobile users. The RRE module residing in each base station dynamically estimates the amount of re source needed by referencing traffic conditions in neighboring cells periodically or upon the call request arrival depending on the design of the system. In 2G TDMA/FDMA mobile systems, network accessibility, controlled by the RRM module, is typically designed based on the number of available channels. Due to the limited channel capacity, preferential treatm ent should be given to high prior ity calls to support them with higher QoS guarantees when the system is congested. Since people expect to receive services, continuously once they are admitted into the system, dropping an ongoing call during handoff is less tolerable than blocking a new 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. call. Thus, it is a widely agreed upon policy that an ongoing handoff call should hold a higher priority than a new call request. One way to provide preferential treat ment is to pre-reserve a certain number of channels for higher priority calls such as handoff calls. This is referred to as the guard channel (GC) scheme [8]. Various GC schemes have been intensively studied, and will be reviewed in Section 4.2. Resource management is also critical in 3G mobile systems. However, the GC approach is not completely suitable for code division multiple access (CDMA) systems for the fol lowing reason. The channel capacity of a 2G TDMA/FDMA system is limited by the number of available channels while the capacity of a CDMA system is limited by maximum tolerable interference in the system. In other words, a new call request in CDMA systems is admitted if it does not introduce excessive interference into the system. Knutsson et al. [30] investigated the CAC scheme for downlink communication for CDMA systems. Due to the asymmetric traffic conditions in the reverse link (from the mobile to the base-station) and the forward link (from the base-station to the mobile), the CAC scheme should admit a call only when the call admission requirements are met in both directions. However, the reverse link capacity is usually more constrained, in CDMA systems [31, 32, 33], and should receive more attention. Huang and Yates [31] and Dimitriou and Tafazolli [32] presented CAC schemes based on transmission power. Liu and El Zarki [33] proposed an SIR-based CAC scheme for the reverse link in DS-CDMA systems to improve the system performance under 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. heavy traffic conditions. In [33], it was assumed that the base station received the same signal power from each of its mobile users, and CAC was designed based on the variation of the SIR (Signal-to-Interference Ratio) value. However, such an assumption does not hold in practical systems, where power control is used to keep SIR close to a target value during the whole operation for each mobile user according to link conditions [34]. Shin et al. [35] proposed an interference-based channel assignment scheme for DS-CDMA Cellular Systems. However, their CAC algorithm was based on fixed resource reservation, where a fixed number of channels is reserved to give the preferential treatment to high priority handoff calls. In this paper, we present a dynamic resource management scheme for multime dia traffic. The system resource is allocated efficiently by using dynamic resource reservation estimation (RRE) and rate-adaptive CAC. In our proposed scheme, a constant target SIR value is assumed due to the use of power control in practical systems. Furthermore, the total interference level in the system is computed by employing the load curve introduced by Holma and Laakso [36]. The use of the load curve makes it possible to handle different levels of interference-increase introduced by heterogeneous traffic with various service rates. There are several distinguished features of our proposed scheme. They are summarized below. First, it supports rate adaptive characteristics for multiple services with flexible QoS guarantees. Second, it takes heterogeneous traffic mobilities into consideration to achieve better resource estimation. Third, by using adaptive resource reservation estimation (RRE), the 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. amount of reserved resource can be dynamically changed by referencing the traffic condition in neighboring cells. Fourth, the proposed scheme bridges two important concepts, the guard channel (GC) and the load curve (LC), to result in a new concept called the interference guard margin (IGM). The resulting CAC scheme gives pref erential treatment to higher priority handoff calls by pre-reserving a certain amount of IGM. The rest of this paper is organized as follows. We provide some background information about priority handoff with guard channels and the load estimation scheme for CDMA systems in Sections 4.2 and 4.3, respectively. In Section 4.4.2, the interference guard margin (IGM) scheme to provide preferential treatment to mobile users in CDMA systems is proposed. It includes the CAC scheme and the associated dynamic RRE method. Several QoS metrics are measured in terms of the cost function, the handoff dropping probability and the new call blocking probability. Section 4.5 shows the simulation results conducted with OPNET (the OPtimized Network Engineering Tool) by using a service model. Finally, concluding remarks and future work are given in Section 4.6. 4.2 Priority H andoff w ith Guard Channels Many different admission control strategies are available for 2G TDMA/FDMA wire less systems to handle the priority handoff calls, without excessive blocking of new connection requests. These schemes fall into two categories: Handoff Queue (HQ) 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A ,n+A ,/, ( m - g ) f i Figure 4.1: The state diagram of a M /M /m /m queue with a total of m channels and g guard channels. and Guard Channel (GC) [8, 9, 22, 23, 24, 25]. To provide mobile users with contin uous connectivity, a system reserves backup channels referred to as ” guard channels” to provide preferential treatment to priority calls and handoff calls. In such a system, call requests with lower priority are rejected if the available resource is less than a certain threshold. Basic GC schemes can be extended to deal with multimedia traf fics with different priorities by using multiple thresholds. Recently, dynamic GC schemes have been discussed in the literature to improve system utilization while providing QoS guarantees to priority calls. These dynamic schemes adaptively re serve resources needed for priority calls and, therefore, accept more lower-priority calls as compared to a fixed GC scheme [22, 23, 24, 25], The QoS performance of a system is often measured by a cost function J in terms of the weighted sum of the rejection probability of each class. When there are only two priority classes (i.e. new-call and handoff-call classes), the system can be 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. modelled by the M /M /m /m queue with a total of m channels and g guard channels as shown in Fig. 4.1. Then, the cost function can be written as J ' P n + ^ h ' P h y (4.1) m Pn = Y , P , (4.2) i— m — g Ph = Pm- (4.3) where Pn is the new call blocking probability and Ph is the handoff dropping proba bility, w n and W h are weighting factors for the new call blocking and the handoff call dropping probabilities, respectively, and Pit 0 < i < m, is the probability that there are i channels busy upon a call arrival. Similarly, Pm stands for the probability that all m channels are busy upon a call arrival. It is clear from (4.2) and (4.3) that new calls can be accepted only in (m — g) channels and remaining channels are reserved for handoff calls only. Furthermore, it is assumed that the new call arrival follows the Poisson distribution with an average call arrival rate of X „, and average call service time l//rn. Similarly, the handoff call arrival follows the Poisson distribution with an average arrival rate of A h and average call service time 1/Hh- Let X — A n + A h, Xh ■ — (X (C < 1) 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and P'n = f^h A We can solve for Pi, which denotes the state probability of i channels being busy. As done in [8], we have where P o = t o ( A / ^ + £ \ i — 1 t — m — g + 1 J However, a mobile system may carry heterogeneous traffics with more compli cated service models, where a closed-form solution may not exist. Some objective function should be defined and measured by simulation for performance comparison. For this purpose, we define a simple cost function as J = wn -Pn + wh - Ph (4.5) n e w _ b l o c k - ^ h a n d o f f -b lo c k = W n ■ T r ----------------------------- 1 - W h - ^ n e w _ r e q u e s t - ^ h a n d o f f - r e q u e s t 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where lV n eW _ b i0c k is the total number of new calls blocked, N new request the total num ber of new calls, requested, iV h a n d o ff_ b io ck the total number of handoff calls blocked, and iV h an d off.req u est the total number of handoff calls requested. 4.3 C apacity and Load E stim ation in C D M A System s Unlike 2G TDMA/FDMA systems, the 3G CDMA system does not have a fixed number of channels. Instead, the capacity of a CDMA system is limited by the total interference it can tolerate, which is called the interference-limit system. In CDMA systems, each new mobile user will increase the overall level of interference. In other words, call blocking occurs when the overall interference level reaches some level above background noise [37]. Normally, the interference level increases rapidly when the system load reaches a certain level. Users with different traffic profiles and attributes, such as the service rate, the signal-to-Interference ratio (SIR) re quirement, media activity, etc., introduce a different amount of interference to the system. These factors are especially important in 3G wireless networks that support multimedia services. Liu and El Zarki [33], Holma and Toskala [36] and Viterbi and Viterbi [37] studied the effect of interference increase for traffics with the following attributes for user v. 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ri : service (data) rate for user i. €i : target signal-to-interference ratio (SIR) is determined according to the QoS requirement in terms of the frame error rate (FER) or the bit error rate (BER) for a specific media type. The SIR value is set to a target value and governed by power control. Here, a perfect power control is assumed to maintain the target value. v : m edia activity level is defined as the ratio of the busy period to the total call holding time and lies in the range between 0 and 1. Overall interference in the system increases with an increase in the media activity level. / : other cell to own cell interference ratio accounts for noise in troduced into the current cell by other cells. Its aggregate value is obtained from field experiments and is assumed to be / = 0.55 in IS-95. To derive the amount of interference e * introduced by user i with data rate Ri and target SIR, we have / j- 1 / \ R i j { y i R i ) f A £ > \ <■* = (Eb/No)i - {Itotal_ S i ) / w ’ (4' ) 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where I t o t a i is the total received power from N active users at the base station, i.e. N I t o t a i = ' y ( + P p j 2 = 1 Eb is the energy per user bit, N 0 the noise spectral density, S', and are the received power at the base station from user i and the activity level of user i, respectively, and PN is background noise. Itotai is limited by an upper-bound Ith for a system. When Itotai is higher than the upper-bound, the system is unstable and the overall interference increases dramatically. By rewriting (4.6), we can express the received power Si for user i at the base station as W = (1 4...... .) 1 ' I t o t a i = A p i • I t o t a i , (4 -7 ) where A pi = (1 + ti-Vi-Ri is called the load factor increment [36]. The current load factor of such an interfer ence system is the sum of load factor increments brought by N active mobile users, i.e. N P = J 2 ^ P i- 4=1 Shapira and Padovani [38] and Holma et al. [36, 39] estimated the interference increase by taking into account the load curve as shown in Fig. 4.2. The ratio of 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Itotai to background noise Pjq is called noise-rise and denoted by rj. The maximum value of 77, denoted by r ] max, is normally set to 10 [37], The noise-rise 77 in Fig. 4.2 can be written as r u p N v = ‘ -l* ± = ^ ~ i h ± J2L = (1 _ = (1 - £ A ft)-1. (48) ■nv r ’ N i=1 By taking the partial derivative of Itotai with respect to pi, we get A ij _ 8Itotai _ <9(-Pjv/( 1 — p)) _ PN _ Itotai A p i ~ dpi dpi ( 1 - p)2 ( 1 - p)' Fig. 4.3 shows the amount of load increments introduced by various data rates Ri and media activity u. Consequently, the interference increment, A/* can be expressed in term of Ri if we replace A pi with (1 + ^ -)_1 as shown below: a j ... A Pi j + j 0\ 7al i — — — • Ito ta i “ Z---------- • R a ta l■ 1 - p 1 - p The load curve serves as a good tool for interference increment estimation in our proposed scheme. The key point here is that given a data rate R , the load increment A pi can be computed to yield interference increment A/j. 78 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.4 Preferential Treatm ent w ith IGM In this section, we will develop an efficient radio resource management scheme based on the concept of interference guard margin (IGM). 4.4.1 Service M od el In a mobile communication system with N active mobile users, the ith (i < N ) user’s traffic profile, which characterizes its services, is described as = { r i, {R m ax, R m in )i, M i } , (4.10) where r,, (R m a X, R m in )i, n* and M, in 9(f), denote user i’s rate adaptivity, ser vice rate range, priority and mobility, respectively. The proposed service model is designed to take advantage of modern coding schemes and advanced mobile commu nication technologies as described below. First, ri is a binary indicator that indicates whether the user can be serviced at reduced bit-rates when the system is congested. To maintain a specified QoS level, a wireless system has to adapt to varying traffic conditions. W ith rate-adaptive features, our proposed call admission control scheme can achieve this goal. Second, the service rate range ( R m ax, R m in )i describes the target bandwidth consumption. If the network has enough resources, the request can be admitted at R m ax■ If the cellular system is overloaded (congested), a rate-adaptive user can be serviced at a 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. lower rate (down to R m ^ ' t or even R mm) with degraded quality of service. Adaptation only takes place at the time of admitting new calls or at handoff epochs. Third, the priority tag 1 1 * helps the system to identify high priority users, who are likely to receive better QoS guarantees. Finally, three mobility types, Mi, are considered in our service model (high, moderate and low mobility). The speed for high, moderate and low mobility traffic are 1, 2 and 4 unit speeds. Each different mobility traffic has a different weighting factor to estimate the amount of resources necessary to be reserved. This is discussed in our proposed resource reservation estimator in Section 4.4.3. It is worthwhile to point out that modern coding schemes such as MPEG-2 [40] , MPEG-4 [41] and JPEG-2000 [42], have rate adaptive abilities for data commu nications. In MPEG-2 video/audio compression standard [40], different layers and profiles are defined to achieve target SNR and the spatial scalability. The base layer (with lowest bit-rates) consists of critical information for decoding the sequence at its lowest visual quality such as DC components. Additional layers provide increas ing quality. Applications using this kind of codecs can adapt to network resource availability by transm itting bit streams coded at different layers. Similarly, MPEG-4 [41], which is the new generation multimedia communication coding standard, has the fine-granular scalability (FGS) mode. Another promising approach for adapta tion is the use of embedded coding schemes, such as the wavelet-based JPEG-2000 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. image coding standard [42]. Instead of a few discrete coding rates provided by a lay ered coding scheme, continuous bit rates can be achieved by cutting a single coded bit stream at almost any bit. Better quality can be obtained by transmitting more bits in the bit stream. 4.4.2 In terferen ce G uard M argin (IG M ) IGM is a natural extension of the guard channel idea developed in the context of TDMA/FDMA systems by considering the load factor for system capacity estimation in CDMA systems. As illustrated in Figs. 4.2 and 4.3, we have the following two operations. First, the load curve is used to estimate the load increase as well as the interference increase. Second, a certain amount of IGM, instead of guard channels, is pre-reserved for high priority calls. The amount of IGM is dynamically adjusted by the RRE module. For a new call to be admitted, the total interference level should not exceed the upper bound of the interference with threshold Ith that the system can tolerate. In addition to the constraint of Irh, a lower priority call should comply with the augmented constraint I'th- The margin between I th and I'th is exactly the guard margin, which provides the preferential treatment to high priority calls by limiting the access to the low priority calls. 81 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Interference level new A h o ld — ► p Load factor P old P i new Figure 4.2: The load curve and the load estimation. CL < 1 ’ .= 0.75 •# * 0 C S3 S s o ■S = 0.25 = 0.5 = 0.75 x 1 0 4 Sou rce rate for u ser i : R. (bps) Figure 4.3: The load increment A pi for different source rates Ri and different media activities Vi. 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.4.3 D yn am ic R esou rce-R eservation E stim ation (R R E ) When a mobile terminal moves toward cell boundaries, the neighboring base stations (BS) receive a stronger signal from it. Each BS in neighboring cells sends messages to the mobile switching center (MSC) to register itself as a handoff candidate for the mobile terminal. This bookkeeping process is done in MSC by using a handoff candidate registration (HCR) table to maintain the registration record and to inform the mobile terminal about where to handoff when its signal fades. The HCR table provides useful information to estimate future handoff calls of a given cell. They are used to estimate the amount of resource, in terms of interference margin, needed to be reserved when admitting a low priority call. In other words, before admitting a new or handoff call j into a cell, the proposed dynamic RRE module estimates the interference guard margin IG M (j) for this call. Call j will be admitted into the cell only when the resulting net interference of the system is less than Irh — IG M (j) after admitting this call. The estimation is based on the weighted sum of estimated minimum interference- increments, A /j, according to the traffic profile, for each potential handoff call from neighboring cells as given by IG M (j) = a ■ (4.11) ies(j) ,&Pri E / j \ * \ ~ ' -^total) 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ^ (1 + — f — )_ 1 — ^ ’ i * * t o t a l iesu) X ~ P where a, 0 < a < 1, is an empirical scaling factor that takes into account the fact that either some calls i from neighboring cells which are likely to handoff in the current cell terminate before they actually arrive or ongoing calls in the current cell terminate (or handoff to other cells). The results for IGM by using a = 1 and a = 0.7 are compared in Section 4.5. Furthermore, the weighting factor w, for user i in (4.11) is proportional to the ratio of mobility Mi to user’ s distance di from the base station, i.e. L O i oc (Mi/di) = where % is represented in unit of time. The factor w * implies that a high speed mobile user could be more likely to handoff into the current cell even when it is farther from the base station as compared to a low speed mobile user. Let us define Trh/Ti, if Ti > TTh, (4.12) 1, if Ti < Trh, where the threshold, Txh, is an empirical value (in time unit) determined by a critical distance d^h of a call from the target cell boundary with a typical mobile speed. 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For instance, if we use a value of dxh = 400 meter and a mobile speed of 30 miles/hr (800 m/min), the value of Tph is approximately 0.5 min. Therefore, any mobile user whose estimated arrival time T is less than 0.5 min is very likely to handoff into the current cell and we use tU i = 1, to reserve minimum resources requested by it. For T greater than Tth, we reserve the resources partially. For another example, if Ti is estimated as 1.0 min, we use oji — 0.5 and reserve one half of the requested minimum resources. The distance d* for user i is measured based on the received signal power Pr at its mobile terminal, by using the following radio propagation model [43] : PT = Pt - 157.7 - 38.41og(di) + 201og(/n) + 101og(/i2) + Gt + Gm, (4.13) where Pt and Pr are the transmitted power from the base station (in dB) and the received power at the mobile terminal (in dB), respectively. Gt and Gm are antenna gains (in dB) for the base station and the mobile terminal, and hi and h2 are antenna heights for base station and mobile terminal (in feet), respectively. Suggested values for these parameters are Gt — 6dB, Gm = OdB, Gt — 6dB, h i—100ft and /i2— 10ft (see Table 4.1 in [43]). Next, we show an efficient way to determine set S(j) in Eq. (4.11). This set consists of all neighboring active calls that meet two criteria. First, the priority 11(f) of call i is higher than that of incoming call j denoted by II(j). Second, the target cell A*(j) of call j is in the set of the handoff candidate cells A(i) of call i in the 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. HCR table. Here, the target cell A*(j) is defined as the cell with maximum SIR among the handoff candidate cells of call j. Note also that the current cell of call i is not the same as the target cell of incoming call j. To conclude, we have S(j) = (*|n(i) > n(j),A*(j) € A(i)}. (4.14) Figs. 4.4 (a) and (b) illustrate criteria n(i) > n (j) and A*{j) € A(i) for set S(j ), respectively. In Fig. 4.4(a), we show an incoming call j that requests a handoff from its current base station B S 3 toward its target cell B S q- We will find out neighboring calls w.r.t. BSo whose priority is higher than that of call j. From this operation, we get (i|H (i) > n (j)} = {i2, * 3 ,^5}- In Fig. 4.4 (b), the target cell B S 0 is chosen as the best candidate cell from j ’ s HCR table. Let us denote cell BSo by A (j)*. In this figure, we see several dotted lines associated to each call, which represent handoff candidate cells for each call i. For example, i2, i3 and i5 calls have {BS$}, {B S 0, B S 6} and {BSo, B S 2}, respectively, as their candidate cells. Here, only calls i3 and i5 satisfy the condition {A*{j) € A(f)}. From the results of (a) and (b) in Eq. (4.14), we get S(j) = 4.4.4 C all A d m ission C ontrol (C A C ) A lgorith m The pseudocode of the CAC algorithm for a media type with three scalable rates is given in Fig. 4.5, where a new call or a handoff call can be admitted into the system with three data rates : R max, Rhaif and Rmin. It can be generalized to a media type 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. BS0 H ig h p rio rity c a ll BS3 © ^ \ L o w priority c a ll l i ~ l 5 N e ig h b o r in g c e ll c a ll (w.r.t. BS0 ) In c o m in g ca ll (to B S 0) B S■ > * I C a n d id a te h a n d o ff c e ll a s s o c ia t io n B e s t c a n d id a te h a n d o ff cell C u rren t h o m e cell (b) Figure 4.4: Set S(j) in resource-reservation estimation: (a) II(z) > II(j) and (b)A(j)* € A(i). 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 01 I f I n c o m in g c a l l s a r e n e w c a l l s 02 I f C a l l s a r e n o n - r a t e a d a p t i v e 03 I f ( I c u r r e n t + A 7j) < ( I x h ~ I G M n e w ) 04 A d m it c a l l r e q u e s t w i t h r a t e R t 0 5 E l s e 06 R e j e c t call r e q u e s t 07 E ls e / * C alls a r e r a t e a d a p t iv e * / 0 8 I f ( ^ c u r r e n t 4“ A I m a x , i ) ^ ( / t / i 7 G M n e w ) 09 A d m it c a l l r e q u e s t w i t h r a t e R m a x , i 10 E l s e I f ( - / c u r r e n t 4“ (7xVi 7tfrA 7ne. lt,) 11 A d m it c a l l r e q u e s t w i t h r a t e R h a i j , i 12 E l s e I f ( ^ c u r r e n t 4” A ^ (TjT/i I G h d y i c w ) 13 A d m it c a l l r e q u e s t w i t h r a t e R minii 14 E l s e 15 R e j e c t call r e q u e s t 16 E ls e / I n c o m in g c a l l s a r e h a n d o f f c a l l s * / 17 I f C a l l s a r e n o n - r a t e a d a p t i v e 18 I f (I c u r r e n t 4* A 7j) (Tp/i I G M h a n d o f f ) 19 A d m it c all r e q u e s t w i t h r a t e R i 2 0 E l s e 21 R e j e c t call r e q u e s t 22 E ls e / * C a lls a r e r a t e a d a p t iv e * / 23 I f ( ^ c u r r e n t 4“ A I m a x , i ) ^ i ^ T h -7G ^ h a n d o f f ) 24 A d m it call r e q u e s t w it h r a t e 2 5 E l s e I f ( I c u r r e n t 4“ A I h a l f , i ) ^ (Tp/i ~~ 7 G A7h a n d o f f ) 26 A d m it c a l l r e q u e s t w i t h r a t e R h a i f , i 2 7 E l s e I f (7 c u r r e n t 4“ A I m i n , i ) *7 (Tj7i " 7 G A7h a n d o f f ) 28 A d m it call r e q u e s t w it h r a t e 2 9 E l s e 30 R e j e c t call r e q u e s t Figure 4.5: The proposed call admission control algorithm. consisting of even more rates. Note that IG M new and IGMhandoff are the estimated bandwidths required to be reserved for new and handoff calls, respectively. The basic concept behind CAC is to test whether there is enough system resource left to serve the current call request at a certain rate, after reserving the necessary resource for preferential treatment. The CAC test is performed according to the following steps: 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Step 1. Step 2. Step 3. Step 4- New or handoff call test: An incoming call is first identified as a new or a handoff call type to decide its priority. Rate adaptivity test: The rate adaptivity of a new call (handoff call) is tested to decide whether it can be serviced at a lower data rate if the system is congested. Non-rate adaptive call test: If the call is rate-adaptive, go to Step 4- Otherwise, test whether the amount of interference after admitting the current call and reserving the estimated IGM will exceed the maximum interference level that the system can tolerate. Rate adaptive call test: If the call is rate-adaptive, the current call could be serviced at rates of R max,i, R haif,i and R min,i, depending on the system traffic condition. The amount of interferences introduced by a call are A /m ax>i, A Ihaif,i and A /m iT l)i when it is serviced at rates Rm ax,i, Rhaif,i and R min,i, respectively. Then, we test the admission criteria by the order of data rates for the highest to the lowest. The call is served at its highest admissible rate. 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.5 Sim ulation R esults 4.5.1 S ystem M od el and Link C h aracteristics Simulations were conducted by using the OPtimized Network Engineering Tool (OP- NET) [10]. The link characteristics of the CDMA system used in simulation are given below. A network topology with seven cells, which cover a region in a non-overlapped fashion, is applied. The maximum interference level Ixh is normally set to ten times of background noise, i.e. rjmax — 10. The same radio frequency band is reused for every cell, and separated frequency bands are used for the reverse link and the forward link. There are 420 mobile terminals with three types of mobility (equally distributed). Each cell has its own home base station. Each base station is connected with the mobile switching center (MSC) via a wired link. There are a number of mobile users with their own traffic profiles in each cell, which can move across two or more cells according to their predetermined trajectories. Along its trajectory, a mobile user can originate connection requests randomly at its call generation rate. The Poisson call arrival rate and the exponentially distributed call holding time are assumed. The call arrival rate and the call holding time are controlled by two parameters, i.e. A (the mean request arrival rate measured in the number of connec tions per hour) and I (the mean call holding time of each flow in minutes, which is set to 15 for each call connection). Increasing the value of A results in the increment of the network traffic load. 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Values used in the traffic profile Sr(i) of user i are listed below: 1 ) n e {YE S, NO}, 2) Rmax,i are, respectively, set to 19.2 Kbps, 38.4 Kbps and 76.8 Kbps jR for voice, audio and video traffics and Rmin,i is set to b e , 3) II G {new, handoff}, 4) M i € {H IGH , M O D , L O W }, 5) Communication system parameters used in simulation include: CDMA chip rate W = 3.84Maps, media activity v = 1, and target SIR Si = 7dB. To illustrate the advantage of dynamic IGM, we compare the QoS performance in terms of the cost function J given in Eq. (4.5) for the following four scenarios: • Non-priority scheme (also referred to as the complete sharing scheme), • Fixed IGM 20% scheme (i.e. IGM is fixed to 20% Irk), • Dynamic IGM scheme with a = 1, • Dynamic IGM scheme with a = 0.7. 4.5.2 N on -rate A d ap tive Traffic Figs. 4.6(a) and (b) show the performance under light to moderate traffic load with A varying from 0.1 to 1.3 (calls per hour per user). Fig. 4.6(a) shows that the dynamic IGM scheme has the best QoS performance (evaluated in terms of cost function J). Given a new-call blocking weighting c u n — 1, we used the handoff-call dropping weighting = 10 for J to reflect the higher cost for dropping a handoff 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. call. Fig. 4.6(b) shows that the dynamic IGM scheme significantly reduces the handoff dropping probability Ph without much increase in the new call blocking probability Pn as compared to the non-priority scheme. Some calls from the neighboring cells will not handoff to the current cell and some other ongoing calls in the current cell will either terminate or handoff to other cells. Thus, we use a scaling factor a to avoid excessive resource reservation in the choice of IGM. Figs. 4.7(a) and (b) show that the dynamic IGM scheme with scaling factor a = 1 has the best performance for moderate to heavy traffic loads (where A varies from 1 to 5) in the cost function J and and the associated probabilities Ph and Pn. The performance of the dynamic IGM scheme with scaling factor a = 0.7 is next to it. The use of scaling factor a = 0.7, however, increases the system utilization as shown in Section 4.5.4. 4.5.3 R a te A d a p tiv e Traffic The rate adaptive traffic can be admitted into a system with a lower data rate when the system is congested. Figs. 4.8(a) and (b) show the system performance under light to moderate traffic load with A varying from 0.1 to 1.3. The performance comparison in terms of the cost function J is given in Fig. 4.8(a). Results show that, the non-priority scheme is better than a fixed scheme (i.e., 20% fixed IGM) for light traffic conditions (A = 0.1 to 1.1). This is due to the fact that, a fixed scheme cannot adapt to the traffic condition and still reserves excessive resources 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Units) Non-Prfority Scheme -© - 20% Fixed IGM Dynamic IGM_______ 2.5 I I 3 a, 3 + a, i i Non-Priority S chem e 20% Fixed IGM Dynamic IGM 0.5 0.5 0.7 X (calls/hr/mobile) (a) 0 .3 5 -M- Ph (Non-priority scheme) .. n ■ ■ Pn (Non-priority scheme) - 0 - Ph (Dynamic IG M scheme) , ^ , Pn (Dynamic IG M scheme) 0.3 a, 0.25 Dynamic IGM sch etpe 0.2 0.15 Non-priority sch em e 0.1 0 .0 5 0:3 X (calls/hr/mobile) (b) Figure 4.6: Performance comparison for non-rate adaptive users under light to mod erate traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph. 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Units) -M- Non-priority scheme -O- Fixed IG M 20% Dynamic IG M (a=1) ■ Dynamic IG M (a=0.7) N on-Priority S chem e I I 3” -c O , 3* + a,= 20% Fixed IGM Dynamic IGM (oc=0.7) Dynam ic IGM (a=1) 3 3.5 X (calls/hr/mobile) 2.5 4.5 (a) Dynamic IGM schem e(a=1) p -N- Non-priority scheme Ph Non-priority scheme Pn Dynamic IG M Ph (a=1) Dynamic IG M Pn (cx = 1 ) _ q_ Dynamic IG M Ph («=0.7) Dynamic IG M Pn (a=0.7) 0.7 a. 0.5 0.4 N on-priority s c h e m e 0.3 Dynamic IGM scheme(a=0.7) Dynamic IGM schem e(a=1) * 3 3 .5 X (calls/hr/mobile) 2.5 4 .5 (b) Figure 4.7: Performance comparison for non-rate adaptive users under moderate to heavy traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph- 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. under the light traffic load, which leads to a higher new-call blocking probability Pn. However, the dynamic IGM scheme can adapt well to each traffic condition, and has the best QoS performance. The new call blocking probability Pn and the handoff call dropping probability Ph are shown in Fig. 4.8(b). Again, the dynamic IGM scheme has the best performance under the light as well as the moderate traffic loads. Fig. 4.9(a) is evaluated under the moderate to the heavy traffic loads with A varying from 1 to 5 (calls per hour per user) for the cost function J, where the proposed dynamic IGM scheme has the best performance. The assoicated new call blocking and handoff dropping probabilities (i.e. Pn and Ph) are shown in Fig. 4.9(b). Results show that proposed dynamic IGM scheme significantly reduces the handoff dropping probability without much increase the new call blocking probability. The use of scaling factor a = 0.7 increases system utilization for the dynamic IGM scheme with a = 1 as discussed in Section 4.5.4. We compare the cost function J for non-rate adaptive as well as rate adaptive cases in Fig. 4.10. It is clear that the proposed dynamic IGM scheme outperforms the non-priority scheme for both rate adaptive and non-rate adaptive cases. Fur thermore, the dynamic IGM scheme achieves a more significant improvement when the rate adaptive mechanism is available. 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Units) Non-Priority Scheme 20% Fixed IG M Dynamic IG M _____ 2.5 i i 3 -s; 3 Non-Priority Schem e + a, l l = : 20% Fixed IGM Dynamic IGM 0.5 0.5 0.7 X (calls/hr/mobile) 0.9 (a) 0 .3 5 Ph (Non-priority scheme) Pn (Non-priority scheme) Ph (Dynamic IGM scheme) Pn (Dynamic IGM scheme) ft. 0 .3 0 .2 5 D ynam ic IGM s c h e m e 0.2 0.15 N o n -p rio rity sch e m e 0.1 0 .0 5 0*3 o r ? X (calls/hr/m obile) 0 T o (b) Figure 4.8: Performance comparison for rate adaptive users under light to moderate traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph- 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Units) -M- Non-Priority Scheme -9 - Fixed IG M 20% -0 - Dynamic IG M w ith a=1 ■ Dynamic IG M w ith a=0.7 o N on-Priority Schem e 20% Fixed IGM Dynamic IGM (a=0.7) Dynam ic IGM (a=1) 2.5 3 3.5 X (calls/hr/mobile) 4.5 (a) Dynamic IGM scheme(tx=1) p -M- Non-priority scheme Ph . . i t . . Non-priority scheme Pn Dynamic IG M Ph (a=1) Dynamic IGM Pn (a=1) q Dynamic IGM Ph (ot=0.7) , ,q, , Dynamic IG M Pn (a=0.7) 0 .7 e a, 0.5 N o n -p rio rity s c h e m e 0.3 Dynamic IGM scheme(a=0.7] 0.2 Dynamic IGM schem e(a=1) P 0.1 2.5 X (calls/hr/mobile) 3 .5 4 .5 (b) Figure 4.9: Performance comparison for rate adaptive users under moderate to heavy traffic load: (a) the cost function J and (b) the new call blocking rate Pn and the handoff dropping rate Ph- 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (Units) ■'*" Non-priority scheme (Non-RA) ■ ' Non-priority scheme (RA) -«•- Dynamic IG M scheme (Non-RA) Dynamic IG M scheme (RA ) 3 Non-RA RA N on-priority s ch em e C ft. I I Dynam ic IGM sch em e Non-RA RA 3 3.5 X (calls/hr/mobile) 4.5 1.5 2.5 Figure 4.10: Performance comparison between rate adaptive (RA) and non-rate adaptive (Non-RA) schemes for the cost function J. 98 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 .5 .4 S y stem U tiliza tio n System utilization for all four schemes w.r.t non-rate adaptive and rate-adaptive cases is compared in Fig. 4.11 under the heavy traffic load (i.e. A = 5). We see that system utilization is higher for traffic with the rate-adaptive capability than that without the rate-adaptive capability. This can be explained by the fact that the system can provide calls with a reduced data rate when the system is congested, thus increasing the overall system utilization. Due to resource reservation adopted by the proposed dynamic IGM scheme, it cannot fully utilize the system resource in order to provide preferential treatment to higher priority calls. The use of scaling factor a — 0.7 in IGM increases the system utilization for non-rate adaptive traffic at the expense of dropping more handoff calls. As shown in the figure, the non-priority scheme has the best system utilization performance since it does not reserve any resources for handoff calls, and accepts the calls on the first-come first-served basis. The fixed IGM scheme gives higher system utilization as compared to the dynamic IGM scheme because the latter reserves more resources to serve handoff calls at the heavy load condition. Finally, we would like to point out that system utilization is about the same for the non-priority scheme, the fixed IGM and the dynamic IGM under the light traffic condition. System utilization is also not sensitive to the rate-adaptive capability of the underlying traffic. Usually, the maximum rate demanded by a user can be served. 99 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.6 C onclusion and future work Effective radio resource management schemes, including dynamic radio resource es timation (RRE) and call admission control (CAC), based on the concept of inter ference guard margin (IGM) for CDMA systems were presented. We considered a service model that included mobile terminals’ service rate, their different levels of priority, rate adaptivity as well as their mobility. The proposed dynamic IGM scheme reserves a certain amount of interference margin for high priority handoff calls by referencing the traffic condition and mobile users’ traffic profile in neigh boring cells. The mobility-aware weighted sum plays an important role in the RRE process so that the effect of different mobility is taken into consideration. It was shown by computer simulation that the proposed fixed and dynamic IGM schemes outperform the non-priority scheme in giving a smaller cost function J under light as well as heavy traffic conditions. In modern CDMA systems, soft handoff is employed to provide a better tran sition process than hard handoff. The benefit of soft-handoff is that a mobile user can connect two or more base stations at the same time, thus greatly reducing the probability of call-dropping due to severe channel impairments. However, soft- handoff should be used only up to a certain extent because an excessive amount of soft-handoff connections increases the downlink interference. In our future work, the IGM scheme will be generalized to CDMA systems with hybrid hard- and soft- handoff schemes to achieve an efficient radio resource management mechanism. 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Non-priority Fixed IGM 20% Dynamic IGM Dynamic IGM ( a=l ) ( a=0.7) Figure 4.11: Comparison of system utilization w.r.t. rate-adaptive and non-rate- adaptive traffic under the heavy load with A = 5. I Non-RA IR A Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C hapter 5 O ptim al R adio Resource M anagem ent w ith M arkov D ecision Process 5.1 Introduction W ith the increasing demand for multimedia services and the stimulus of the Interna tional Mobile Telephony 2000 (IMT-2000) standard [28, 29], industry and academia are actively working on efficient methods for providing multimedia services over wireless channels. In the Universal Mobile Telecommunications System (UMTS) and other next generation standards, the quality of service (QoS) provisioning for emerging multimedia services is one major challenge. To meet the large bandwidth requirement of multimedia traffic, it is important to utilize system resources effi ciently. The Radio Resource Management (RRM) module in the cellular network system is responsible for efficient utilization of air interface resources to guarantee a certain QoS level for different users according to their traffic profiles. The call 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. admission control (CAC) policy is vital, since it affects the resource usage efficiency and QoS provided to users. Many CAC schemes in mobile networks have been proposed in the literature [ 8 ], [9], [44], [45], They were discussed in previous chapters. One of the major challenges in the design of CAC policies is to provide preferential treatm ent among users while utilizing the system resource efficiently. When a mobile user moves from one cell to another, a handoff occurs. The handoff process makes QoS provisioning even more difficult due to the volatile resource availability from one cell to the other and the potential delay as a result of complicated message routing and forwarding. In the code division multiple access (CDMA) system, soft handoff is employed to provide a better transition process than hard handoff. The benefit of soft-handoff is that a mobile user can connect two or more base stations at the same time, thus greatly reducing the probability of call-dropping in the presence of severe channel impair ments. However, soft-handoff should be kept at a reasonable level since an excessive amount of soft-handoff connections may result in higher downlink interference. Dropping an ongoing call during the handoff process is less tolerable than block ing a new call. Higher QoS guarantees should be given to handoff calls in terms of a lower handoff dropping rate, since people expect to receive services continuously once they are admitted to the system. The guard channel (GC) scheme proposed in [ 8 ] is one of the popular schemes in providing preferential treatment. It is simple and easy to implement. Various GC schemes have been intensively studied. We 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. also proposed a dynamic CAC scheme with a rich service model in our previous work [44], The GC scheme was proved by Ramjee et al. [45] to be optimal for an objective function formed by a linear weighting of the new call blocking and the handoff dropping probabilities. However, the guard channel (GC) call admission control scheme is not optimal for multiple traffic classes. In order to achieve better QoS, it is necessary to have a stochastic control policy. One well known stochastic control approach is the Markov decision process (MDP), which is a branch from operation research. MDP models are widely used in diverse research areas and practical applications such as ecology [46, 47], eco nomics [48] and network routing [49]. Extensive application and examples are given in [50, 51]. In this chapter, we focus on the application of MDP to the design of the call admission control policy. Stochastic control with dynamic programming is often used in determining the optimal call admission control policy. Rezaiifar, Makowski and Kumar [52] developed a dynamic programming algorithm to design the optimal handoff strategy in cellular radio systems. They studied the problem of choosing the optimal fixed threshold that minimizes the cost function associated with switching the cell site and maximizes the reward for improving the quality of the call. While dynamic programming can be used to solve simplified problems as given in [53], the high computational cost as a result of the large state space size makes dynamic programming practically infeasible. 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Unlike dynamic programming, MDP can be used to derive an optimal call ad mission policy in a stationary sense. In most cases, where the traffic condition does not change rapidly, the MDP-based call admission policy provides a better trade-off between optimality and complexity. In the literature, MDP was used to determine the optimal call admission policy for priority treatment. Two types of calls, new call and hard-handoff traffic, were considered by Ho [54] to maximize system utilization in deriving the optimal call admission policy. In this work, the accepting preference was given to handoff calls by limiting the dropping rate of handoff calls. Choi et al. [55] studied the highway traffic control system with multiple traffic classes to max imize the revenue. Xiao, Chen and Wang [56] applied the model to rate-adaptive multimedia traffic to re-allocate the system resource to different media types to maximize the revenue. The Semi-Markov Decision Process (SMDP) generalizes the Markov Decision Process (MDP) in the following characteristics [50]. • (1) SMDP allows the decision maker to choose actions whenever the system state changes. • (2) SMDP Models the system evolution in continuous time. (MDP also has a continuous-time version.) • (3) SMDP allows the time spent in a particular state to follow an arbitrary proba bility distribution. 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. According to the description above, the MDP can be viewed as a special case of a SMDP in which inter-transition times are exponentially distributed and actions are chosen at every transition. More details can be found on pp. 530 in Puterm an’s book [50]. In this chapter, we formulate the optimal RRM design as the MDP optimization problem. The stationary optimal CAC policy, which can be used in the hybrid handoff scenario (i.e. hard and soft handoff schemes are both) is determined by solving a set of linear programming equations. The CAC policy can be controlled by choosing appropriate actions to accept or reject calls of different classes according to the current system state. In this work, we examine the relationship between optimal policy and traffic parameters. To be more specific, we address the preferential treatment problem by assigning high priority calls with a larger weighting factor. Experimental results show that CAC can be derived to achieve the best result in terms of weighted system utilization by using the MDP approach. Our work can be easily extended to a rich service model with multiple traffic types and models. Several distinguishing features of the proposed scheme are highlighted below. • The MDP model is used for system modeling while the linear programming (LP) technique is adopted to find the optimal CAC policy in our work. The MDP model was used to determine the optimal CAC policy for new call and hard-handoff traffic [54] as well as rate-adaptive traffic [56] before. However, 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. previous work did not address the behavior of the optimal CAC policy. The CAC policy will be studied for two types of calls in Section 5.5. Unlike previous work, we focus on the relationship between the optimal CAC policy and traffic parameters. • We propose an advanced CAC policy, which is used in the hybrid handoff sce nario (with both hard-handoff and soft-handoff traffic coming from neighboring cells and modelled by MDP). In the CDMA system, hard-handoff is applied when the target cell and its adjacent cells operate at a different frequency. This is referred to as inter-frequency handoff. Soft handoff is adopted when the target cell and its adjacent cells operate at the same frequency, which is referred to as intra-frequency handoff. If there are multiple active CDMA car rier frequencies, independent frequency synthesizers would be required in soft handoff. This is however costly, and hard handoff is chosen in such a scenario as well. • The proposed scheme is suitable for traditional channel-limit systems as well as interference-limit CDMA systems using system capacity estimation as de scribed in Section 5.2. Thus, both systems can be handled within the same framework under one CAC policy. • Soft handoff is implemented in modern CDMA systems to take advantage of macro-diversity to enhance signal quality. However, more system resources are 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. required since soft handoff calls connect two or more base stations simultane ously during handoffs. • From the perspective of CAC precision, the system state of the multiple thresh old guard channel scheme as described in [9] can be viewed as a subset of the system state model proposed in this chapter. The proposed optimization scheme has a higher controlling precision (or a higher system state resolution) and it outperforms the complete sharing and the multiple threshold guard channel schemes. The rest of this chapter is organized as follows. First, system capacity estimation used in the proposed model is discussed for interference-]imit systems in Section 5.2. With capacity estimation, we are able to measure how much resource to allocate, how to conduct the system plan and how to perform the optimal CAC policy accordingly. The MDP formulation is introduced in Section 5.3. It is followed by the LP solution in Section 5.4. In Sections 5.5 and 5.6, optimal CAC policies are derived using the MDP formulation for homogeneous and hybrid handoff systems, respectively, and numerical results are given and discussed for both cases. In Section 5.7, the complexity of the MDP approach for 2-dimension and 3-dimension cases is analyzed. Finally, concluding remarks and future work are described in Section 5.8. 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.2 System Capacity in Interference-lim ited System s Most optimal CAC schemes have been developed for the second generation (2G) cellular systems, in which system resources are divided into ’ ’channels” in the unit of time slots with time division multiple access (TDMA) or in the unit of non overlapping narrow frequency bands in frequency division multiple access (FDMA). The system capacity in a code division multiple access (CDMA) system is however quite different. The capacity of the third generation (3G) CDMA system is lim ited by the overall interference the system can tolerate. Such a system is called the interference-limit system, in which each mobile user admitted to the system con tributes a certain amount of interference to the overall interference. Normally, the interference level increases rapidly when the system load is beyond a threshold level. In other words, call blocking occurs when the overall interference level reaches the maximum level that the system can tolerate. This behavior leads to two issues: (1) how to estimate the system capacity for the CDMA system and (2) how to design CAC policies for CDMA systems according to a finite system state rather than a continuous value of system load, which is proportional to the overall interference level. In the following, we briefly derive the system capacity in terms of the overall interference and the minimum service rate (or the data.rate). The CAC policy will be designed based on the bandwidth 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. consumption state, which belongs to a finite and countable set. Thus, the system load derivation simultaneously solves the above two issues. It is assumed that the system can provide services to user i at data rate R4 = riiR, where R is the minimum data rate and n, is an integer. Let us focus on user 1 , and let Ptotal be the total interference power experienced by user 1 from other N — 1 active users, and be the background noise power. We have N P t o t a l T . Sj + Pjy. (5.1) i= 2 The maximum value of P totai is limited by an upper-bound that is determined by the system stability threshold rj as defined in [37], [2]. ( 77 — 0.1 in [37], [2]). The system is unstable and the overall interference increases dramatically when the ratio of the overall interference power to the background noise power exceeds the value of i. Therefore, for the system to be stable, we have the following constraint: Pn V As a result of (5.1) and (5.2), we have P t o t a l < 1 ^ N N P t o t a l — n i ' R ' ( E b ) i + P n i= 2 i = 2 N > Y , n i .R . (Eb)i + P t o t a l ■ rj, (5.3) i = 2 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where 5) is the received power at the base station from user i and Eb is the energy per user bit. Under the assumption of perfect power control, a fixed target energy to-interference-ratio [37] e * for user i can be achieved. The parameter e * is defined as €i = (Eb/Io)i, and Ptotai = Iq -W. Then, one can rewrite (5.3) to be N N R - { E t ) , « ]T rii ■ R ■ { E b) i < Ptotal ■ (1 - rj) = I0 - W (1 - rj) »=i i=2 N J 2 n i ' ei ^ (W / R ) ' (1 - 7l)- (5- 4 ) i— 1 where Jo is the interference spectral density and W is the chip rate for CDMA system. In the case of the same target value e for all users, we can rewrite the above equation as i > < w m L - A . (5.5, i= 1 6 Eq. (5.5) indicates that the total capacity of a CDMA system should not exceed (5.6) which can be viewed as the allowed maximum bandwidth of the system. We will use this system capacity estimation to design the CAC policy in the next Section. I l l Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.3 M D P and O ptim al CAC Policy MDP is a dynamic process that can model an optimization problem in which the time intervals between consecutive decision epochs are not identical but follow a probability distribution. In our proposed model, the consecutive decision epochs are assumed to follow the exponential distribution. Consider a dynamic process which is observed at discrete time points t = 0,1,2,.... At each observation, the system is classified to be one of the possible states, which are finite and countable. The set of possible states forms the state space denoted by X. For each state x € X, a set of actions denoted by Ax is given, which is again finite and countable. Note that, for a continuous-time Markov de cision process (CT-MDP), a standard uniformization [51] technique can be used to handle CT-MDP with the solution developed for the discrete time case (e.g. the linear programming technique). It is worthwhile to point out that choosing a stationary CAC policy means to find a mapping from the state space to the action space. When a call event happens, we can determine the optimal action according to the current system state based on such a mapping. In this chapter, we discussed handoff scenarios in Section 5.5 and Section 5.6 for homogeneous and hybrid handoff systems, respectively. 112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.3.1 M arkov D ecision P rocess M od el The proposed MDP model can be uniquely identified by the following five compo nents: the decision epochs, the state space, the action space, the reward function and the transition probabilities. • Decision epochs The decision is made only at the occurrence of a call arrival. Call arrival events include new call and handoff call arrivals. At events of call termination or handoff to other cells, decisions will not be made. • State space The state space A is a set of all possible combinations of occupied channels of k types in the system, i.e. k X = {x|x = (x1,x2,---,xky,x1,---,xk > < C}, (5.7) i = 1 where Xi is the number of calls for call type z, and the maximum system capacity can be expressed as C = _ • Action space The action space A is a set of vectors consisting of k binary elements, i.e. A = { a |a = (a1,a2,---,afc);ai,o2,---,afe € (O(reject), 1 (accept)}}, (5.8) 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where a* are actions for each type of calls, respectively. They take the value of 0 for rejecting and 1 for accepting that type of calls. The action space Ax for state x € X can be written as {a = (1,1,---, 1)}, if x = (0,0, • • •, 0), {a =(0,0,-■•,())}, if E h x i = C, {a|a = (ax, a2, • • •, a*);a1; a2, • • •, ak € {0,1}}, Otherwise. Ax — « (5.9) Reward function Let us denote call arrival event as a vector e of k binary values, k {e = (ex, e2, • • ■ , eu ■ ■ -, ek)\ei 6 {0,1},]T et = 1} (5.10) i=1 The above equation means that only one of the k binary values is equal to one, say e* = 1. It indicates that the current arrival event is call type i. Other values of ej, j G {1, • ■ ■ , k},j ^ i, are set to 0. This notation will be convenient for the following reward definition. The reward r(x, a) defined below is earned if the system state is in state x, and the CAC policy is configured as a: k -(x,a) = ^2wi(xi + ei ■ a{), (5.11) »=i 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where W i s are weighting factors for all call types. When the weighting fac tors are equal to one, the objective reward function is to maximize system utilization. • The transition probabilities. Let r(x , a) be the sojourn time, the expected time until a new state is entered [57, 54, 56], when the system is in the present state x € X when action a € Ax is chosen. The value of sojourn time can be expressed by r (x > a ) = 7 ^ —7 TT77 ’ (5-12) * C j = l 4 " X v i = 1 % if^i where o, ’ s represent actions determined from the optimal CAC policy for each call type. They take binary values, i.e. with 1 for accepting a call and 0 for rejecting a call. 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The transition probability from state x with action a to state y can be written as ai ■ h ■ r(x,a), if y = x + (1,0, • • ■ , 0), a2 ■ X2 ■ r(x, a), if y = x + (0,1, ■ ■ •, 0), ak ■ \ k ■ r(x, a), if y = x + (0, 0, • ■ -, 1), P{y|x, a)=^ xi •/ii ■ r(x, a), if y = x — (1, 0, • • ■, 0), (5.13) x 2 • M2 ■ r(x,a), if y = x - (0 ,1, • • •, 0), xk ■ i±k ■ r(x, a), if y = x - (0,0, • ■ •, 1), 0, if y = x. 5.3.2 U n iform ization Technique In this section, we describe a uniformization technique [51], which transforms a continuous-time Markov chain with non-identical decision times, denoted by M, into an equivalent continuous-time Markov process, denoted by M, in which decision epochs are generated by a Poisson process at a uniform rate. After the use of uniformization, the transition process from one state to another can be described by a discrete-time Markov chain which allows fictitious transitions from a state to itself, while process M does not allow transition back to the starting state. By observing the transition probability defined in Eq. (5.13), we see that the state transition rate from state x to y is either A or p depending on the action at the 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. transition epoch. The time spent in state x before transition to state y is r(x, a), if x 7^ y. It is obvious that the time spent in each state is varying according to the value of r(x , a). To transform the non-identical mean of transition times into an equivalent continuous time Markov process, let us define Tc = (J2 k ai + C T i — 1 (5.14) Furthermore, let us define a continuous-time Markov process M with an identical state transition duration of mean rc with the following transition probability «i • Ax■ rc, if y = x + (1,0, • ■ •, 0), a2 -A2 -rc, if y = x + (0,1, ■ - ■ , 0), ak -Xk -rc, if y = x-f- (0,0, - - •, 1), P(y\x,a) = x x -hx-t c, if y = x - (1,0, • • •, 0), x2 -p2 - t c, if y = x — (0,1, • • - ,0), x k ■ n k ■ t c, if y = x - (0 , 0 , • ify = x ^ S j - P ( y K a), if Xy4y •,1), r(x,a) ’ if x = y (5.15) 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. a(x, a) • P(y\x, a), if x ^ y (5.16) 1 — a(x, a) if x = y where a(x, a) = tc/ t (x , a). Note that the net transitional probability of P(y|x, a) from state x to y, with x 7^ y, is the same as P(y|x, a) as shown below. For x ^ y, we have 5.4 Solution via Linear Program m ing The linear programming algorithm is a well known technique to find out Markov decision policies. It has several advantages. First, it is convenient to add more constraints without modifying the structure significantly. Second, it allows us to analyze the sensitivity of the obtained solution. The simplex method is commonly adopted to find the optimal solution. Instead of evaluating the objective function for all candidates satisfying the constraints, this method examines only ” better” candidates, which are known in advance that the objective function will have a larger value [58]. If the decision process M is designed with an identical transition duration distribution, say rc = 1, the linear programming algorithm can be used to OO P (y |x ,a ) = (1 - a(x, a))1 ■ a(x, a) • P(y|x, a) (5.17) = ^(y|x,a). (5.18) 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. find the maximum reward function. It is given below with decision variables 7r(x, a)', x € X and a € Ax: Maximize ]T] r(x, a)7r(x, a)' (5.19) x€EX s l GAx subject to E ^ (y -a )'~ E I] -P(y|x,a)yr(x,a)' = 0 ,V y € X, (5.20) neAy x e x ae A x E E T T ^a)'- 1, (5.21) XeX aej4 ;j; 7r(x, a)' > 0 , x 6 X, a € Ax. (5.22) For the discrete-time MDP case, the term 7r(x, a) can be interpreted as the long term fraction of the decision epochs at which the system is in state x and action a is taken ([51], pp. 181). By using the uniformization technique, which converts the transition probability structure from P (y |x , a) to probability structure P (y |x , a), we can find that solving the standard Linear Programming formulation for the discrete-time MDP problem above is equivalent to that for an uniformized continuous-time MDP problem. Let 7r(x, a) = 7r(x, a )'/r(x , a), we can verify that (5.19) is equivalent to (5.26), and (5.21) is equivalent to (5.28). Furthermore, we will show below that (5.20) is equivalent to (5.27). E ’ ’ ■ ( y > a ) ' - E E p ( y l x> a M x- a )' = ° , v y g x , (5.23) a x ( E . X ^ 2 L & A .X 119 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. E ~ Y P (x = y |x = y, a)7r(x, a)' a €Ay a6Ax - E Y P(y|x,a)7r(x,a)' = 0,Vy e X , (5.24) x ^y,x,yeX agAx ]T 7r(y,a)'(l - (1 -a (a :,a ))) aeA„ • E Y P (y lx> a M x,a )' = 0 ,V y e X, x^y,x,y6X agAr Y n(y> a)) - X) ]T P (y |x , a)7r(x, a)' = 0, V y G X, agAy xjAy.x.yeX a.eAx E ^ W ) ) - E E ^ ^ g = 0 , v y e x , agA,, ' c x # y ,x ,y e X &eAx r(x ,a ) 1 V X ’ v - ^(y.a)' v - V- P(ylx > a) tt( x , a)' n w „ _ v , § „ “ J b « , £ «(*,«) r W - ° ’V y 6 'x ' V 7r(y1a y _ ^ ^ P (y |x ,a)7 r(x ,a)/ £ l y T(X’a ) x ^ y w a a ^ «(*>“) T(X> a) E E E P(y|x,a)7T(x,a) = 0,V y € X , a e A j, ' V x > A ) x,x ,y 6 X a e A * E ^(y> a )- E E P(ylx >a M *,a ) = o,vyex a .eA y x e X a e A t (5.25) where a(x, a) = rc/r(x , a) and P (y|x, a) = 0, if y = x. We can therefore solve the following modified Linear Programming for the continuous-time MDP problem with the standard Linear Programming formulation for the discrete-time MDP problem after the uniformization transformation: Maximize Y Y r (x ; a )r (x > a)7r(x, a) (5.26) X ^ ~ E S - ^ A - x 120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. subject to J2 n(y, a) - E E lx > aM x > a) = 0, V y 6 X (5.27) 3-£E .A jy x £ A Si^i-Ax E E r(x , a)vr(x, a) = 1 , (5.28) xex a eAx tt(x, a) > 0, x € X, a € Ax. (5.29) The term r(x, a)7r(x, a) can be viewed as the long term fraction of the decision epochs at which the system is in state x and action a is taken [56]. The optimal action in each state can be chosen among all actions in each state since the value of 7r(x, a) will be zero for all but one action in each state. This implies that the optimal policy is non-randomized admission policy and the action a chosen for each state is a deterministic function of state x. Next, in Section 5.5, an optimal CAC policy used in a homogeneous system is modelled using MDP, and the behavior of the optimal CAC policy is discussed. 5.5 2-D M D P Optim al CAC P olicy in H om ogeneous H andoff System s In this section, the CAC policy is studied for two types of calls. In contrast with previous work, we focus on the relationship between the optimal policy and traffic parameters. 121 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.5.1 S y stem M od el Consider a cellular system with the maximum system capacity C as given in Eq. (5.6) for a CDMA system. Two types of calls are considered here: new call and handoff calls. In the homogeneous handoff system, only one type of handoff occurs. As described in [59], a soft handoff occurs while the mobile terminal is under the mobile station control and in the traffic channel state. This handoff is characterized by commencing communications with a new base station with the same CDMA frequency assignment before terminating communications with the old base station, a hard handoff occurs when the mobile terminal is transferred between disjoint active sets, where the CDMA frequency assignment changes, the frame offset changes, or the mobile station is directed from a CDMA traffic channel to an analog voice channel. Here, we focus on the hard handoff. Consider that call arrivals follow the Poisson distribution with the following pa rameters (note that in a homogeneous handoff system, only a single type of handoff is present): A n : the arrival rate of new calls Ah : the arrival rate of handoff calls : the service rate of new calls Hh ■ the service rate of handoff calls 122 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The MDP model for the proposed handoff scheme used in a homogeneous hand off system is stated below. • Decision epochs correspond to time instances only at call arrivals. • The state space X is defined as X = {x|x = (xn,xh),xn > 0 ,x h > 0, xn + xh < C}, (5.30) where xn and x^ are numbers of new calls and handoff calls, respectively, and c = ^(w/R)(i-n)^ • The action space is a set of vectors consisting of two binary elements, i.e. A = {a|a = (an, ah); an, ah € (O(reject), 1 (accept)}}, (5.31) where an and a* are actions for new calls and handoff calls, respectively. They take the value of 0 for rejecting and 1 for accepting th at type of calls. The action space Ax for state x 6 X can be written as (a = (1,1)}, if x = (0,0), Ax = (a = (0,0)}, if xn + xh = C, (5.32) {a|a = (on, ah)] an, ah € (0,1}}, if otherwise. 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Let r(x , a) be the sojourn time in the present state x € X when action a € Ax is chosen. We obtain r(x,a) = - ■ j ------------------, (5.33) T '^hQ 'h. " t ~ 3 > n fJ"n T ^hf^h where an and ah represent actions for new calls and handoff calls, respectively. They take a binary value, i. e. with 1 for accepting a call and 0 for rejecting a call. • Reward function The call arrival event can be represented by a vector e consisting of two binary values, { (1,0), if new call arrival, (5.34) (0,1), if handoff call arrival The reward r(x, a), defined below, is earned if the system state is in state x and the CAC policy is configured as a. The reward function can be written as r(x, a) = wn(xn + en ■ an) + wh{xh + eh ■ ah), (5.35) where wn and Wh are weighting factors for new and handoff calls, respectively. 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The transition probability from state x with action a to state y can be written as an ■ Xn ■ r(x,a), if y = x + (1,0), ah -Xh - r(x,a), if y = x-P (0,1), P (y|x > a) = xn ■ • r(x,a), if y = x - (1,0), (5.36) xh - Hh- t(x, a), if y = x - (0,1), 0, if y = x. 5.5.2 M od ified Linear P rogram m ing By using the uniformization technique, we can derive a modified linear programming algorithm associated with the MDP for the maximum reward function. It is given below with decision variables 7r(x, a), x £ X and a € A*. Maximize E E r(x, a)r(x, a)7r(x, a) xeX aeAx (5.37) subject to E ^ a ) - E E p(y lx> a )7 r ( xJ a) = 0,Vy e X (5.38) Q-G-Ay x.& X E E t ( x , a)7r(x, a) = 1, x e X a e A x 7 r(x, a) > 0, x e l , a. E Ax (5.39) (5.40) The variables 7r(x, a) satisfying (5.38)-(5.40) can be viewed as the steady-state probabilities of being in state x and choosing action a. The optimal action in each 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. state can be chosen among all actions in each state since the value of tt(x, a) will be zero for all but one action in each state. This implies that the optimal policy is non randomized admission policy and the action a chosen for each state is a deterministic function of state x. 5.5.3 N u m erical R esu lts for C A C P o licy Trend In this section, the CAC policy on two types call is investigated. Compared with previous work in [54, 56], our work is focused on how the optimal CAC policy varies with traffic parameters. The optimal CAC policy is illustrated (with the total system capacity equal to a bandwidthe of 10 units) in Figures 5.1 and 5.2 for light and heavy traffic scenarios, respectively. 5.5.3.1 L ight Traffic Scenario If the total service rate is larger than the total arrival rate, the system is in the light traffic scenario. In such a case, the CAC policy rejects to accept calls only when the system exceeds its capacity limit. The resulting CAC policy performs similarly to the “complete sharing scheme” as shown in Figures 5.1 (a), (b) and (c). In Figure 5.1 (a), service rates pn — ^ = 1 are greater than A n = Ah = 0.5. Similarly, in Figure 5.1 (b), the service rates jin = Hh — 2 are greater than A„ = \ = 1. In Figure 5.1 (c), the service rates fJ -n — ^h = 1 are greater than A„ = A /, = 0.5. 1 2 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2-dimension Cali Admission Control 2-dimenaion Cali Admission Control \: New,Handoff] }: Accept,Aacept :: Accept,Reject : Reject.Accept - : Reject, Rejacl >: New,Handoff] >: Accept,Accept t: Accept, Reject 1: Rejec!,Accept h: Reject,Reject (a)A n — A h 0.5 j ( i n j-i} i 1 2-dimension Call Admission Control New .Handoff] Accept,Accept Aocept.ReJec! Reject,Accept Reject, Reject (b )A n — A^ — lj/^ n f t h 2 2-dlmension Call Admission Control i : New,Handoff] i : Accept.Accept :; Acccpt,Reject : Reject,Accept y : Reject,Reject (c)A n — 2, \ h 0.5, jJjn f l h 1 (d) An — 0.5, A/j — 2,/x„ Figure 5.1: Optimal CAC policies in the light traffic scenario. 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Because of the use of weighting factors wn = 1 and wh — 10, the dominant parameter is the handoff arrival rate, The comparison between Figure 5.1 (c) and (d) provides a clear proof. In these two cases, their service rates are kept the same as those in (a), i.e. fin — Hh = 1- Even though the new call arrival rate A n is increased to 2 in Figure 5.1 (c), a CAC policy similar to th at in (a) is obtained due to the unchanged handoff arrival rate (Ah = 0.5), which is the dominant arrival parameter. On the other hand, if Ah, the dominant parameter, is increased as shown in Figure 5.1 (d), the optimal CAC policy becomes the guard channel type as discussed later. 5.5.3.2 Heavy Traffic Scenario If the total service rate is smaller than the total arrival rate, the system is in the heavy traffic scenario, and the optimal CAC policy should be designed based on the guard channel (GC) scheme as shown in Figures 5.2 (a) and (b). In Figure 5.2 (a), service rates fxn = \ih = 0.5 are less than arrival rates An — Ah = 1. Similarly, in Figure 5.1 (b), service rates fj,n = jih — 1 are less than arrival rates An = A ^ = 3. It is worthwhile to point out that, because of the use of weighting factors wn = 1 and Wh — 10, the dominant parameter is the handoff arrival rate Ah- Figures 5.2 (c) and (d) confirms this claim, where the service rates fj,n = ji = 1 and the dominant parameter Ah = 3 are kept the same as those in (b). The increase of the new call arrival rate, as done in (c) with A n = 5 or the decrease of it, as done in (d) with An — 0.8, does not have an impact on the guard channel nature of the CAC policy. 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2-dimension Call Admission Control 2-dimension Call Admission Control i : New.Handoff] i : Accept,Accapt t : Accept, Reject : Reject,Accept i-: Reject,Reject t : New.Handoff) >: Accept,Accept i : Accept, Reject : Reject.Aceept ► : Reject,Reject 1 2 3 4 3 (a')An = A * . = 1, /in = nh = 0.5 2-dimension Call Admission Control t : New.Handoff] i : Accept,Accept :: Aecapt.Reject : Reject,Accept •: Reject,Reject (b)An \}l 3,/in {ifi 1 2-dimension Call Admission Control New.Handoff] Accept,Accept Accept, Reject Reject.Aceept Reject,Reject (c)'^n 5, A /. 3, fjin 1 (cl)An — 0.8, Xh — 3, fin fJ'h Figure 5.2: Optimal CAC policies in the heavy traffic scenario. 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The advantage of using MDP is that MDP provides an accurate number of guard channels and a finer level of policy adjustment. 5.5.3.3 D iscussion on R elationship betw een CAC Policies and Traffic C onditions Figure 5.3 show how the optimal CAC policy changes with a different handoff service rate fih under the heavy traffic condition. Sufficient Resource As shown in Figure 5.3 (a), faster service rates (larger /x/, and /x„) help the system to have enough resource even in a heavy traffic situation. A larger number of fih increases the system reward more efficiently due to a higher weight of the handoff service rate /x/, on the reward function. As a result, the system blocks both calls only when all resources are in use. In-sufficient R esource Figures 5.3 (b) to (d) show the evolution of slowing down the handoff service rate /p* from 3 to 2 and then to 1. Smaller service rates result in longer system occupation time and reduce the average residual system resource. The effect of slower handoff service rate /x contributes to a lower reward function. As the service rate decreases, the CAC policy will change according to the following two rules to optimize the system reward under the heavy traffic situation. • Rule 1 (General Trend of the CAC Policy): 130 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2-dimension Cali Admission Control 2-dlmansion CaH Admission Control [a : New.Handoff] a ; Accept,Accept s : Accapt,Reject * : Reject.Aceept + ■ : Reject,Reject >: New.Handoff] i : Accept,Accept (: Accept,Reject : Reject.Aceept h: Reject,Reject 5 6 7 (&)nh = 4 2-diroension Cali Admission Control i : New.Handoff] i : Accept,Accept i : Accept,Reject : Reject,Accapt -: Reject,Reject 2 3 4 5 6 7 (b )fih = 3 2-dlmension Call Admission Control ■: New.Handoff] >: Accept,Accept c: Accept,Reject : RejecLAccept i-: Reject.ReJect (c)fih = 2 (d )nh Figure 5.3: Optimal CAC policies for (a) p* = 4, (b) fih — 3, (c) p,h — 2, (d) /j,h — 1, with the total system capacity equal to C = 10 unit bandwidth, A„ = Xh = 4 and M n 1- 131 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. When the resource becomes in-sufficient due to heavy arrivals and/or slow service rates, the general trend of the CAC policy is to have two types of states: (1) reject all calls when system states satisfy x={(:r„, Xh)\xn+Xh — C = 10}, and (2) accept only handoff calls for states x = { ( x n,x h)\xn + Xh = i,t < i < 9}, where parameter t is called the CAC policy trend parameter. • .Rule 2 (Fine-Tuning of the CAC Policy): On top of the general trend, we also observe some fine tuning in the CAC policy according to the number of pre-occupied handoff calls. To give an example, let us focus on those states satisfying x = { ( x n . Xh)\xn + Xh = t = 9} as shown in Figure 5.3 (b) (/x^ = 3). States (xn,Xh) = (4,5), (3, 6) have a different CAC policy. This state-dependent CAC policy only occurs with the MDP model and will not be allowed in traditional GC schemes. If all parameters are fixed except new call service rate f i n , the CAC policy follows the same general trend, but differs in the fine-tuning rule along boundary states. For those states satisfying x = {(:!;„, Xh,)\xn+Xh = t — 8} as shown in Figure 5.4 (a), states ( .xn, xh) = (6, 2) and (5, 3) have a different policy since the CAC policy would like to reject new calls only when the system is not served fast enough (no enough resource). With the same total number of occupied calls, the former state (6,2) consists of more 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2-dimension Call Admission Control 2-dimension CaH Admission Contro! i : Naw,Handoff| }: Accept,Accept (: Accept,Hejact : Reject,Accept k : Reject,Rejec! S ► O S t O 0 <Jtt O 0 o * * + 2 < ► O O O O O O 1(t o o o o o o o * (a)/in = 4 i : New,Handoff] i : Aocept.Accept ;: Accept,Reject : Reject.Accept 4 tt O O O o o o * 2tt Q O O O O * O O O O O 0 O 5 6 7 8 (b )/j,n = 3 Figure 5.4: Optimal CAC policy under heavy traffic with different new call service rates jin\ (a) = 4 and (b) /z„ = 3, where the total system capacity C — 10, A n = \ — 4 and — 1 are fixed.) fast service new calls and, therefore, the system resource in this state is considered sufficient to accept all calls. For every given state, we use the linear programming technique to compute the optimal value of 7r(x, a) that maximizes the reward function as given in (5.49). Then, an optimal action for each state can be determined by comparing values of 7 r(x, a) among actions of each state. These actions can be then tabulated a CAC policy, which is a mapping function from system states to actions. The policy is implemented in the OPNET simulator as a lookup table so that one can make the proper action based on the system state and the occurrence of the call arrival or departure event. Figure 5.5 shows the implementation steps for the MDP-LP policy (Markov Decision Process and Linear Program), it also shows the performance 133 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. , Traffic parameters Transition probability i A,p... P (x ,a ) , i s D.S.A.R.T. tt(x,a) : decision variables i Markov Decision Process OPNET CAC policy Guard Channel Scheme LP solutions Complete Sharing Scheme Modeling Figure 5.5: Implementation steps for the MDP-LP CAC policy, comparison of the MDP-LP policy with other two schemes, i.e. the complete sharing (CS) scheme and the guard channel (GC) scheme with OPNET. 5 .5.4 P erform ance C om parison 5.5.4.1 Traffic Param eters The Poisson call arrival rate and the exponentially distributed call holding time are assumed in the experiment. The call arrival rate is controlled directly by \ n and Ah for the new call and hard handoff arrivals, respectively. The call holding time for a new call and a hard handoff are directly controlled by 1 //j,n and fih, respectively. A large value of A results in the increment of the network traffic load. Values used in 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the simulation models are listed below. • Xn = 4 : the arrival rate of new calls; • A /j = 3.1 : the arrival rate of hard handoff calls; • [A n — 1 : the service rate of new calls; • Hh — 1.4 : the service rate of hard handoff calls; • C — 10 : the total system bandwidth units; • wn — 1, Wh = 10 : weighting factors for new call and hard handoff. We compare the performance of MDP-LP with various guard channel schemes GC(TH) with different threshold value TH = i, i G 0,1,2, 3. Let us describe the call admission control policy in GC(TH) scheme as follows. If there is less or equal to TH channels left, only hard handoff calls can be admitted. The complete sharing scheme is equivalent to the guard channel scheme with 0 unit guard channel, which is denoted as G C ( 0 ) . For the CDMA system, the “channel” is replaced by the term “unit bandwidth”. To approximate the definition given by (5.53), the QoS metric is measured by the average weighted reward < f> which is defined as I N e < t> = j r E r (x > a ) e rt= 1 I N e = TT ^ ' e" + Xn) + wh(ah ■ eh + xh)}, (5.41) e n= 1 where Ne is the total number of events in the simulation, including all arrival events e = (en, e^, r(x, a) is the reward function for the current event, x = (xn, x^) and 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2-dim ension Call Admission Control 7 (i 6 0 *= 5 ( 5 3(5 O * o o o o o o [ a : New,Handoff] o : Accept,Accept x : Accept,Reject •j-: Reject,Reject • a t - O * o o o o o o o o o o 1(5 O o o o + -$■ + o 0 1 Figure 5.6: The MDP CAC policy for the 2D Case, a = (an, ah) are the current state and the action chosen by CAC, respectively, where 0 < xn,Xh < C and an and an take the binary value. The optimal policy derived by the MDP-LP decision process is shown in Figure 5.6. Each point represents one possible system state of x = (xn, Xh), 0 < xn, Xh < C. The action vector a = (an, %) chosen for each state is represented by a different symbol. In this example, four possible actions are found in all coordinates. If action a = (0,0) is chosen (symbol all types of calls are rejected. This happens at coordinates of x = {(xn,Xh)\xn + Xh = C}. If action a = (0,1) is chosen (symbol only the hard handoff call is accepted. Similarly, action a = (1,1) (symbol ’ o’) denotes accepting all types of calls, and action a = (1,0) (symbol ’ x ’) denotes accepting the new call only. 136 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. £ £ 3 E ® w > * C O 1 x : o > '3 5 V s MDP 0 \ 3 (1 } GC(2 ) O C V S ) ( \ \ x C A Q ) V - \ ... j ©ettt GC(2) | ■C 23-31 Simulation Time (a ) Simulation Time (b) Figure 5.7: The performace comparison by weighted reward (a) MDP-LP, GC(0), GC(1), GC(2) and GC(3) schemes, (b) a close look at the performance among MDP- LP, GC(1) and GC(2) schemes. The average weighted reward, as defined in (5.53), for different schemes are shown in Figures 5.7 (a) and (b). We see that the MDP-LP decision rule gives the best result when compared with all guard channels with various thresholds. Figure 5.7 (b) gives a close look at the performance comparison among MDP-LP, GC(1) and GC(2) schemes. The new call blocking and the handoff dropping probabilities under different control schemes are shown in Figures 5.8 (a) and (b), respectively. Results show that the proposed MDP-LP scheme optimizes the reward function using the optimal CAC policy for each system state. Such an optimal CAC policy provides a better controlling resolution that cannot be achieved by the GC scheme only. 137 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Simulation Time Simulation Time (a) (b) Figure 5.8: (a) The new call blocking probabilities for various schemes, and (b) handoff dropping probabilities for various schemes. 5.6 3-D M D P and O ptim al CAC P olicy 5.6.1 S y stem M od el In this Section, we propose a 3-D MDP model for the hybrid handoff system with three types of calls. They are new calls, hard-handoff calls and soft-handoff calls. The system capacity for CDMA with a hybrid handoff system is defined in Eq. (5.6). New, hard-handoff and soft-handoff call arrivals follow the Poisson distribution with parameters: 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A„ : the average arrival rate of new calls; Xs : the average arrival rate of soft handoff calls; A /j : the average arrival rate of hard handoff calls; /j,n : the average service rate of new calls; jj,„ : the average service rate of soft handoff calls; fX h- the average service rate of hard handoff calls. The proposed MDP model is defined by the following five components: the deci sion epochs, the state space, the action space, the reward function and the transition probabilities. The MDP model for the proposed hybrid handoff system is stated be low. • The decision is made only at the occurrence of a call arrival. Call arrival events include new call and handoff call arrivals. • The state space X is a set of all possible combinations of occupied channels of each type in the system, i.e. X = {x|x = (xn, xs,xh),xn > 0 ,x s > 0, xh > 0 ,x n + x 3+ x h < C}, (5.42) where xn, xs, Xh are numbers of new calls, soft handoff calls and hard handoff calls, respectively, and C = 2lj _ 139 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • The action space is a set of vectors consisting of three binary elements, i.e. A — {a|a = (a„, as, ah);an, as, ah e (O(reject), l(accept)}}, (5.43) where an, as, ah are actions for new calls, soft handoff calls and hard handoff calls, respectively. They take the value of 0 for rejecting and 1 for accepting that type of calls. The action space Ax for state x € X can be written as { a = (1,1,1)}, if x = (0,0,0), Ax =< {a = (0,0,0)}, if xn + xs + xh > C, (a|a — (an,a3,ah)-, an, a3, ah € {0,1}}, if otherwise. (5.44) Let r(x, a) be the sojourn time in the present state x G X when action a € Ax is chosen. The sojourn time can be expressed as r(x, a) = (5.45) where an, as, ah represent actions for new calls, soft and hard handoff calls, respectively. They take a binary value, i.e. with 1 for accepting a call and 0 for rejecting a call. 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • The reward r(x ,a ) of state x when action a is taken is expressed as r(x, a) = wn(xn + en ■ an) + wn(xs + es ■ as) + wh(xh + eh ■ ah), (5.46) where wn, ws and Wh are the weighting factors for each call type, respectively. When the weighting factors are equal to one, the objective reward function is to maximize system utilization. The call arrival event can be represented by a vector e consisting of two binary values, (1,0,0), if new call arrival, (0,1,0), if soft handoff call arrival (0,0,1), if hard handoff call arrival (5.47) The transition probability from state x with action a to state y can be written as P (y |x ,a ) = an -An - r(x, a), if y = x + (1,0,0), as ■ As ■ r(x, a), if y = x + (0 ,1, 0), ah ■ Ah ■ r(x, a), if y = x + (0,0,1), 2r n -/rn -r(x, a), if y = x - (1,0,0), xs -ijls ■ r(x, a), if y = x — (0 ,1, 0), xh ■ jih ■ r(x, a), if y = x - (0 ,0 ,1), 0, i f y = x. It is also shown in Figure 5.9. (5.48) 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5.9: Illustration of the transition probability P (y |x , a). 5.6.2 S olu tion via Linear P rogram m in g By using the uniformization technique, we can derive a modified linear programming associated with the MDP for the maximum reward function. It is given below with decision variables 7r(x,a), x € X and a € Ax. Maximize E E r(x, a)r(x , a)7r(x, a) (5.49) xex aeAx subject to E ^ a ) ~ E E P(y lx> a )7r(x )a ) = ° > V y e X (5.50) a e A y xex a e A x E E T ( x'a W x> a ) = 1> (5-51) xeX aeAx 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7r(x, a) > 0, x € X, a 6 Ax. (5.52) The variables 7r(x, a) satisfying Eqs. (5.50)-(5.52) can be viewed as the steady- state probabilities of being in state x and choosing action a. The optimal action in each state can be chosen among all actions in each state since the value of 7r(x, a) will be zero for all but one action in each state. This implies that the optimal policy is non-randomized admission policy and the action a chosen for each state is a deterministic function of state x. For every given state, we use the linear programming technique to compute the optimal value of 7r(x, a) that maximizes the reward function as given in Eq. (5.49). Then, an optimal action for each state can be determined by comparing values of 7r(x, a) among actions of each state. These actions can be then tabulated a CAC policy, which is a mapping function from system states to actions. The policy is implemented in the OPNET simulator as a lookup table so that one can make the proper action based on the system state and the occurrence of the call arrival or departure event. Similar to 2D case, Figure 5.5 also shows the implementation steps for the MDP-LP policy (Markov Decision Process and Linear Program), it also shows the performance comparison of the MDP-LP policy with various schemes, i.e. the complete sharing (CS) scheme and the guard channel (GC) schemes using OPNET simulator. 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.6.3 N u m erical R esu lts 5.6.3.1 Traffic Param eters The Poisson call arrival rate and the exponentially distributed call holding time are assumed in the experiment. The call arrival rate is controlled directly by A n, A /, and A s for the new call, hard handoff and soft handoff arrivals, respectively. The A values are in the units of calls per minute. The mean request arrival rate is measured in the number of connections per minute. The call holding time for a new call is directly controlled by 1 /fin. A large value of A results in the increment of the network traffic load. The call holding time for hard and soft handoff calls is controlled by p* and fis, which are the service rate for hard and soft handoff calls, respectively. Values used in the simulation models are listed below. • A n = 2.5 : the arrival rate of new calls; • Xg = 2.5 : the arrival rate of soft handoff calls; • Xh = 2.5 : the arrival rate of hard handoff calls; • fin = 0.5 : the service rate of new calls; • = 1 : the service rate of soft handoff calls; • yih = 1.5 : the service rate of hard handoff calls; • C = 10 : the total system bandwidth units; • wn = 1, ws = 5, Wh — 10 : weighting factors for new, soft handoff and hard handoff calls. 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. S Q fi_ h a n tjo ff_ a riiv a l h a rd £ h a n d d o ff_ arriv a l new_call_ges^ rsc_martager Figure 5.10: The OPNET simulation system. 5.6.3.2 O PN E T Im plem entation Simulations were conducted by using the OPtimized Network Engineering Tool (OP NET) [10], and the CAC policy was evaluated in a distributed manner. Figure 5.10 shows the simulation system. Three call generating processes are deployed to generate traffics for new calls, soft handoff calls and hard handoff calls, respectively. The resource management module is responsible for resource manage ment, where call admission control (CAC) is included in this module. Figure 5.11 (a) and (b) illustrate the call generator process and the resource management process, respectively. Detail state transition diagrams for them are depicted in these figures. For each call type, a call generation process is generating calls according to Poission distribution with a given traffic parameters as shown in Figure 5.11 (a). Each call’s birth and death is controlled by the call generating process while the call admission control is governed by the resource management process as shown in 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5.11 (b) according to the CAC policy such as the complete sharing scheme, the guard channel scheme and the optimal control policy. Events and states are listed below to describe the finite state machine of the resource management process. In addition to the IDLE state, there are six states: (1) the N EW -C A C state, (2) the CACLSOFT-HANDOFF state, (3) the CACJfARD-HANDOFF state, (4) the N EW _C A L L _T E R M state, (5) the DEPART_SOFT_HANDOFF state and (6) the DEPART-HARD-HANDOFF state. Events trigger the resource management process to transit from one state to the other as explained below. • Event NEW _ARRIVAL occurs when a new call is requesting access to the target cell. This event will trigger the transition from the IDLE state to the N EW -C A C state. The admission control for the new call will then be performed. The system returns to the IDLE state after the admission decision is made. • Event SO FT -A R R IV A L occurs when a soft handoff call is requesting access to the target cell. This event will trigger the transition from the IDLE state to the CAC_SOFT_HANDOFF state. The admission control for soft handoff call will then be performed. The system returns to the IDLE state after the admission decision is made. • Event H ARD -A RRIV A L occurs when a hard handoff call is requesting access to the target cell. This event will trigger the transition from the IDLE state to the C A CLH ARD-H ANDO FF state. The admission control for hard handoff call will 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. then be performed. The system returns to the IDLE state after the admission decision is made. • Event N EW _TER M occurs when a new call is terminating. This event will trigger the transition from the IDLE state to the N EW _C A LL_TE R M state. The system resource used by that new call will be released. The system returns to the IDLE state afterwards. • Event SO FT JD E PA R T occurs when a soft handoff call is handed off to an other cell or terminates. This event will trigger the transition from the IDLE state to the DEPART_SOFT_HAND0FF state. The system resource used by that soft handoff call will be released. The system returns to the IDLE state afterwards. • Event H A R D JD E PA R T occurs when a hard handoff call is handed off to another cell or terminates. This event will trigger the transition from the IDLE state to the DEPART_HARDJH ANDQFF state. The system resource used by that hard handoff call will be released. The system returns to the IDLE state afterwards. 5 .6 .3 .3 P e r fo r m a n c e C o m p a riso n Let us compare the performance of MDP-LP with two other schemes, i.e. the complete sharing scheme (CS) and the guard channel scheme with two thresholds, denoted by GC(THi, THi). In the simulation, we used TH\ = 1 and TH 2 = 3 or 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 1 \ \ \ V (DISABLED)\ (PA C K ET _G EN ER A TE )/ss_packeLgenerate(); X7 (STOP) \ Q (a) The call generating process GACvSDFT ( N E W jiC A C ) (S(jlFT_ARRlVAL) TNEW _ARRIVAL) (HARO.ARFiiVAU DEPART. S [SCFTDEPARTj _ (htW TERM) [HARD DEPART’) (b) The resource management process Figure 5.11: The OPNET simulated processes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. GC(1,3) for comparison. With a total of C — 10 unit bandwidths, the call admission scheme of GC(1,3) does not block any types of calls until there are 3 channels left. If there are less or equal to TH 2 = 3 channels left, only soft handoff and hard handoff calls can be admitted. If there is less or equal to THi = 1 channel left, only hard handoff calls can be admitted. For the CDMA system, the “channel” is replaced by the term “unit bandwidth” . To approximate the definition given by (5.11), the QoS metric is measured by the average weighted reward ( j> which is defined as 1 N* < t> = TT J 2 r (x > a ) e 7 1 = 1 1 Ne = TT X] K K ■ en + Xn) + ws(as ■ es + xs) + wh(ah ■ eh + xh)}, (5.53) 7 7 = 1 where Ne is the total number of events in the simulation, including all arrival events e = (en, es,eh), r(x, a) is the reward function for the current event, x = (xn,xs,Xh) and a = (an, aSj a*) are the current state and the action chosen by CAC, respectively, where 0 < xn, xs,Xh < C and a„, as and ah take the binary value. The optimal policy derived by the MDP-LP decision process is shown in Fig ure 5.12. Each point represents one possible system state of x = (xn,xs,Xh), 0 < xn, xs,Xh < C. The action vector a — (an, a3, ah) chosen for each state is represented by a different symbol. In this example, four possible actions are found in all coordinates. If action a = (0,0,0) is chosen (symbol ’ x ’), all types of calls are rejected. This happens at coordinates of x = {(xn, xs,Xh)\xn + x 3 + Xh = C}. 149 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5.12: The MDP CAC policy. If action a = (0,1,1) is chosen (symbol ’*’), only soft and hard handoff calls are accepted. Similarly, action a = (1,1,1) (symbol ’o’) denotes accepting all types of calls, and action a = (0, 0,1) (symbol ’+ ’) denotes accepting hard handoff only. The normalized average weighted reward, as defined in Eq. (5.53), is shown in Figure 5.13. We see that the MDP-LP decision rule gives better results than the complete sharing (CS) scheme and the GC(1,3) guard channel scheme. As mentioned in Section 5.1, we can view the multiple-threshold guard channel scheme as a control subset of the proposed high resolution decision rule based on system states, whose solution was obtained by linear programming and proved to be optimal as done in [51] and [60]. 150 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. _ I I I I I I Z b 1h 2h 3h 4h 5h Simulation Time (hour) Figure 5.13: The average reward function. I ¥ f t h 9 1 £ X 0 0 0 Simulation Time (hour) (a) s __ ( 1 ) ( 1 > - ( 2 ) — N e » c a l S o f l h a n M ( 3 ) — H a r d h a r d o f f ( 2 ) 1 1 ! 1 ( 3 ) i i i Sim ulation Time (hour) (b) ! M a c (1 ) (1)- N e w call (2)— Sotthandoff (3)— Hard handoff w O l ! h I Simulation Time (hour) (c) Figure 5.14: The blocking probablities for (a) the complete sharing scheme, (b) the GC(1,3) guard channel scheme and (c) the M D P stationary policy. 151 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The dropping probability for each class under different control schemes is shown in Figure 5.14. Figure 5.14 (a) shows that the blocking probabilities are at a similar level because system resources are completely shared among all types of users. Fig ure 5.14 (b) shows that the GC(1,3) guard channel scheme provides a preferential treatm ent to hard and soft handoff users. Such a preferential treatm ent reserves more resources for the handoff usage so that it leaves a part of the system idle in some cases if not controlled carefully. On the other hand, the proposed scheme as shown in Figure 5.14 (c) optimizes the reward function well using the optimal CAC policy for each system state. 5.7 C om plexity of th e M arkov D ecision Process using Linear Program m ing Approach The complexity of the Markov Decision Process was discussed in literature [61], [62], [63], [64],[65] and [66], In this section, we focus on the complexity of the Lin ear Programming method. We consider two parameters related to the computation complexity: (1) the number of total states, 1 V X , and (2) the number of total deci sion variables, Nw. Since the optimal CAC policy for each state is obtained from the Linear Programming solution by referencing the value of decision variables, NT determines the problem size and the computation complexity of the MDP problem. 152 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.7.1 C om p lexity Issue for Linear P rogram m in g Let X denote the state space consisting of k types of calls in the system, i.e. X = {x|x = (x1,x2, • ■ ■ ,xk)]Xi, ■ ■ ■ ,xk > 0 ,J 2 xi ^ G}, (5.54) i = 1 where Xi is the number of calls for call type i, and the maximum system capacity can be expressed as C = _ The action space Ax for state x 6 X can be written as {a =(1,1, - - •, 1)}, if x = (0, 0, ■ ■ • , 0), {a = (0,0, • • •, 0)}, if H h x i ^ C , {a|a = (a!,a2, • - ■ , af e ); ai, a2, ■ • •, af c € {0,1}}, Otherwise. Ax = (5.55) Let iVa(x) denote the total number of actions for state x, it can be written as JVa(x) = « 1, if x — (0,0, • • ■ ,0), 1, i f E t i Xi = C, 2k, O therw ise. (5.56) Following the same notation, let us first define the following vectors. The size of all following vectors is Nn x 1. For the indexing purpose, let r,, t , C j and 7 T ; be the 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. i-th indexed elements in r, r, c and 7r, respectively. Each is associated with possible (x, a) values. Thus, we have r = [ri,r2, " ' ) r i) ' Z — h , r 2,' ' j Vj, Q — - h n , r-2-72,' ’ • ; rsT i, ‘ ’ ■ 7 T = Kl) 7 r2 ) ' ' ' > T T j > ' • • , TbvJ2 We can map our problem in (5.19)- (5.22) into to a standard-form linear program as L P : min c 7r s.t. Gn = b, 7 T > 0. where 5(^+1)xi ^ [0,0, • ■ •, 0, l]r , and matrix G(wx4-i)xWT can be defined as the following form: PNxxNw - D jV ^ -X iV v r *(Nx+l)xN„ = 1 x J V , (5.57) 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where PjvlXw, is the state transition probability matrix with elements of P (y |x ,a ) defined in Eq. (5.13). Dnxxn* is an block diagonal matrix with diagonal blocks lu of all ones, and zeros for non-diagonal blocks lij,Vi ^ j. Thus, ^11 o ................... Q I22 0 ............ D NxxN, = ............ 0 la 0 2 In*Nx (5.58) where if there are k types of calls, the diagonal block 1 ^ is a vector of all ones of the following dimension, i lxl, if indexed i-th state satisfies x = (0, 0, • ■ •, 0), or En=l xn = C, 1 lx2f c , Otherwise. (5.59) A well known solution using the Interior Point Method was proposed by Kar- markar [63]. This algorithm solves an LP problem in the polynomial time, and has been adopted by Matlab for large-scale LP problems. The complexity merits for this algorithm are usually expressed as an upper bound of the magnitude of the 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. number of arithmetic operations, and they are a function of the problem size. One can interpret the size as a measure of the computation loading involved. It is obvious that computation loadings are different if the elements in G, 7r, and b are different. The more 0 and 1 elements in equations, the less computation involved. In order to capture the problem size denoted by L, it requires an agreed upon definition, which results in rather subtle work than just the number of decision variables (Nn) and constraints (Nx + 1). One commonly accepted measure of the size of an LP problem is L os (Nx + 1) • Nv + |"log2 |v|], where v is the product of all of nonzero elements in the problem (in matrix G, vector 7r, and b) [67], For this measure, these matrices and vectors are assumed to contain only integers (which can always be achieved by appropriate scaling) and [log2 |u|] reflects the number of bits needed to represent the problem in binary notation for a computer. A numerical illustration of the problem size can be found on pages 494-496 in Miller’ s book [67], Karmarkar’s algorithm can solve the LP problem in the order of 0(N 7 C 4L), where L is the size of the problem. The overall complexity for Karmarkar’s basic algorithm requires 0(N 1 T 4L) operations in ”big-0” notation. Here, the opera tion is referred to as arithmetic operations and comparisons in infinite precision as described by Anstreicher [68], in which a comprehensive survey on the complexity issue for other algorithms was presented. We summarized his comparison results in Table 5.1. 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5.1: Complexity for Solving LP Algorithm Complexity Authors Projective (basic) o (N w 4L) Karmarkar [63] Projective partial updating 0 (N J -5L) Karmarkar [63] Path following 0 (N w 3- 5L) Renegar [64] Partial updating + Path following 0 (N v3L) Gonzaga [65] and Vaidya [66] Partial updating + Preconditioned conjugate gradient Anstreicher [68] 5.7.2 D erivation o f th e P rob lem Size 5.7.2.1 The 2D Case In a system with two call types and of capacity C, and the total number of actions for each state can be represented as in Eq. (5.56), in which k = 2. The total number of decision variables Nx can be derived as Nx = 1 + 2 + • • ■ + (C + 1) = +1^ C - + 2 ). (5.60) Figure 5.15 shows the state diagram of the 2-D MDP model, where each node stands for a state. Next, we determine the number of decision variables in the model. To reduce the number of decision variables, we make two reasonable assumptions for boundary states: • At state x = (0, 0), accept all call types i.e., a = (1,1). • At states with xn + Xh = C , reject all incoming calls i.e. a = (0, 0). 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X„+Xh ~ @ x„ +xi,=l x„+xh~2 • • • Xn +Xh =^ ' o o Figure 5.15: The computation of the total number of states N x in a 2-D MDP model. Table 5.2: Total number states and decision variables in a 2-D MDP model c Ax Nv C Nx 5 21 63 30 496 1888 10 66 228 40 861 3318 15 136 493 50 1326 5148 20 231 858 60 1891 7378 Thus, only one decision variable is allowed for each boundary state. For the remain ing states, there are 22 = 4 decision variables, each for one possible action in that state. The total number of decision variables can be computed by N v = l + _ i ) . 22 + (C + 1). (5.61) Table 5.2 shows the values of jV x, N* with different system capacities C. 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.7.2.2 The 3D Case In a system with two call types and of capacity C, and the total number of actions for each state can be represented as in Eq. (5.56), in which k = 3. The total number of decision variables N x can be derived as N x = Number of states on surface S(xh = C) + Number of states on surface S(xh = C — 1) + ........ + Number of states on surface S(xh — i) + Number of states on surface S{xh = 0) 1-2 2 -3 (C + l) • (C + 2) 2 + ~ 2~ + ' " + 2 ° (C - i + 1) ■ (C - i + 2) — ^ 2 i= C z (C + 1) • (C + 2) • (C + 3) 6 (5.62) Note that there are (C —i+ l) ( C — i+ 2)/2 states on the surface S (xh — i), 0 < i < C. Figure 5.16 gives the state diagram of a 3D MDP model, where each node stands for a state. Next, we determine the number of decision variables in the proposed model. In order to reduce the number of decision variables, we make two assumptions for boundary states: 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If, V ! I i -J o i., (0,0,0) c a ,0,0) (2,0,0) C ) (C,0,0) Figure 5.16: The total number of states iV x in a 3-D MDP model • At state x = (0,0, 0), accept all types of call, i.e. a = (1,1,1). • At states where x n + x s + x h — C, reject all incoming calls, i.e. a = (0, 0, 0). Therefore, there is only one decision variable for each boundary state. For the remaining states, there are 23 = 8 decision variables, each for one possible action in that state. As a result, The total number of decision variables is W „ = l + (M ^ .± g 1^ g + 2) — 1). 2s + { C + 1 ^C + 2 \ Table 5.3 shows the values of N x and Nx under different system capacities. 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5.3: Total number states and decision variables in a 3-D MDP model c iV x K C iV x N* 5 56 294 30 5456 40169 10 286 1819 40 12341 92694 15 816 5569 50 23426 178119 20 1771 12544 60 39711 304444 5.7.2.3 C om plexity D iscussion for Higher D im ension In a state with k call types and of capacity C, the total number of actions for each state is N& = 2k and the total number Nx of states is of 0{Ck). For real time implementation, a large number of C result in a larger amount of memory and computational time requirements in a polynomial of order k. In the case of a higher dimension (say k > 4), an approximation model has to be developed to reduce the computational complexity. A good approximation model should find a good balance between accuracy and efficiency. 5.8 Conclusion and Future Work Effective radio resource management schemes depend greatly on the CAC policy. The MDP was used to construct a system model, and the linear programming (LP) technique was used to find the optimal CAC policy at a given state in this chapter. We considered the CAC policies for a system with the hard handoff only or with a hybrid handoff process, which includes new, soft handoff and hard handoff calls. 161 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The state concept in the traditional multiple threshold guard channel schemes can be viewed as a special case of the system state model discussed in this work. Our proposed scheme has a higher controlling precision. It was shown by computer simulation that the proposed scheme outperformed the complete sharing and the multiple threshold guard channel schemes. Our work focused on the unconstrained Markov decision problem. The goal was to maximize the average weighted system reward without a constraint imposed on the dropping probability. If the dropping probability has a hard constraint, the CAC policy can be randomized as described in [50]. In other words, the policy in a specific state will choose a certain action based on a probability distribution rather than having a predicted action. In the future, it is worthwhile to consider CAC policies under some dropping probability constraints. The tradeoff between hard and soft handoff calls should also be studied. 162 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C hapter 6 Conclusion and Future Work 6.1 S u m m ary o f th e R esearch In this research, we addressed the problem of efficient resource management in or der to provide Quality of Service (QoS) and preferential treatm ent to traffic with different QoS requirements. The first part of the research (Chapter 3) focused on the development of an efficient dynamic resource management for channel-based TDMA/FDMA systems. The second part of the research (Chapter 4) considered a dynamic call admission control system. The proposed system is suitable for interference-based CDMA systems by the use of interference guard margin (IGM). The third part of the research (Chapter 5) examined an optimal stationary call ad mission control (CAC) scheme based on the semi-Markov decision process (SMDP) model and the linear programming (LP) techniqes under different handoff processes. The proposed system is suitable for channel-based as well as interference-based sys tems. 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In Chapter 3, we presented dynamic CAC and associated resource reservation schemes based on the concept of guard channels to adapt the resource access prior ity by the signal-to-noise ratio (SNR) and the distance information of the potential higher-priority calls in the neighboring cells, which are likely to handoff. Under light as well as heavy traffic conditions, our CAC scheme outperformed the fixed GC scheme. The cases with different traffic profiles of mobile terminals under var ious traffic conditions were also discussed. We considered a comprehensive service model, which includes mobile terminals’ bandwidth requirements and their different levels of priority, rate adaptivity as well as their mobility. Our RR scheme provides more accurate estimation of potential higher-priority call arrivals, thus increasing the system reward while providing QoS guarantees to higher-priority calls. The higher system reward implies that our proposed scheme can get a good balance between resource sharing and resource reservation to achieve the opposing goals of accommodating more calls while providing QoS guarantees for high-priority class connections. In Chapter 4, the interference-guard margin (IGM) approach has been developed by following the idea of the GC scheme. It turns out that there is a straightforward mapping from GC to IGM using the loading factor concept. A comprehensive service model was adopted. The model covered different bandwidth requirements, mobilities of mobile terminals, flexible service rates, as well as priority classes. The QoS perfor mances in terms of several system objective functions were evaluated in the presence 164 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of various traffic characteristics. This research examined the provision of connection- level QoS from the viewpoint of the service provider (i.e. the system operator) as well as mobile terminals (i.e. users). From the perspective of the service provider, the degree of QoS is evaluated in terms of system utilization and maximum rewards. From the perspective of users, QoS is measured in terms of the new call blocking probability and the handoff dropping probability. Mathematical models for the con ventional fixed GC scheme was extended to multi-threshold GC schemes to take care of the scenario with traffic of multiple priority classes. Advanced dynamic schemes were also studied by OPNET simulation. In the proposed model, we extended the analysis and simulation of traditional wireless communication networks with only one type of application (voice) to multimedia applications with different bandwidth requirements. Besides, we examined an important feature, i.e. rate-adaptability, for emerging multimedia compression schemes. The QoS performance was studied in the presence of rate-adaptive applications under different traffic scenarios. Based on the proposed service model and the conducted simulation analysis, we extended the preferential treatm ent to the 3G CDMA system. The main tool to bridge the channel-based TDMA/FDMA system and the interference-based CDMA system is the loading factor concept. It serves as a good tool to relate the received service of a mobile terminal to the amount of system resource consumption. The loading factor converts a mobile user’s bandwidth, priority attribute and other characteris tics, such as rate-adaptability, into a practical loading increment. Consequently, a 165 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. preferential treatment can be realized in the CDMA system via the loading factor. Thus, the resource reservation scheme {i.e. the guard channel scheme) adopted by the TDMA/FDMA system can be applied to the CDMA system (i.e. interference guard margin scheme) in the same fashion. In Chapter 5, an effective radio resource management scheme was developed based on the semi-Markov decision process (SMDP) model and the linear program ming (LP) solution technique. They were used to find the optimal CAC policy at a given state. The proposed CAC scheme was designed for CDMA systems under a hybrid handoff scenario. The proposed stochastic control scheme determined the optimal stationary CAC policy for three traffic types {i.e. the new call, soft handoff and hard handoff). From the viewpoint of the precision of CAC, the state concept in the traditional multiple threshold guard channel schemes can be viewed as a sub set of the system state model discussed here. The proposed scheme has a much higher controlling precision. It was shown by computer simulation that the pro posed scheme outperformed the complete sharing and the multiple threshold guard channel scheme. . 6.2 Future Work There are some interesting topics to be studied in the near future as an extension of this research. 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6.2.1 E m bedded S y stem D esig n for R esou rce M anagem ent Embedded systems have received a large amount of interest in recent years. There are some unique characteristics that distinguish the embedded system from other computing systems [69]. First, an embedded system is often dedicated to a single function rather than general-purpose tasks. Second, an embedded system usually has tighter constraints in the cost, size and power constraints. Third, it should process tasks and respond to actions in real time. Pagers, PDA, satellite phones and cellular phones are some examples of embedded systems with functions described above. W ith the quick growth of Integrated Circuits (IC) capacity, which doubles every 18 month, it is possible to implement the resource management and call admission control functions on an embedded system. Several types of processors can be used to realize the functionality needed in the resource management embedded system, including general purpose processors (GPPs), single purpose processors (SPPs) and application specific instruction-set processors (ASIPs). The GPP has a larger size, but it often reduces the design effort and cost on required common functions. Using a G PP in an embedded system provides designer a great flexibility by working on software programming only (rather than a re-design of digital circuits). The SPP has a much faster processing speed because digital circuits are hard-wired and optimized to perform the target task. Using an SPP in an embedded system provides a better performance in terms of a faster processing speed and low power consumption, yet with little flexibility on 167 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. its functionality. The third type of microprocessors are ASP, it is designed for a class of similar functionality rather than a general purpose GPP, nor too specific as SPP. Therefore, ASIP can provide the benefit of flexibility while maintaining a good performance. ASIP is a good candidate for the hardware implementation of the proposed optimal call admission control algorithm. W ith the advantages of flexibility and a good performance, it can be applied widely in base stations for distributed CAC and mobile switching center (MSC) for centralized CAC systems. Figures 6.1 (a) and (b) give the I/O view of the optimal CAC cases using the Semi-Markov Decision Process (SMDP) model and the linear programming tech nique for 2-D and 3-D systems, respectively. The input traffic parameters for the 2-D case include system capacity C, the new call arrival rate A„, the handoff arrival rate Athe new call departure rate fxn and the handoff departure rate +/,. The clock (CLK) serves for the data update and other timing purposes. The CAC module also contains N x outputs, where each represents a call admission policy (action) for each state as described in Chapter 5. Recently, powerful compilers have made the embedded system design feasible with high-level processor-independent languages such as C, C + + . The integrated design environments (IDEs) significantly decreases the complexity of microprocessor code development and assembly language programming [69]. This trend has enabled an easy porting of our developed algorithms to hardware design. We have the 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CAC - • Vl.il* 'I Vj i = K - Hn - Ms - /** .... C7.il' CA C won-Lr J N x (b) I/O of 3-D CAC Embedded System. Figure 6.1: I/O of embedded CAC systems for (a) 2-D and (b) 3-D cases. software design ready, and the gate-level design shall be carried out with the help of synthesis tools and high-level languages in the near future. 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R eference List [1] Y.-B. Lin and I. Chlamtac, Wireless and Mobile Network Architectures. John Wiley < f e Sons, Inc, 2001. [2] J. Zander, S.-L. Kim, M. Almgren, and O. Queseth, Radio Resource Manage ment for Wireless Networks. Artech House Publishers, 2001. [3] C. E. Shannon, “A mathematical theory of communication,” Bell System Tech nical Journal, vol. 27, pp. 379-423, 623-656, July and October 1948. [4 ] ITU, “International mobile telecommunications-2000,” http://www.itu.org, 2000 . [5] B. Lab, “American telephone and telegraph company,” The Bell System Tech nical Journal, vol. 58, January 1979. [6 ] L. Harte, R. Levine, and S. Prokup, Cellular and PCS : The Big Picture. McGraw-Hill Series, 1997. [7 1 L. L. Peterson and B. S. Davie, Computer Networks: A Systems Approach. Morgan Kaufmann Publishers, 1999. [8 ] D. Hong and S. S. Rapport, “Traffic model and performance analysis for cellular mobile radiotelephone systems with prioritized and nonprioritized handoff pro cedures,” IEEE Transactions on Vehicle Technology, vol. 35, pp. 77-92, Aug. 1986. [9] H. Chen, S. Kumar, and C.-C. J. Kuo, “Differentiated QoS aware priority handoff in cell-based multimedia wireless network,” in IS& T /SP IE ’ s 12th Int. Symposium, Electronic Imaging 2000, vol. 3974, (San Jose, CA), pp. 940-948, January 2000. [10] I. Karzela, Modeling and simulating communication networks: a hands-on ap proach using OPNET. Prentice Hall, New Jersey, Auguest 1998. [11] H. Chen, S. Kumar, and C.-C. J. Kuo, “Radio resource management for cdma systems with interference guard margin (igm),” IS& T /SP IE ’ s International Symposium, Electronic Imaging 2002, Auguest 2002. 170 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [12] 3GPP2, “The third generation partnership project 2 (3gpp2),” http://www.3gpp2.org, 2000. [13] K. Richardson, “Umts overview,” Electronics & Communication Engineering Journal, vol. 12, pp. 93-100, June 2000. [14] T. O. a and R. Prasad, “An overview of third-generation wireless personal com munications: A european perspective,” IEEE Personal Communication Maga zine, 1998. [15] H. Aghvami and B. Jafarian, “A vision of umts/imt-2000 evolution,” Electronics & Communication Engineering Journal, vol. 12, pp. 148-152, June 2000. [16] Z. Zheng and W. Lam, “Analytical methods to calculate performance of hand off prioritisation in dynamic channel assignment,” Electronics Letters, vol. 37, pp. 978-979, July 2001. [17] S. Boumerdassi, “An efficient reservation-based dynamic channel assignment strategy,” First International Conference on 3G Mobile Communication Tech nologies, pp. 352-355, 2000. [18] P. O. Gaasvik, M. Cornefjord, and V. Svensson, “Different methods of giving priority to handoff traffic in a mobile telephone system with directed retry,” 41st IEEE Vehicular Technology Conference, Gateway to the Future Technology in Motion, pp. 549-553, 1991. [19] T. Kwon, Y. Choi, C. Bisdikian, and M. Naghshineh, “Call admission con trol for adaptive multimedia in wireless/mobile networks,” in IEEE Wireless Communications and Networking Conference, pp. 540-544, 1999. [20] S. S. Rapport and C. Purzynski, “Prioritized resource assignment for mobile cellular communication systems with mixed services and platform types,” IEEE Transactions On Vehicular Technology, vol. 45, pp. 443-458, Aug. 1996. [21] B. Epstein and M. Schwartz, “Reservation strategies for multimedia traffic in a wireless environment,” in Vehicular Technology Conference, 1995 IEEE 45th, (Chicago, IL), pp. 165-169, July 1995. [22] M. Naghshineh and M. Schwartz, “Distributed call admission control in mo bile/wireless networks,” IEEE Journal Selected Areas Communications, vol. 14, pp. 711-717, May 1996. [23] P. Ramanathan, K. M. Sivalingam, P. Agrawal, and S. Kishore, “Dynamic resource allocation schemes during handoff for mobile multimedia wireless net works,” IEEE Journal on Selected Areas in Communications, vol. 17, pp. 1270- 1283, July 1999. 171 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [24] A. S. Acampora and M. Naghshineh, “Control and quality of service provision ing in high-speed micro-cellular networks,” IEEE Personal Communications, no. 2nd Quarter, pp. 36-43, 1994. [25] A. Sutivong and J. M. Peha, “Novel heuristics for call admission control in cel lular systems,” 1997 IEEE 6th International Conference on Universal Personal Communications Record, vol. 1, pp. 129-133, 1997. [26] Y. B. Lin, “Performance modeling for mobile telephone networks,” IEEE Net work, vol. 45, pp. 63-68, Nov.-Dee. 1997. [27] W. C. Y. Lee, Mobile Cellular Telecommunications- analog and digital systems. McGraw-Hill, 2 ed., 1995. [28] K. Buchanan, R. Fudge, D. McFarlane, T. Phillips, A. Sasaki, and H. Xia, “IMT-2000: service provider’s perspective,” IEEE Personal Commun., vol. 4, no. 4, pp. 8-13, 1997. [29] R. Pandya, D. Grillo, E. Lycksell, P. Mieybegue, H. Okinaka, and M. Yabusaki, “IMT-2000 Standards: network aspects,” IEEE Personal Commun., vol. 4, no. 4, pp. 20-29, 1997. [30] J. Knutsson, P. Butovitsch, M. Persson, and R. D. Yates, “Downlink admission control strategies for CDM A systems in a M anhattan environment,” in Proc. VTC, vol. 2, pp. 1453-1457, May 1998. [31] C. Huang and R. Yates, “Call admission in power controlled CDMA systems,” in Proc. VTC, vol. 3, pp. 1665 - 1669, May 1996. [32] N. Dimitriou and R. Tafazolli, “Resource management issues for UMTS,” in Proc. 1st IEEE Int. Conf. 3G Mobile Commun. Technol., pp. 401 - 405, March 2000 . [33] Z. Liu and M. E. Zarki, “Sir-based call admission control for ds-cdma cellular systems,” IEEE Journal on Selected Areas in Communications, vol. 12, pp. OSS- Odd, May 1994. [34] K. Gilhousen, R. P. I.M. Jacobs, and et. al, “On the capacity of a cellular cdma system,” IEEE Transactions On Vehicle Technology, vol. 40, no. 2, pp. 303-312, 1991. [35] S. M. Shin, C.-H. Cho, and D. K. Sung, “Interference-based channel assign ment for ds-cdma cellular systems,” IEEE Transactions On Vehicle Technology, vol. 48, pp. 233-239, January 1999. 172 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [36] H. Holma and J. Laakso, “Uplink admission control and soft capacity with mud in cdma,” in Proceedings of V TC ’ 99 Fall, (Amsterdam, Netherlands), pp. 19-22, September 1999. [37] A. M. Viterbi and A. J. Viterbi, “Erlang capacity of a power controlled cdma system,” IEEE Journal on Selected Areas in Communication, vol. 11, no. 6, pp. 892-900, 1993. [38] J. Shapira and R. Padovani, “Microcell engineering in cdma cellular networks,” IEEE Transctions on Vehicular Technology, vol. 43, no. 4, pp. 213-216, 1994. [39] H. Holma and A. Toskala, WCDMA for UMTS, Radio Access For Third Gen eration Mobile Communications. John Wiley & Sons, June 2000. [40] I. JTC1/SC29/W G11, “Generic coding of moving pictures and associated audio information,” ISO /IE C International Standard 13818, Nov. 1994. [41] I. JTC1/SC29/WG11, “Overview of the MPEG-4 standard,” ISO /IE C N3747, Oct. 2000. [42] I. JTC1/SC29/W G1, “JPEG2000 part 1 final committee draft version 1.0,” ISO /IE C International Standard N l6f 6R, Mar. 2000. [43] W. C. Y. Lee, Mobile Cellular Telecommunications - Analog and Digital Sys tems. McGraw-Hill, 2000. [44] H. Chen, S. Kumar, and C.-C. J. Kuo, “Dynamic call admission control scheme for QoS priority handoff in multimedia cellular systems,” in IEEE Wire less Communications and Networking Conference, vol. 1, (Orlando, Florida), pp. 114-118, March 2002. [45] R. Ramjee, R. Nagaxajan, and D. Towsley, “On optimal call admission control in cellular networks,” in IEEE INFOCOM’ 96, pp. 43-50, 1996. [46] C. W. Clark, “The lazy adaptable lions: a Markovian model of group foraging,” Anim. Behav., vol. 35, pp. 361-368, 1987. [47] I. Houston and J. M. McNamara, “Singing to attract a mate - a stochastic dynamic game,” J. Theor. Biol, vol. 129, pp. 171-180, 1986. [48] R. Howard, Dynamic Programming and Markov Processes. MIT Press, 1960. [49] R. E. Bellman, Dynamic Programming. Princeton University Press, 1957. [50] M . L. Puterman, Markov Decision Processes: Discrete Stochastic Dynamic Pro gramming. John Wiley k, Sons, 1994. 173 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [51] H. C. Tijms, Stochastic Modelling and Analysis: A Computational Approach. John Wiley & Sons, 1986. [52] R. Rezaiifar, A. M. Makowski, and S. P. Kumar, “Stochastic Control of Handoffs in Cellular Networks,” IEEE J. Select. Areas Commun., vol. 13, pp. 1348-1989, 1995. [53] M. Sidi and D. Starobinski, “New call blocking versus handoff blocking in cel lular networks,” Wireless Networks, vol. 3, pp. 15-27, 1997. [54] C.-J. Ho and C.-T. Lea, “Finding better call admission policies in wireless networks,” in Proc. VTC, pp. 2135-2139, 1998. [55] J. Choi, T. Kwon, Y. Choi, and M. Naghshineh, “Call admission control for multimedia services in mobile cellular networks: a markov decision approach,” in ISCC 2000, pp. 594 -599, 2000. [56] Y. Xiao, C. L. P. Chen, and Y. Wang, “An optimal distributed call admission control for adaptive multimedia in wireless/mobile networks,” in IEEE, pp. 447- 481, 2000. [57] K. W. Ross and D. H. Tsang, “Optimal circuit access policies in an ISDN environment: A Markov Decision approach,” IEEE Trans. Commun., vol. 37, pp. 934-939, 1989. [58] R. E. Miller, Optimization. John Wiley & Sons, 2000. [59] A. H. M. Ross, The CDMA Technology Site, h ttp : / / www.amug.org/ ahmrphd/, 1996. [60] S. Ross, Introduction to Stochastic Dynamic Programming. ACADEMIC PRESS, INC, and Harcout Brace Jovanovich, 1993. [61] M. L. Littman, T. L. Dean, and L. P. Kaelbling, “On the complexity of solv ing Markov decision problems,” in Proceedings of the Eleventh Annual Con ference on Uncertainty in Artificial Intelligence (UAI-95), (Montreal, Quebec, Canada), pp. 394-402, 1995. [62] J. Goldsmith and M. Mundhenk, “Complexity issues in markov decision pro cesses,” in IEEE Conference on Computational Complexity, pp. 272-280, 1998. [63] N. Karmarkar, “A New Polynomial-Time Algorithm for Linear Programming,” Combmatorica, vol. 4, pp. 373-395, 1984. [64] J. Renegar, “A polinomial-time algorithm, based on Newton’ s method, for linear programming,” Math. Programming, vol. 40, pp. 59-93, 1988. 174 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [65] C. C. Gonzaga, “An algorithm for solving linear programming problems in 0 (n 3L) Operations,” Math. Programming, vol. 40, pp. 59-93, 1988. [66] P. M. Vaidya, “An algorithm for linear programming which requires O(((to + n)n2 + (m + n )l^ n ) L ) ” Math. Programming, vol. 47, pp. 175-201, 1990. [67] R. E. Miller, Optimization: Foundations and applications. Wiley-Interscienee, John Wiley & Sons, 2000. [68] K. M. Anstreicher, “Linear Programming in O (t^-L ) Operations,” SIAM J. OP TIM., vol. 9, pp. 803-812, 1999. [69] F. Vahid and T. Givargis, Embedded System Design: A Unified Hard ware/Software Introduction. John Wiley Sons, 2002. 175 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R eference List [3GP00] [AJOO] [AN94] [Ans99] [aP98] [Bel57] [BFM+97] [BouOO] [CKCNOO] [CKKOO] 3GPP2. The third generation partnership project 2 (3gpp2). http://www.3gpp2.org, 2000. H. Aghvami and B. Jafarian. A vision of umts/imt-2000 evolution. Electronics & Communication Engineering Journal, 12(3): 148-152, June 2000. A. S. Acampora and M. Naghshineh. Control and quality of service pro visioning in high-speed micro-cellular networks. IEEE Personal Commu nications, (2nd Quarter):36-43, 1994. K. M. Anstreicher. Linear Programming in O (j^-L) Operations. SIAM J. OP TIM., 9:803-812, 1999. - n" T. Ojanper a and R. Prasad. An overview of third-generation wire less personal communications: A european perspective. IEEE Personal Communication Magazine, 1998. R. E. Bellman. Dynamic Programming. Princeton University Press, 1957. K. Buchanan, R. Pudge, D. McFarlane, T. Phillips, A. Sasaki, and H. Xia. IMT-2000: service provider’s perspective. IEEE Personal Commun., 4(4):8-13, 1997. S. Boumerdassi. An efficient reservation-based dynamic channel assign ment strategy. First International Conference on 3G Mobile Communi cation Technologies, pages 352-355, 2000. J. Choi, T. Kwon, Y. Choi, and M. Naghshineh. Call admission control for multimedia services in mobile cellular networks: a markov decision approach. In ISC C 2000, pages 594 -599, 2000. H. Chen, S. Kumar, and C.-C. Jay Kuo. Differentiated QoS aware prior ity handoff in cell-based multimedia wireless network. In IS& T /SP IE ’ s 12th Int. Sym posiumElectronic Imaging 2000, volume 3974, pages 940— 948, San Jose, CA, January 2000. 176 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [CKK02a.] H. Chen, S. Kumar, and C.-C. Jay Kuo. Dynamic call admission control scheme for QoS priority handoff in multimedia cellular systems. In IEEE Wireless Comm,unications and Networking Conference, volume 1, pages 114-118, Orlando, Florida, March 2002. [CKK02b] Huan Chen, Sunil Kumar, and C.-C. Jay Kuo. Radio resource management for cdma systems with interference guard margin (igm). IS& T /SP IE ’ s International Symposium, Electronic Imaging 2002, Au guest 2 0 0 2 . [Cla87] C. W. Clark. The lazy adaptable lions: a Markovian model of group foraging. Anim. Behav., 35:361-368, 1987. [DT00] N. Dimitriou and R. Tafazolli. Resource management issues for UMTS. In Proc. 1st IEEE Int. Conf. 3G Mobile Commun. Technol., pages 401 - 405, March 2000. [ES95] B. Epstein and M. Schwartz. Reservation strategies for multimedia traffic in a wireless environment. In Vehicular Technology Conference, 1995 IEEE 45th, pages 165-169, Chicago, IL, July 1995. [GCS91] P. O. Gaasvik, M. Cornefjord, and V. Svensson. Different methods of giving priority to handoff traffic in a mobile telephone system with di rected retry. J is i IEEE Vehicular Technology Conference, Gateway to the Future Technology in Motion, pages 549-553, 1991. [GIJea91] K.S. Gilhousen, R. Padovani I.M. Jacobs, and et. a!. On the capacity of a cellular cdma system. IEEE Transactions On Vehicle Technology, 40(2):303-312, 1991. [GM98] Judy Goldsmith and Martin Mundhenk. Complexity issues in markov decision processes. In IEEE Conference on Computational Complexity, pages 272-280, 1998. [G0 1 188] C. C. Gonzaga. An algorithm for solving linear programming problems in 0 ( n 3L) Operations. Math. Programming, 40:59-93, 1988. [HL98] ChiJui Ho and Chin-Tau Lea. Finding better call admission policies in wireless networks. In Proc. VTC, pages 2135-2139, 1998. [HL99] H. Holma and J. Laakso. Uplink admission control and soft capacity with mud in cdma. In Proceedings of V T C ’99 Fall, pages 19-22, Amsterdam, Netherlands, September 1999. .............. [HLP97] Lawrence Harte, Richard Levine, and Steve Prokup. Cellular and PCS : The Big Picture. McGraw-Hill Series, 1997. 177 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [HM86] I. Houston and J. M. McNamara. Singing to attract a mate - a stochastic dynamic game. J. Theor. Biol., 129:171-180, 1986. [How60] R. Howard. Dynamic Programming and Markov Processes. MIT Press, 1960. [HR86] D. Hong and S. S. Rapport. Traffic model and performance analysis for cellular mobile radiotelephone systems with prioritized and nonpri oritized handoff procedures. IEEE Transactions on Vehicle Technology, 35:77-92, Aug. 1986. [HT00] Harri Holma and Antti Toskala. WCDMA for UMTS, Radio Access For Third Generation Mobile Communications. John Wiley & ; Sons, June 2000 . [HY96] C.Y. Huang and R.D. Yates. Call admission in power controlled CDMA systems. In Proc. VTC, volume 3, pages 1665 - 1669, May 1996. [ITU00] ITU. International mobile telecommunications-2000. http://www.itu.org, 2000. [JTC94] ISO/IEC JTC1/SC29/W G11. Generic coding of moving pictures and associated audio information. ISO /IE C International Standard 13818, Nov. 1994. [JTCOOa] ISO/IEC JTC1/SC29/W G1. JPEG2000 part 1 final committee draft version 1.0. ISO /IE C International Standard N I 64.6R, Mar. 2000. [JTCOOb] ISO/IEC JTC1 /SC29/WG11. Overview of the MPEG-4 standard. ISO /IE C N3747, Oct. 2000. [Kar84] N. Karmarkar. A New Polynomial-Time Algorithm for Linear Program ming. Combinatorica, 4:373-395, 1984. [Kar98] I. Karzela. Modeling and simulating communication networks: a hands- on approach using OP NET. Prentice Hall, New Jersey, Auguest 1998. [KBPY98] J. Knutsson, P. Butovitsch, M. Persson, and R. D. Yates. Downlink admission control strategies for CD M A systems in a M anhattan envi ronment. In Proc. VTC, volume 2, pages 1453-1457, May 1998. [KCBN99] T. Kwon, Y. Choi, C. Bisdikian, and M. Naghshineh. Call admission control for adaptive multimedia in wireless/mobile networks. In IEEE Wireless Communications and Networking Conference, pages 540-544, 1999. 178 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [Lab79] [LC01] [LDK95] [Lee95] [LeeO O ] [Lin97] [LZ94] [M ilOOa] [MilOOb] [NS96] [PD99] [PGL+97] [Put94] [Ren88]' Bell Lab. American telephone and telegraph company. The Bell System Technical Journal, 58(1), January 1979. Yi-Bing Lin and Imrich Chlamtac. Wireless and Mobile Network Archi tectures. John Wiley & Sons, Inc, 2001. Michael L. Littman, Thomas L. Dean, and Leslie Pack Kaelbling. On the complexity of solving Markov decision problems. In Proceedings of the Eleventh Annual Conference on Uncertainty in Artificial Intelligence (UAI-95), pages 394-402, Montreal, Quebec, Canada, 1995. William C. Y. Lee. Mobile Cellular Telecommunications- analog and digital systems. McGraw-Hill, 2 edition, 1995. William C. Y. Lee. Mobile Cellular Telecommunications - Analog and Digital Systems. McGraw-Hill, 2000. Y. B. Lin. Performance modeling for mobile telephone networks. IEEE Network, 45:63-68, Nov.-Dee. 1997. Zhao Liu and Magda El Zarki. Sir-based call admission control for ds- cdma cellular systems. IEEE Journal on Selected Areas in Communica tions, 12(4):638-644, May 1994. Ronald E. Miller. Optimization. John Wiley & Sons, 2000. Ronald E. Miller. Optimization: Foundations and applications. Wiley- Interscience, John Wiley & Sons, 2000. M. Naghshineh and M. Schwartz. Distributed call admission control in mobile/wireless networks. IEEE Journal Selected Areas Communica tions, 14:711-717, May 1996. Larry L. Peterson and Bruce S. Davie. Computer Networks: A Systems Approach. Morgan Kaufmann Publishers, 1999. R. Pandya, D. Grillo, E. Lycksell, P. Mieybegue, H. Okinaka, and M. Yabusaki. IMT-2000 Standards: network aspects. IEEE Personal Commun., 4(4):20-29, 1997. Martin L. Puterman. Markov Decision Processes: Discrete Stochastic Dynamic Programming. John Wiley & Sons, 1994. J. Renegar. A polinomial-time algorithm, based on Newton’ s method, for linear programming. Math. Programming, 40:59-93, 1988. 179 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [RicOO] [R.MK95] [RNT96] [Ros93] [Ros96] [RP96] [RSAK99] [RT89] [SCS99] [Sha48] [SP94] [SP97] [SS97] K.W. Richardson. Umts overview. Electronics & Communication Engi neering Journal, 12(3):93-100, June 2000. Ramin Rezaiifar, Armand M. Makowski, and Srikanta P. Kumar. Stochastic Control of Handoffs in Cellular Networks. IEEE J. Select. Areas Commun., 13:1348-1989, 1995. R. Ramjee, R. Nagarajan, and D. Towsley. On optimal call admission control in cellular networks. In IEEE INFOCOM’ 96, pages 43-50, 1996. Sheldon Ross. Introduction to Stochastic Dynamic Programming. ACA DEMIC PRESS, INC, and Harcout Brace Jovanovich, 1993. Arthur H. M. Ross. The CDMA Technology Site. http://www.am ug.org/ ahmrphd/, 1996. S. S. Rapport and C. Purzynski. Prioritized resource assignment for mobile cellular communication systems with mixed services and platform types. IEEE Transactions On Vehicular Technology, 45(3):443-458, Aug. 1996. P. Ramanathan, K. M. Sivalingam, P. Agrawal, and S. Kishore. Dy namic resource allocation schemes during handoff for mobile multimedia wireless networks. IEEE Journal on Selected Areas in Communications, 17(7): 1270-1283, July 1999. K. W. Ross and D. H. Tsang. Optimal circuit access policies in an ISDN environment: A Markov Decision approach. IEEE Trans. Commun., 37:934-939, 1989. Sungmoon M. Shin, Cheol-Hye Cho, and Dan Keun Sung. Interference- based channel assignment for ds-cdma cellular systems. IEEE Transac tions On Vehicle Technology, 48(l):233-239, January 1999. C. E. Shannon. A mathematical theory of communication. Bell System Technical Journal, 27:379-423, 623-656, July and October 1948. J. Shapira and R. Padovani. Microcell engineering in cdma cellular net works. IEEE Transctions on Vehicular Technology, 43(4):213-216, 1994. Arak Sutivong and Jon M. Peha. Novel heuristics for call admission control in cellular systems. 1997 IEEE 6th International Conference on Universal Personal Communications Record, 1:129-133, 1997. M. Sidi and D. Starobinski. New call blocking versus handoff blocking in cellular networks. Wireless Networks, 3:15-27, 1997. 180 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. [Tij86] [Vai90] [VG02] [VV93] [XCWOO] [ZKAQOl] [ZLOl] Henk C. Tijms. Stochastic Modelling and Analysis: A Computational Approach. John Wiley & Sons, 1986. P. M. Vaidya. An algorithm for linear programming which requires 0(((m + n)n2 + (m + n )1 '5n)L). Math. Programming, 47:175-201, 1990. Frank Vahid and Tony Givargis. Embedded System Design: A Unified Hardware/Software Introduction. John Wiley & Sons, 2002. A. M. Viterbi and A. J. Viterbi. Erlang capacity of a power con trolled cdma system. IEEE Journal on Selected Areas in Communication, 11 (6) :892— 900, 1993. Yang Xiao, C. L. Philip Chen, and Yang Wang. An optimal distributed call admission control for adaptive multimedia in wireless/mobile net works. In IEEE, pages 447-481, 2000. Jens Zander, Seong-Lyun Kim, Magnus Almgren, and Olav Queseth. Radio Resource Management for Wireless Networks. Artech House Pub lishers, 2001. Z.H. Zheng and W.H. Lam. Analytical methods to calculate performance of handoff prioritisation in dynamic channel assignment. Electronics Let ters, 37(15):978-979, July 2001. 181 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Design and analysis of MAC protocols for broadband wired/wireless networks

PDF

Adaptive video transmission over wireless fading channel

PDF

Design and analysis of server scheduling for video -on -demand systems

PDF

Error resilient techniques for robust video transmission

PDF

Efficient acoustic noise suppression for audio signals

PDF

Contributions to image and video coding for reliable and secure communications

PDF

Dynamic voltage and frequency scaling for energy-efficient system design

PDF

Contributions to coding techniques for wireless multimedia communication

PDF

A CMOS frequency channelized receiver for serial-links

PDF

Adaptive packet scheduling and resource allocation in wireless networks

PDF

Call admission control and resource allocation for quality of service support in wireless multimedia networks

PDF

Code assignment and call admission control for OVSF-CDMA systems

PDF

Direction -of -arrival and delay estimation for DS -CDMA systems with multiple receive antennas

PDF

Color processing and rate control for storage and transmission of digital image and video

PDF

Algorithms for streaming, caching and storage of digital media

PDF

A template-based standard-cell asynchronous design methodology

PDF

Design and applications of MPEG video markup language (MPML)

PDF

Digital processing for transponding FDMA/TDMA satellites

PDF

Algorithms and architectures for robust video transmission

PDF

Intelligent image content analysis: Tools, techniques and applications

Asset Metadata

Creator Chen, Huan (author)

Core Title Dynamic radio resource management for 2G and 3G wireless systems

School Graduate School

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag engineering, electronics and electrical,OAI-PMH Harvest,operations research

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Kuo, C.-C. Jay (committee chair), Lee, Daniel C. (committee member), Zimmermann, Roger (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-245584

Unique identifier UC11339373

Identifier 3093745.pdf (filename),usctheses-c16-245584 (legacy record id)

Legacy Identifier 3093745.pdf

Dmrecord 245584

Document Type Dissertation

Rights Chen, Huan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA