UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Channel resource managemant strategies for low earth orbit mobile satellite systems Wang, Zhipeng 2006

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
831-ubc_2006-201176.pdf [ 6.38MB ]
Metadata
JSON: 831-1.0064902.json
JSON-LD: 831-1.0064902-ld.json
RDF/XML (Pretty): 831-1.0064902-rdf.xml
RDF/JSON: 831-1.0064902-rdf.json
Turtle: 831-1.0064902-turtle.txt
N-Triples: 831-1.0064902-rdf-ntriples.txt
Original Record: 831-1.0064902-source.json
Full Text
831-1.0064902-fulltext.txt
Citation
831-1.0064902.ris

Full Text

Channel Resource Management Strategies for Low Earth Orbit Mobile Satellite Systems By Zhipeng Wang B. Sc., Sun Yat-sen (Zhongshan) University, 1995 M . Sc., Sun Yat-sen (Zhongshan) University, 1998 A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE D E G R E E OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES (Electrical and Computer Engineering) UNIVERSITY OF BRITISH C O L U M B I A September 2006 © Zhipeng Wang, 2006 Abstract Low Earth Orbit Mobile Satellite Systems (LEO-MSS) are promising solutions for global-coverage mobile communications systems, which are expected to provide users with advanced telecommunication services, including multi-party and multimedia services, anytime and anywhere. For these systems, one of the major difficulties in providing such advanced telecommunication services on a global scale is efficient channel resource management. The very high speed of LEO satellites and their relatively small spotbeams make. channel resource management for LEO-MSS a very challenging technical issue. With this problem as our main motivation, we propose several novel and efficient resource management strategies in this thesis and provide a combined analytical and computer simulation framework through which the performance and limitations of conventional and newly proposed resource management strategies can be better understood and evaluated. First, we develop a more accurate analytical method to evaluate Fixed Channel Reservation (FCR) with First Input First Output Queuing of Handover requests (FIFO-QH). In order to improve the performance of FCR, we propose an efficient Traffic-Dependent Dynamic Channel Reservation (TDDCR) scheme that exploits the high-speed deterministic movement property of LEO satellites and reserves channels according to the estimated number of handover requests and the positions of the Mobile Terminals (MTs). Second, motivated by the fact that the exact analysis of Dynamic Channel Allocation (DCA) with FIFO-QH is highly complex due to the dynamic nature of channel allocation to different cells, ii we develop an approximate but accurate analytical method to evaluate the performance of DC A in conjunction with FIFO-QH in LEO-MSS. Based on our previous study of D C A , we optimize channel reservations through both handover estimation and D C A and propose a novel Traffic-Dependent Dynamic Channel Allocation and Reservation (TDDCAR) technique that improves the efficiency of channel reservation and the handover performance of D C A . Third, we develop an analytical method for evaluating the performance of channel resource management strategies for LEO-MSS supporting multi-party traffic. To improve the overall performance, we propose and analyze the performance of an efficient Adaptive Channel Reservation (ACR) scheme that allows priority to be given to handover requests generated by multi-party traffic. When A C R is used in conjunction with a New Call Queuing (NCQ) policy, extremely low blocking and handover failure probabilities can be achieved for multi-party traffic. Finally, we generalize our research to include multi-class traffic in LEO-MSS. We develop an analytical methodology to evaluate the performance of two channel resource partitioning schemes, namely Complete Partitioning (CP) and Complete Sharing (CS), with and without FCR. Using multi-dimensional Markov chain techniques, we solve the analytical models and derive explicit expressions for Quality of Service (QoS) parameters such as call blocking and dropping probabilities. In addition, we introduce a more efficient Threshold Call Admission (TCA) scheme. This scheme ensures fair access to channel resources by setting proper call admission thresholds for incoming traffic and achieves better efficiency of channel utilization with its multiplexing gain. i i i Table of Contents Abstract u* Table of Contents iv List of Tables viii List of Figures ix Acronyms xii Symbols xiv Acknowledgements xix 1 Introduction 1 1.1 LEO/MEO Mobile Satellite Communication Systems 1 1.2 Technical Challenges for Resource Management 5 1.2.1 Handovers in LEO Mobile Satellite Systems.. 5 1.2.2 Resource Management Strategies 7 1.3 Research Motivations 11 1.4 Research Contributions of the Thesis 14 1.5 Thesis Type 15 1.6 Organization of the Thesis 15 1.7 References 20 2 Analysis and Performance Evaluation of Fixed and Traffic-Dependent Dynamic Channel Reservation Techniques 24 2.1 Introduction 24 iv 2.2 System Models and Parameters 26 2.3 Analysis of the FCR Scheme with FIFO-QH Policy 30 2.4 Traffic-Dependent Dynamic Channel Reservation Scheme 38 2.5 Conclusions 43 2.6 References 51 3 Performance Analysis of Dynamic Channel Allocation with FIFO Handover Queuing 52 3.1 Introduction... 52 3.2 System Model 53 3.3 Performance Analysis 53 3.4 Conclusions 58 3.5 References 61 4 A Novel Traffic-Dependent Dynamic Channel Allocation and Reservation Technique 62 4.1 Introduction 62 4.2 System Model 64 4.3 Traffic-Dependent Dynamic Channel Allocation and Reservation Technique 66 4.4 Conclusions. 71 4.5 References 76 5 Channel Resource Management Strategies for Multi-party Traffic: Performance Analysis and Improvements 78 5.1 Introduction 78 5.2 System Model and Parameters 80 5.3 Performance Analysis 82 5.4 Performance Improvement Techniques 88 5.4.1 Adaptive Channel Reservation (ACR) Scheme 89 5.4.2 New Call Queuing (NCQ) Scheme 90 5.5 Performance Evaluation Results and Discussion 92 5.6 Conclusions 94 5.7 References 1 Q 0 Analysis and Performance Evaluation of Channel Partitioning Policies for Multi-class Traffic 102 6.1 Introduction 102 6.2 System Model and Parameters 104 6.3 Performance Analysis of Complete Sharing 106 6.3.1 CS without FCR 108 6.3.2 CS with FCR 112 6.4 Performance Analysis of Complete Partitioning 114 6.4.1 CP without FCR 114 6.4.2 CP with FCR 116 6.5 Threshold Call Admission Policy 117 6.6 Performance Evaluation Results and Discussion 119 6.7 Conclusions 122 vi 6.8 References 130 Conclusions and Topics for Future Research 133 7.1 Conclusions 133 7.1.1 Analysis and Performance Evaluation of FCR and TDDCR Techniques 133 7.1.2 Performance Analysis of D C A with FIFO-QH 134 7.1.3 A Novel T D D C A R Technique 134 7.1.4 Analysis and Performance Evaluation of Channel Resource Management Strategies for Multi-party Traffic 135 7.1.5 Analysis and Performance Evaluation of Channel Resource Management Strategies for Multi-class Traffic 135 7.2 Topics for Future Research 136 7.2.1 Optimization of Channel Reservation Mechanism 136 7.2.2 Efficient Channel Resource Management Techniques for Multi-class Traffic 137 7.2.3 Optimization of Channel Resource Management Strategies ..137 7.2.4 System Utilization and Packet Level System Performance 137 7.2.5 Resource Management Strategies for Integrated Satellite-Terrestrial Mobile Telecommunication Networks 138 7.2.6 Resource Management over Fading and Shadowing Channels 138 vii List of Tables Table 1.1 General system characteristics of GEO and LEO satellite systems 17 Table 1.2 Space segment characteristics of Globalstar and Iridium 17 Table 1.3 Mobile user link characteristics of Globalstar and Iridium 18 Table 1.4 System services and costs of Globalstar and Iridium 18 Table 2.1 Simulation performance results for various non-uniform traffic conditions and comparison with uniform traffic performance 45 viii List of Figures Figure 1.1 Channel reuse and allocation 19 Figure 1.2 Channel reservations for handover calls 19 Figure 2.1 Rectangular cell model for LEO-MSS 46 Figure 2.2 Proposed non-uniform traffic model for LEO-MSS 46 Figure 2.3 State-transition diagram for the FCR scheme with FIFO-QH 47 Figure 2.4 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for FCR scheme with FIFO-QH 47 Figure 2.5 Traffic-dependent dynamic channel reservation procedures 48 Figure 2.6 Pbhck vs. traffic intensities for the different resource management schemes 48 Figure 2.7 Pfau vs. traffic intensities for the different resource management schemes 49 Figure 2.8 Pdrop vs. traffic intensities for the different resource management schemes 49 Figure 2.9 Pns vs. traffic intensities for the different resource management schemes 50 Figure 3.1 Rectangular cell system model for the considered LEO-MSS network 59 Figure 3.2 State-transition diagram for D C A with FIFO-QH and co-channel interference 59 Figure 3.3 Analytical and computer simulation evaluation results of the various performance probabilities for D C A with FIFO-QH technique in conjunction with LEO-MSS network 60 Figure 4.1 Rectangular cell model for LEO-MSS 73 ix Figure 4.2 Traffic-dependent dynamic channel reservation procedures 73 Figure 4.3 Puock vs. traffic intensities for different resource management techniques 74 Figure 4.4 Pfau vs. traffic intensities for different resource management techniques 74 Figure 4.5 Pdwp vs. traffic intensities for different resource management techniques 75 Figure 4.6 Pns vs. traffic intensities for different resource management techniques 75 Figure 5.1 Rectangular cell system model for the LEO-MSS under consideration 95 Figure 5.2 State-transition diagram for FCR in LEO-MSS 95 Figure 5.3 Adaptive channel reservation procedures 96 Figure 5.4 Performance evaluation results for the various probabilities as a function of traffic intensity for 7-party traffic with FCR (Ch = 2) 96 Figure 5.5 Performance evaluation results for the various probabilities as a function of traffic intensity for single-party traffic with FCR (C/, = 2) 97 Figure 5.6 Pb\ and Pus\ as a function of traffic intensity for FCR and A C R 97 Figure 5.7 Pf\ and Pd\ as a function of traffic intensity for FCR and A C R 98 Figure 5.8 Pbi and Pusi as a function of traffic intensity for FCR and A C R 98 Figure 5.9 Pbk and Pusk as a function of traffic intensity for N C Q and A C R 99 Figure 5.10 Pjk and Pdk as a function of traffic intensity for N C Q and A C R 99 Figure 6.1 Rectangular cell system model for the considered LEO-MSS network 123 Figure 6.2 Markov state diagram for two incoming traffic with b\=\ and Z?2=2 123 Figure 6.3 State-transition diagram for CP without FCR in LEO-MSS 124 Figure 6.4 State-transition diagram for CP with FCR in LEO-MSS 124 x Figure 6.5 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CS policy 125 Figure 6.6 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 2-channel traffic under CS policy 125 Figure 6.7 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CS with FCR 126 Figure 6.8 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 2-channel traffic under CS policy with FCR 126 Figure 6.9 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CP policy 127 Figure 6.10 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CP policy with FCR 127 Figure 6.11 PJK as a function of traffic intensity for CS and T C A 128 Figure 6.12 Pusk as a function of traffic intensity for CS and T C A 128 Figure 6.13 Pjk as a function of traffic intensity for CS and T C A 129 Figure 6.14 Pusk as a function of traffic intensity for CS and T C A 129 xi Acronyms 3G - Third Generation ACR - Adaptive Channel Reservation CAC - Connection Admission Control CCI - Co-Channel Interference CP - Complete Partitioning CS - Complete Sharing DCA - Dynamic Channel Allocation DCR - Dynamic Channel Reservation FCA - Fixed Channel Allocation FCR - Fixed Channel Reservation FIFO - First Input First Output FPLMTS - Future Public Land Mobile Telecommunications Syst GEO - Geostationary Earth Orbit GSM - Global System for Mobile Communications IS-54 - Interim Standard 54 IS-95 - Interim Standard 95 LEO - Low Earth Orbit LHS - Left-Hand Side LUI - Last Useful Instant M E O - Medium Earth Orbit MSS - Mobile Satellite Systems M T - Mobile Terminal NCQ - New Call Queuing for multi-party traffic QH - Queuing of Handover requests QN-Queuing of New calls QoS - Quality of Service R & D - Research and Development RHS - Right-Hand Side T C A - Threshold Call Admission TDDCR - Traffic-Dependent Dynamic Channel Reservation T D D C A R - Traffic-Dependent Dynamic Channel Allocation and Reservation UMTS - Universal Mobile Telecommunications System Symbols a - Variable to adjust the costs of reserved channels in T D D C A R scheme 8- Variance of the Gaussian distribution of the distance between traffic centers (Chapter 2) or correction factor (Chapter 5). Smax - Maximum value of the variance of Gaussian distribution 8k - Correction factor for class A: traffic A - New call arrival rate of each cell (Chapters 2, 4) or total new call arrival rate (Chapter 5) Ah — Handover call arrival rate of each cell (Chapters 2, 4) or total handover request arrival rate (Chapter 5) heii - New call arrival rate of each cell Ajn - Arrival rate of calls that enter the cluster Aout - Arrival rate of calls that leave the cluster Ai - New call arrival rates for state i Ani - Handover call arrival rates for state i Ak - New call arrival rates of A>party or class k traffic Ahk - Handover requests arrival rates of &-party or class k traffic Ak(n) - Arrival rate of class k calls when the system is in state n /Jk(n) - Departure rate of class k calls when the system is in state n \/jUc - Mean call duration time l/juw - Mean maximum queuing time xiv l/ju- M e a n channel ho ld ing t ime i n a cel l for both new arrivals and handovers l//Uk - M e a n channel ho ld ing t ime i n a cel l fo r both new arrivals and handovers o f class k t raf f ic p— T ra f f i c intensi ty o f new cal l arr ivals Pmax - M a x i m u m value o f t raf f ic intensi ty Pk— T ra f f i c in tensi ty o f new &-party or class k ca l l arr ivals per cel l A(x) - Set o f id le channels i n cel l x b(i) - Probab i l i t y that arr iva l calls cannot f i nd any avai lable channel w h e n the cluster has / < C channels i n use bk - B a n d w i d t h o f class k t raf f ic B(x) - Set o f busy channels i n cel l x C- N u m b e r o f channels assigned to one cel l or one cluster o f cells Ch - N u m b e r o f reserved channels i n one cel l Ck - Number of channels allocated for class k traffic Chk - N u m b e r o f reserved channels for class k t ra f f ic Cx(i) - Cost o f channel i i n cel l x Cx(k, i) - Cost cont r ibu t ion fo r channel i e A(x) due to the in ter fer ing cel l k e I(x) D - Channel reuse distance Di - Distance between the t raf f ic centers i and (z+1). Dmax - M a x i m u m distance between the adjacent t w o t raf f ic centers Cr - Channel reservat ion number E[thi] - Expected value o f channel ho ld ing t ime i n source ce l l x v E[thi] - Expected value of channel holding time in transit cell Ek[th\k\ - Expected value of channel holding time in a source cell for &-party or class k calls Ek[thik\ - Expected value of channel holding time in a transit cell for &-party or class k calls jit) - Approximate probability that handover calls in queue will fail because of CCI in system state i F(x) - Set of channels assigned to x by F C A I(x) - The set of cells interfering with cell x k - Number of parties or classes of traffic K - Maximum number of parties or classes of traffic L - Cell length L0 - The average length of the overlap area between two adjacent cells rihk - Mean number of times that a newly arriving &-party or class k call is handed over during its lifetime nue - Mean number of handovers for all incoming calls rik - Number of ongoing new and handover calls of class A:, calls in a cell N- Number of cells in the MSS network model Nk - Nominal capacity assigned to class k traffic rtik - Maximum number of class k calls that can be admitted with Ck channels Pbiock - Blocking probability of new call attempts Pfaii - Handover failure probability Pdmp - Call dropping probability xvi Pns - Unsuccessful call probability Phi - Handover probability of M T in its source cell Phi - Handover probability of M T in its transit cell PHI - Handover probability of a new call i Pho - Handover probability Pj - Probability of state j of the M/M/C/S queuing system Pfaii]/ - Handover failure probability conditioned on the system queuing state (C+j) P(i\i+l) - Probability that a handover request in queuing position i+l moves to position i before its M T leaves the handover area Pbk - Blocking probability of the &-party or class k new call attempts Pfk - Handover failure probability of the &-party or class k calls Pdk - Call dropping probability of the &-party or class k calls; Pusk - Unsuccessful call probability of the &-party or class k traffic Ph ik - Handover probability of a M T in its source cell for &-party or class k calls Phik - Handover probability of a M T in its transit cell for A>party or class k calls Pbe - Mean value of Pbk, Phie -Mean value of Pnu Ph2e -Mean value of Phik P(n) - Probability that the system is in state n q{c) - Probability that c channels are in use R - Cluster size xvii R(x) - Set of reserved channels in cell x S - State space of the Markov process modeling multi-class traffic Sk - Subset of states in which the system admits an arriving class k call S(c) - Set of states when c channels are in use Sf,k - Subset of states in which the system blocks an handover class k call tu -Time that a call i has already spent tjt -Time for M T to reach the cell boundary tc - Deterministic inter-arrival time for the subsequent handovers td - Call duration time tdk - Call duration time for &-party or class k traffic tsrc - Dwell time of a M T in its source cell twmax - Maximum queuing time for handover requests Tc - User sojourn time in a cell Tcaii — Mean call duration time Td- Mean call duration time Tdk - Mean call duration time for the &-party or class k calls Tqmax - Maximum queuing time for new multi-party calls Tw - Average maximum queuing time Vtrk~ Satellite ground-track speed Wj - Position factor for TDDCR xi - MT's position in a cell xviii Acknowledgements First and foremost I would like to express my sincere and deep gratitude to my supervisor, Professor P. Takis Mathiopoulos, for his tremendous support, invaluable advice, and insightful comments, which benefit not only the work presented in this thesis, but also my career in the long term. His optimism and constant encouragement helped me get through some difficult times of my research. I am eternally grateful to my co-supervisor, Professor Robert Schober. I would like to thank him for welcoming me into his research group in January 2004 and for his continued assistance in all aspects that are essential to me. Moreover, I thank him for his friendship and many invitations to join the group gatherings with his family. I would like to thank the members of my thesis examination committee, Dr. Cyril Leung, Dr. Hussein Alnuweiri, Dr. Lutz Lampe, and Dr. Vincent Wong, as well as the external examiner, for their time and effort spent reviewing this thesis. Their comments have enhanced my research in innumerable ways. I would also like to thank Mr. Craig Wilson from ICICS at U B C for his professional proofreading of this thesis. Special thanks are owed to my parents, who have supported me throughout my years of education, both morally and financially. I am forever indebted to my wife, Sumin L i , for her love, support, understanding, and encouragement, which helped me through this difficult yet rewarding process. Last but not least, I offer my enduring gratitude to the faculty, staff and my fellow students in the Department of Electrical and Computer Engineering, who have inspired me to continue my work in this field. x x 1 Introduction 1.1 LEO/MEO Mobile Satellite Communication Systems In recent years, a large body of R&D work has dealt with investigations of global-coverage Mobile Satellite Systems (MSS), which aim to provide users with advanced telecommunication services, including multimedia and multi-party services, anytime and anywhere. It is well known that the existing terrestrial digital radio networks (e.g., GSM, IS-95, IS-54) provide mobile communications services within large but limited geographical regions. In order to supplement the coverage area of these terrestrial systems, a number of satellite mobile communication systems for global personal communications have been proposed or developed. Third generation (3G) mobile telecommunications systems (e.g., UMTS, FPLMTS) with a fully integrated satellite component will globally provide seamless personal communications [1]. Several satellite orbital constellations have been considered for MSS, including the Low Earth Orbit (LEO), the Medium Earth Orbit (MEO) and the Geostationary Earth Orbit (GEO) constellations: • LEO: A non-geostationary satellite system that operates in Low Earth Orbit. LEOs can be further divided into Little LEO and Big L E O satellite systems. A Little LEO is a small satellite system, providing mainly mobile data services. A Big LEO is a larger satellite system, providing mainly mobile telephony services. Many of the "global mobile phone" 1 services are provided by Big LEO satellite systems. They are located between 700 - 1500 km above the earth. • M E O : A non-geostationary satellite system that operates in Medium Earth Orbit, again providing mobile telephony services. These satellites have also been proposed to be used as part of a new global mobile telephone system. They are located more than 10,000 km above the earth. • GEO: Geostationary satellites ocpupy an orbital position approximately 36,000 km above the earth, and remain in a stationary position relative to the earth itself. The world's major existing telecommunications and broadcasting satellites fall into this category. Table 1.1 presents some important general system characteristics of GEO and LEO satellite systems [2]. The non-GEO constellations have attracted a considerable amount of attention within the MSS space industry. This is primarily due to the perceived demand for smaller, terrestrial-like hand-held terminals, capable of providing mobile services with global coverage [3]. As is well known, because of their low altitudes, LEO-MSS have relatively low transmit power requirements and short transmission delay, thus offering promising solutions for the global-coverage hand-held mobile communications systems. In this thesis, we consider LEO satellites, which can provide cellular telephony with relatively small size Mobile Terminals (MTs). The development of LEO/MEO satellite constellations for global wireless communication to small user terminals began with the launch of a number of Motorola's Iridium satellites in 1997 2 [4] and the implementation of a number of Qualcomm's Globalstar satellites, which were launched in 1999 [5]. Currently, Globalstar provides affordable, dependable high-quality satellite voice and data service across the United States and to over 120 countries worldwide [5]. It offers good prospects for successful business because of the low cost of its space segment and because the cost of the ground segment is borne by the franchised operators of the system. The price for the handsets and the call rates are also reasonable [6]. However, Globalstar does not offer truly global coverage as Iridium does. The satellite constellation does not have inter-satellite links and the satellites are essentially flying repeaters in a "bent-pipe" architecture. Iridium was the first system launched. It is also the most expensive and most technically complex system, since essentially the largest portion of the network resides in the satellites and their cross links. The Iridium system began to offer the world's first handheld global satellite telephone and paging service in November 1998. However, with the fast expansion of terrestrial cellular networks and the rise of roaming agreements between cellular providers, Iridium L L C went bankrupt in 1999 because of its high operating costs and insufficient demand for the service. Iridium satellite services were re-established in 2001 by the newly founded Iridium Satellite L L C . Recently, the company posted significant growth in both subscribers and revenue. This strong growth reflects Iridium's continued penetration into the focused, vertical segments of the maritime, aeronautical, enterprise and military market. Their combined users now number approximately 148,000. 3 According to [4], currently Iridium Satellite L L C is the only provider of truly global satellite voice and data solutions with complete coverage of the earth (including oceans, airways and Polar Regions). Iridium delivers essential communications services to and from remote areas where no other form of communication is available. The Iridium constellation of 66 cross-linked L E O satellites, operates as a fully meshed network and is the largest commercial satellite constellation in the world. The Iridium service is ideally suited for industries such as maritime, aviation, government/military, emergency/humanitarian services, mining, forestry, oil and gas, heavy equipment, transportation, and utilities. Iridium provides service to the U.S. Department of Defense. The company also designs, builds and sells its services, products and solutions through a worldwide network of more than 100 partners. Similar to other researchers [7]-[10] and in order to enable us to fairly compare our performance results, the Iridium MSS is adopted as our system model in this thesis. Tables 1.2-1.4 provide important technical information about the system characteristics of these two leading LEO-MSS, i.e., Globalstar and Iridium. In particular, Table 1.2 presents the space segment characteristics, Table 1.3 presents the characteristics of the communications links between the mobile users and the satellite systems, and Table 1.4 presents the services provided by these two LEO-MSS and compares the system and service costs. Tables 1.2 - 1.4 are a compilation of data from a number of diverse sources (e.g., [4], [5], [7]-[9], [11]-[13]). 4 1.2 Technical Challenges for Resource Management Compared to the GEO satellite systems, the advantages of the LEO/MEO satellite systems include relatively low transmit power and short transmission delays, thus permitting direct and reliable communications between low-power handheld MTs and the satellites. Moreover, LEO/MEO systems ensure earth coverage with relatively small cells, thus achieving overall higher traffic capacities [14]. However, due to the high frequency of inter-beam handovers in LEO-MSS, the selection of suitable strategies for managing handover requests becomes a significant technical issue that needs to be addressed in order to make full use of the above features and guarantee the required Quality of Service (QoS) [8]. This requirement is our motivation for introducing and evaluating the performance of more efficient channel resource management strategies, such as handover queuing policies, channel allocation techniques and channel reservation schemes for LEO-MSS, which reduce the probability of handover failure. 1.2.1 Handovers in LEO Mobile Satellite Systems Handover, from beam-to-beam or from satellite-to-satellite is necessary in each of the LEO/MEO global MSS. This procedure is similar to the handover situation arising in terrestrial cellular systems, only in reverse: the M T is essentially stationary and the cells are moving fast. Due to the high-speed movement of LEO satellites and their relatively small size spotbeams, inter-beam handover requests occur rather frequently during a call's lifetime in the LEO-MSS. For example, in the Iridium system, assuming moderate M T speeds (e.g., 70 km/h), and with the 5 LEO satellite velocity typically exceeding 26,600 km/h, handover intervals for a cell size with radius 212.5 km occur in less than 1 minute. Every time that a call changes the spotbeam, there is a risk that it may be dropped because of the unavailability of channels in the destination cell. When no priority is given to handover call attempts over new call attempts, no difference exists between these attempts: the probabilities of blocking and handover failure are the same. However, the occurrence of a call being forced to terminate is considerably less desirable from the user's viewpoint than is the occurrence of blocking. On the basis of ITU-T requirements for land mobile services, the values of the call dropping probability and the new call blocking probability should not exceed 5 x 10"4 and 10~2 [15], respectively. In LEO-MSS, this problem is far more critical than the equivalent one in terrestrial mobile telecommunications systems, and will make the QoS provided unacceptable. When multi-party1 and multi-class2 traffic is involved, the issue of channel resource management becomes even more critical. For example, in multi-party calls, each party wil l be requesting channels from different cells in LEO-MSS simultaneously and will hold all channels during the conference. Such calls fail i f one party is blocked or encounters a handover failure. Moreover, the duration time of a multi-party conference call is usually longer than that of a two-party call, which results in more handovers during the call's lifetime. Multi-class calls that require more bandwidth in LEO-MSS will also face more blocking and handover failures than those calls that require only one voice channel. 1 A Multi-party call consists of more than one party in the L E O mobile telecommunications network and requires simultaneous use of more than one LEO-MSS channel. 2 Multi-class traffic consists of traffic with different characteristics, such as bandwidth and mean call duration. 6 The probability of forced termination can be decreased by giving priority (for channels) to handover attempts over new call attempts. Therefore, techniques that prioritize the handover requests with respect to the new call attempts are essential in order to reduce, as much as possible, the call dropping probability and attain a satisfactory QoS [16]. In this thesis, different resource management strategies, which prioritize the handover attempts and can be efficiently applied to LEO-MSS, are thoroughly investigated. Some literature reviews on the existing channel resource management strategies will be presented next. 1.2.2 Resource Management Strategies Resource management strategies for LEO-MSS have been an active research topic for the past few years, and thus quite a few papers on resource management techniques have been published (e.g., [7]-[10], [17]-[25]). In the following subsections, the most important resource management strategies and related research results previously published in the open technical literature are briefly summarized. 1.2.2.1 Handover Request Queuing (QH) Policies A n inter-beam handover strategy based on the suitable queuing of handover requests is essential to guarantee an acceptable QoS to mobile users in LEO-MSS, where there are frequent inter-beam handovers during a call's life time. When a M T with a call in progress leaves cell x and enters an adjacent cell y, there is an overlap area where it can receive a signal with an 7 acceptable power level from both cells. The time the M T spends crossing the overlap area, twmax, can be used to queue the related handover request, i f no channel is available in cell y. Different queuing schemes can be applied depending on how the handover requests are ordered in the waiting queue of a cell. The most common queuing policy is the First Input First Output (FIFO) scheme, where handover requests are queued according to their arrival instants [16]. Another scheme, known as Last Useful Instant (LUI), has been proposed in [8]. The main idea behind LUI relies on the fact that a handover request is stored in a queue position before (after) all handover requests having a greater (lower) residual value of twmax. In such a way, the system tries to serve the most urgent handover request first. According to [8], the ideal LUI scheme represents the best scheduling strategy for handover requests. 1.2.2.2 Call Admission Control (CAC) Call Admission Control (CAC) is one of the fundamental tasks performed by satellite networks at the new call setup phase, and determines i f the connection request can be accepted into the system without violating QoS commitments of the ongoing connections [18]. In a LEO satellite network, due to the frequent handovers, the communication channel between the M T and the satellite changes very often. Thus, the C A C function should ensure that the spotbeams have sufficient resources to support future handover calls. A new call setup request is rejected i f the original cell has no available channel for the MT. Otherwise, the system performs the C A C algorithm to investigate what kind of QoS impacts the new call attempt wil l bring to the future 8 system i f it is admitted into the network. The C A C algorithm will then ensure that the handover call failure probability for the new and existing calls is not above the required QoS threshold. If no QoS violation occurs in the test, the new call will be admitted into the network. If a QoS violation occurs, the new call is rejected. Quite a few efficient C A C algorithms have been proposed for LEO-MSS (e.g., [18]-[22]). 1.2.2.3 Channel Allocation Schemes Channel allocation schemes play a very important role in resource management strategies for LEO-MSS. Channel allocation techniques have to fulfill the following constraint: two different cells may reuse the same channel provided that they are at a suitable distance, called reuse distance, D, that allows tolerable levels for Co-Channel Interference (CCI). Fig. 1.1 illustrates the concept of cellular frequency reuse. A cluster consists of 7 cells that collectively use the complete set of available channels and each cell contains 1/7 of the total number of available channels. The cells with the same letter use the same group of channels. Two categories of channel allocation schemes have appeared in the literature: Fixed Channel Allocation (FCA) and Dynamic Channel Allocation (DCA) (see for example [8, 9], [10]). In FCA, a set of channels is permanently assigned to each cell according to the reuse distance D. A call can only be served by an available channel of the cell in which its M T resides. If an arriving call does not find a free channel in the cell, the call is blocked and therefore lost. As shown in Fig. 1.1, all of the system channels are divided into 7 groups and each cell in the cluster 9 is allocated one group of the channels. A newly arrival call in cell A will be admitted only when there is an unused channel in group A . As the call cannot use any of the available channels belonging to group B to G, it will be blocked i f there is no available channel in group A . A D C A strategy allows that any system channel can be temporarily assigned to any cell, provided that the constraint on the reuse distance D is fulfilled. Different D C A techniques have been proposed on the basis of the strategy used to select the channel to be allocated to a cell when a new call occurs in it. Two D C A schemes for LEO-MSS have been proposed in [9]. Compared to the F C A , they can greatly improve system performance, although the implementation in LEO-MSS is much more complex. 1.2.2.4 Channel Reservation Schemes Channel reservation schemes are widely used techniques for giving priority to handover attempts or more important users/services. As shown in Fig. 1.2, for Fixed Channel Reservation (FCR), among all the C channels in each cell, a fixed set of Ch channels will be exclusively assigned for handover calls. The remaining (C - Ch) channels are shared by both new calls and handover calls. A new call is blocked i f the number of available channels in the cell is less than or equal to Ch when the new call arrives. A handover attempt is unsuccessful i f no channel is available in the destination cell [16]. For Dynamic Channel Reservation (DCR), the number of the reserved channels Ch can be adjusted frequently according to different traffic conditions to make the reservation more efficient (e.g., [23] - [25]). 10 1.2.2.5 Channel Resource Partitioning Policies for Multi-class Traffic In order to provide satisfying wide-area wireless multimedia services, channel resource partitioning policies that determine the number of channels to be assigned to different classes of traffic (e.g., voice, video) need to be analyzed. There are two most primary but important channel resource partitioning schemes: Complete Partitioning (CP) and Complete Sharing (CS) [26]. The CP channel partitioning scheme divides the available bandwidth into separate subsets according to user type and each class of traffic can only make use of the idle channels in the subset of channels allocated to it. For CS, the incoming call can be admitted into the system as long as there are enough channels to accommodate it. 1.3 Research Motivations As discussed in Section 1.2.1, it is critical to implement effective channel resource management strategies that can reduce handover failures, ensure the required QoS for each call and efficiently make use of the precious wireless channel resources in available LEO-MSS. Several resource management techniques for reducing handover failures have been reported in the literature. Although most of them have achieved this goal, they have limitations and require further improvements. Both D C A schemes and QH policies can improve system performance. However, the implementation of D C A techniques in MSS tends to be expensive [7]. The performance of QH schemes can be significantly influenced by several system parameters, including satellite speed and the size and propagation conditions of the overlap area. If the 11 satellite has a very high speed and the overlap area is relatively small, the performance improvement of the queuing schemes is not very significant. For example, in our studies, we found that the ideal LUI scheme achieved almost the same performance as the FIFO scheme [25]. Moreover, the LUI scheme introduces a number of control complexities due to the frequent reordering of the handover requests in queue. The C A C scheme is very useful for the purpose of reducing handover failure probability. However, for some multi-party applications, such as teleconferences, the C A C scheme is not effective. A l l the parties in the teleconference should be admitted into the network, and the QoS of their connection should be maintained during the teleconference. More importantly, 3G wireless systems are expected to provide wide-area wireless multimedia and multi-party conferencing services such as multimedia games, video-on-demand, distance learning, multi-party voice and video conferences, online multiplayer gaming, etc. It appears that most of the published research on resource management strategies for LEO-MSS is restricted to the "mobile-to-fixed" voice service (e.g., [7]-[10]). Very limited research has been done on multi-class and multi-party traffic [27]-[33]. In addition, most of the past research work has focused on proposing new techniques and evaluating system performance through computer simulations. Theoretical studies have not received sufficient research attention yet, which stimulates our interest in developing efficient and accurate analytical methodologies to verify the simulation model and results, and thereby improve our understanding of resource management techniques in LEO-MSS. Motivated by the above, the research in this thesis aims at proposing new efficient channel 12 resource management strategies that can effectively improve the performance of LEO-MSS. Furthermore, it aims at providing a combined analytical and computer simulation framework through which the performance and limitations of conventional and newly proposed resource management strategies can be better understood and evaluated. The following specific research topics are examined: • Performance evaluation of newly proposed and existing techniques that prioritize handover requests with respect to new call attempts. These techniques include handover request queuing techniques, channel allocation schemes, channel reservation schemes and channel partitioning policies. Some combinations of these resource management schemes are also applied to LEO-MSS to further improve system performance, and their effect on overall system performance is evaluated. • Investigation of channel resource management techniques to support multi-party and multi-class traffic in LEO-MSS, including channel reservation techniques and channel partitioning policies. • Development of analytical methods to verify the system model and simulation results. The theoretical analysis covers handover queuing, channel reservation, channel allocation, channel partitioning, multi-class and multi-party traffic. To summarize, in this thesis, we focus on channel resource management strategies for LEO-MSS, propose novel and more efficient strategies, and evaluate their performance through analytical and computer simulation methods. Some typical traffic scenarios, such as multi-party and multi-class, are also considered. Overall, this thesis aims at providing a thorough 13 understanding of known and newly proposed channel resource management strategies under the LEO-MSS platform. 1.4 Research Contributions of the Thesis The previous discussions on the characteristics of LEO-MSS, the existing different channel resource management strategies, and the technical challenges of reducing handover failures in LEO-MSS, especially for multimedia and multi-party traffic, established the context for the following research contributions of this thesis: • We develop a more accurate analytical method to evaluate the performance of FCR with FIFO-QH in LEO-MSS. • We propose a novel and more efficient Traffic-Dependent Dynamic Channel Reservation (TDDCR) technique to reduce handover failures and improve system performance. Its performance is evaluated and compared with other resource management techniques. • We develop a new theoretical method to study the D C A with FIFO-QH in LEO-MSS. • We investigate the D C A schemes for LEO-MSS and propose a Traffic-Dependent Channel Allocation and Reservation (TDDCAR) scheme to further reduce handover failures in LEO-MSS. • We analyze and simulate multi-party traffic in LEO-MSS with FCR. We also propose a novel and more efficient Adaptive Channel Reservation (ACR) technique that outperforms the traditional FCR in supporting multi-party traffic for LEO-MSS. 14 • We study CS and CP channel partitioning policies with or without FCR for multi-class traffic in LEO-MSS through both computer simulation and mathematical analysis. We also introduce a Threshold Call Admission (TCA) policy, which ensures the fair and efficient use of channel resources and improves the performance of CS and CP. 1.5 Thesis Type Manuscript-based format is adopted as the thesis style. Conforming to the guidelines3 on the website of the Faculty of Graduate Studies at the University of British Columbia, each manuscript chapter is a version of an individual manuscript published or submitted. Therefore, except for Chapters 1 and 7 (i.e., the Introduction and Conclusions), Chapters 2 to 6 are self-contained manuscripts of our research, in which some similar descriptions exist, especially in sections relating to introductions, system models and parameters. 1.6 Organization of the Thesis The organization of this thesis is as follows. In Chapter 2, a mathematical analysis of the FCR scheme with FIFO-QH policy in LEO-MSS is presented, and a TDDCR scheme is proposed. In Chapter 3, an analytical approach to evaluate the performance of D C A with FIFO-QH in LEO-MSS is developed, and in Chapter 4, a novel T D D C A R technique is elaborated. An analysis and performance evaluation of FCR for multi-party traffic in LEO-MSS 3 More information on manuscript-based theses at U B C can be found at "http://www.grad.ubc.ca/students/thesis/index.asp?menu=002,000,000,000" 15 is provided in Chapter 5. A more efficient A C R scheme for multi-party traffic is also proposed and evaluated in the same chapter. The performance of CS and CP for multi-class traffic in LEO-MSS is analyzed in Chapter 6, followed by a performance evaluation and comparison of a novel and efficient T C A scheme. Finally, conclusions are drawn and topics for future research are summarized in Chapter 7. 16 Table 1.1 General system characteristics of GEO and LEO satellite systems LEO GEO Time delay Small Large Power Requirements Low High Launcher Low Cost High Cost Coverage Global Low elevation angle for latitude greater than 50° Satellite Cost Low High Satellite Number Many (for global coverage) Three (for continuous coverage up to 75° latitude) Satellite Configuration Complex Simple Satellite Maintenance and Control Complex Simple Handover More Frequent Less Frequent Satellite Tracking Yes No Doppler Shift High Low Table 1.2 Space segment characteristics of Globalstar and Iridium SPACE SEGMENT CHARACTERISTICS GLOBALSTAR IRIDIUM Orbit class LEO LEO Altitude (km) 1414 780 Number of satellites 48 active, 8 in-orbit spares 66 active, 6 in-orbit spares Satellite output power (W) 1000 1200 Satellite mass (kg) 450 689 Number of planes 8 6 Inclination (°) 52 86.5 Period (minutes) 114 100.1 Satellite visibility time (minutes) 16.4 11.1 On-board processing (regeneration) No Yes Satellite antenna Fixed, moving cells Fixed, moving cells Coverage Within ±70° latitude Global 17 Table 1.3 Mobile user link characteristics of Globalstar and Iridium MOBILE USER LINK GLOBALSTAR IRIDIUM Multiple access method C D M A T D M A / FDMA/TDD Modulation QPSK QPSK Uplink frequency (GHz) 1.62135 - 1.6265 1.619-1.6215 Downlink frequency (GHz) 1.62135 - 1.6265 2.4835-2.4985 Maximum bandwidth (MHz) 5.15 11.35 Beams per satellite 16 48 Total number of beams 768 3168 Satellite antenna Fixed, moving cells Fixed, moving cells Reuse pattern (cells per cluster) 1 7 Reuse factor 768 180 Dual or higher satellite path diversity exploited Yes No Minimum mobile terminal elevation angle 10° 8.2° Minimum earth-space link one-way propagation delay (ms) 4.63 2.60 Maximum earth-space link one-way propagation delay (ms) 11.5 8.22 Minimum earth station elevation angle (°) 10 N A Table 1.4 System services and costs of Globalstar and Iridium SERVICES AND COSTS GLOBALSTAR IRIDIUM Service types Voice, data, fax, paging, messaging, position location Voice, data, fax, paging, messaging, position location Voice (kbps) 2.4 2.4 Data (kbps) 9.6 2.4 Voice circuits / satellite 2400 1100 Dual-mode mobile terminals Yes Yes Hand-held mobile terminals Yes Yes System cost (million US$) 2600 4700 Mobile terminal cost (US$) From $625 From $900 rates (USS/minute) 0.14-1.00 0.99-1.50 18 F i g u r e 1.1 C h a n n e l r e u s e a n d a l l o c a t i o n . / / / / New call / ' arrivals / V / \ / \ / \ / \ / \. / \ / \ Handover ' \ calls \ \ \ v \ \ \ \ \ \ \ \ \ \ \ \ \ \ C-Ch C-1 F i g u r e 1.2 C h a n n e l r e s e r v a t i o n s f o r h a n d o v e r c a l l s . 1.7 References [1] M . Werner, A . Jahn, E. Lutz and A . Bottcher, "Analysis of system parameters for LEO/ICO-satellite communication networks", IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 371-381, 1995. [2] E. Del Re, " A coordinated European effort for the definition of a satellite integrated environment for future mobile communications", IEEE Commun. Mag., pp. 98-104, 1996. [3] M . Mohorcic and R. E. Sheriff, "Non-geostationary satellite constellations for provision of mobile broadband services", COST 252 TD (97) 06 Thessaloniki, July 7-8, 1997. [4] http://www.iridium.com [5] http://www.globalstar.com [6] J. V . Evans, "Satellite systems for personal communications", Proc. of the IEEE, vol. 86, no. 7, pp. 1325-1341, 1998. [7] E. Del Re, R. Fantacci and G. Giambene, "Different queuing policies for handover requests in low earth orbit mobile satellite systems", IEEE Trans. Veh. Technol, vol. 48, no. 2, pp. 448-458, 1999. [8] E. Del Re, R. Fantacci and G. Giambene, "Handover queuing strategies with dynamic and fixed channel allocation techniques in low earth orbit mobile satellite systems", IEEE Trans. Commun., vol. 47, no. 1, pp. 89-102, 1999. 20 [9] E. Del Re, R. Fantacci and G. Giambene, "Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks", IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 397-405, 1995. [10] G. Maral, J. Restrepo, E. Del Re, R. Fantacci and G. Giambene, "Performance analysis for a guaranteed handover service in an LEO constellation with a 'satellite-fixed cell' system", IEEE Trans. Veh. Technol, vol. 47, no. 4, pp. 1200-1213, 1998. [11] http://www.ee.surrey.ac.Uk/Personal/L.Wood/ [12] M. Richharia, Mobile Satellite Communications: Principles and Trends, Addison Wesley, 2001. [13] T. Pratt, C. Bostian and J. Allnutt, Satellite Communications, Wiley, 2003. [14] A. Ganz, Y. Gong and B. Li, "Performance study of low earth orbit satellite systems", IEEE Trans. Commun., vol. 42, pp. 1866-1871, 1994. [15] ITU-E.771, "Network grade of service parameters and target values for circuit-switched land mobile services", Blue Book, 1995. [16] D. Hong and S. S. Rappaport, "Traffic model and performance analysis for cellular mobile radio telephone systems with prioritized and nonprioritized handoff procedures", IEEE Trans. Veh. Technol, vol. VT-35, no. 3, pp. 77-92, 1986. [17] N. Luoras and T. Le-Ngoc, "Dynamic capacity allocation for quality-of-service support in IP-based satellite networks", IEEE Wireless Communications, vol. 12, no. 5, 2005. [18] B. Lee and Y. Kim, "Call admission method for call dropping probability guarantee in LEO satellite networks", IEEE GLOBECOM'03, vol. 2, pp. 1168-1173, 2003. [19] K. Lee, "Variable-target admission control for nonstationary handover traffic in LEO satellite networks", IEEE Trans. Veh. Technol, vol. 54, no. 1, pp. 127-135, 2005. [20] K . Lee, D. Oh and H . Lee, "Selective load sharing for handover admission in overlapping coverage", IEEE PIMRC 2004, vol. 2, pp. 1491-1495, 2004. [21] H. Uzunalioglu, J. W. Evans and J. Gowens, " A connection admission control algorithm for low earth orbit satellite networks", IEEE International Conf. Commun., pp. 1074-1078, 1999. [22] S. Olariu, R. Shirhatti and A. Y . Zomaya, "OSCAR - an opportunistic call admission protocol for L E O satellite networks", IEEE ICPP 2004, vol. 1, pp. 548-555, 2004. [23] F. Huang, S. Wu, H . Xu , J. Liu and B. Xiao, "Probability Based Dynamic Channel Reservation Strategy for Reliable Handoff in Multimedia LEO Satellite Communications", IEEE International Symposium on MAPE, vol. 2, pp. 1567 - 1570, 2005. [24] R. Dhaou, A . Beylot and M . Becker, "ACTR: A n adaptive time-based channel reservation mechanism for LEO satellite fixed cell systems", IEEE VTC 2003 Fall, vol. 4, pp. 2688-2692, 2003. [25] Z. Wang and P. T. Mathiopoulos, "Analysis and performance evaluation of dynamic channel reservation techniques for LEO mobile satellite systems", IEEE VTC 2001 Spring, vol. 4, pp. 2985-2989, 2001. [26] K W. Ross, Multiservice loss models for broadband telecommunication networks, Springer-Verlag, 1995. 22 [27] M . Nofal, "Engineering aspects and performance evaluation of a multi-service low earth orbit mobile satellite communication system", IEEE VTC 2000, vol. 4, pp. 1879 - 1886, 2000. [28] A . L. Beylot and S. Boumerdassi, "Adaptive channel reservation schemes in multitraffic LEO satellite systems", IEEE GLOBECOM '01, vol. 4, pp. 2740 - 2743, 2001. [29] K. Ke and C. Tzeng, "Optimal resource allocation for low-earth orbit (LEO) satellite networks with multirate traffics", IEEE GLOBECOM '01, vol. 3, pp. 2093-2097, 2002. [30] S. Karapantazis and F.-N. Pavlidou, "Design issues and QoS handover management for broadband LEO satellite systems", IEE Proceedings Commun., vol. 152, Issue 6, pp. 1006 - 1014, 2005. [31] C. Tzeng, K . Ke and H . Wu, "Resource allocation and adaptive routing in multimedia low earth orbit satellite mobile networks", IEEEICME'04, vol. 3, pp. 1795-1798, 2004. [32] B. S. Yeo and L . F. Turner, " A multi-class LEO satellite network", IEEE VTC 2001 Fall, vol. 4, pp. 2202-2205,2001. [33] L . Wood, H . Cruickshank and Z. Sun, "Supporting group applications via satellite constellations with multicast", IEE Conf. Telecomm. 1998, pp. 190-194, 1998. 23 2 Analysis and Performance Evaluation of Fixed and Traffic-Dependent Dynamic Channel Reservation Techniques1 2.1 Introduction Low Earth Orbit Mobile Satellite Systems (LEO-MSS) are promising solutions for providing wireless communications with global coverage or complementing the terrestrial wireless network. Because of their low altitudes, they offer quite a few appealing characteristics, such as low propagation delay, low transmit power, small cells and high frequency reuse efficiency. However, for LEO-MSS with "satellite-fixed" cell (i.e., cells move on the earth according to the satellite's motion), the high-speed movement of LEO satellites and their relatively small size spotbeams cause inter-beam handover requests to occur rather frequently during a call's lifetime. The occurrence of a call being forced to terminate is considerably less desirable from the user's viewpoint than the occurrence of blocking. Therefore, techniques that prioritize the handover requests with respect to the new call attempts are essential in order to reduce, as much as possible, call dropping probability and attain a satisfactory Quality of Service (QoS) [1]. In the recent past, there have been some papers investigating Dynamic Channel Allocation (DCA) schemes and Queuing policies of Handover requests (QH) (e.g., [2]-[5]). Although both the D C A and QH schemes can improve system performance, they also have their own limitations. The implementation of D C A techniques in MSS tends to be expensive [2]. For the QH schemes, their performance can be greatly influenced by several system parameters, including satellite speed, the size and the propagation conditions of the overlap area. If the satellite has a very high 1 A version of this work has been published. Z. Wang and P. T. Mathiopoulos, "Analysis and performance evaluation of dynamic channel reservation techniques for LEO mobile satellite systems", IEEE VTC 2001 Spring, vol. 4, pp. 2985-2989, 2001. 24 speed and the overlap area is relatively small, the performance.improvement of the QH schemes becomes not so significant. Therefore, it is more important to investigate the channel reservation schemes and find more efficient dynamic channel reservation schemes. Motivated by the above, in this chapter, we first present the theoretical analysis for the Fixed Channel Reservation (FCR) scheme with First Input First Output (FIFO) QH policy in the LEO mobile satellite network and compare the analytical results with equivalent performance evaluation results obtained by means of computer simulation. Based upon these results for FCR, we propose a new and more efficient Traffic-Dependent Dynamic Channel Reservation (TDDCR) scheme for LEO-MSS and evaluate its performance by means of computer simulation. We also study the performance of TDDCR together with other resource management schemes, such as different QH schemes and Queuing of New call attempts (QN). Moreover, while the above-mentioned performance evaluation results are obtained using a uniform traffic model, a non-uniform traffic model is proposed to evaluate the performance of TDDCR for LEO-MSS. The simulation results show that the proposed TDDCR scheme can effectively limit the call dropping probability and handle various non-uniform traffic distributions over the coverage area of the LEO mobile satellite network. The rest of this chapter is structured as follows. In Section 2.2, the system models and parameters are introduced. Section 2.3 presents the theoretical analysis for the FCR scheme with FIFO-QH in the LEO mobile satellite network, and then the analytical results are compared with the simulated performance evaluation results. In Section 2.4, the proposed new TDDCR scheme is presented and analyzed, followed by a comparison of the simulated performance evaluation results among several popular resource management strategies. Finally, our conclusions are drawn in Section 2.5. 25 2.2 System Models and Parameters Although there are several LEO-MSS models available, we consider Iridium as the model for our work as other researchers have done [l]-[5] and in order to make fair comparisons. It should be noted, however, that the proposed methodology can be applied to any LEO-MSS with satellite-fixed cells. The Iridium system consists of 66 satellites that are equally distributed in six near-polar circular orbits at about 780 km of altitude. The satellite ground-track speed Vtrk is approximately 26,600 km/h [2]. Due to beam-forming techniques, each multi-spotbeam antenna from a satellite irradiates a regular honeycomb cellular network on the earth. Many regular-shaped footprints of the L E O satellites cover the earth's entire surface [1]. In our study, similar to [4], each cell is modeled as a rectangular bounded cell. The cell length L is about 425 km. There is an overlap area between two subsequent cells with the average length La approximately 50 km [1]. A model of this network is illustrated in Fig. 2.1. Because the satellite ground-track speed is much greater than the typical speed of the M T on earth, the relative satellite-MT motion can be approximated only by the satellite ground-track speed. That is, the M T will cross the cellular network with a constant speed of 26,600 km/h in a single direction. Therefore, the handover destination cell will always be the neighboring cell in the direction of the relative satellite-MT motion. Similar to [l]-[4], we have assumed an ideal propagation link and considered only the user mobility and network topology. We also assume that there is a positioning system integrated into the satellite system [2, 4] so that the user position can be determined with sufficient accuracy at the beginning of the call setup. Because the relative satellite-MT motion is dominated by the deterministic L E O satellite motion, the M T position can be tracked easily during the call lifetime. 26 We consider as our reference point the Fixed Channel Allocation (FCA) technique, which has been adopted for LEO-MSS such as Iridium and Globalstar. With FCA, a set of channels is permanently assigned to each cell. A M T in a cell can only be served by the channels belonging to that cell. For voice traffic, i f the call does not find any free channel in the cell, it will be queued for a certain length of time or blocked immediately, depending upon the different policies, i.e., with or without queuing. We also make the usual assumption that the new call arrival process is an independent Poisson process. The call duration time tj is exponentially distributed, with an average value of Tcau. As previously mentioned, we consider that there is an overlap area between the two adjacent cells, where the M T with a call in progress can receive signals with acceptable power level from both cells. As illustrated in Fig. 2.1, when a M T with a call in progress leaves Cell A and enters an adjacent/destination cell B, there is an overlap area where it can receive a signal with an acceptable power level from both Cells A and B. The time the M T spends crossing the overlap area, twmax, can be used to queue the related handover request, i f no channel is available in Cell B. The handover procedure starts as soon as an active M T enters the overlap area between Cells A and B. If there is a free channel in Cell B, the handover will be performed immediately. The call will be served in Cell B and the channel in Cell A will be released. If there is no free channel available in Cell B, the call is still served by Cell A and the handover request will be queued. Once a channel in Cell B becomes available, it will first serve the handover request in queue and then the new call. If the call is completed before getting a new channel in Cell B , the corresponding handover request will be cleared from the queue. If the maximum queuing time for handover request, twmax, expires and there is still no channel available in Cell B, the call will be dropped and the handover will fail. When the location of the M T that requires a new 27 connection is in the overlap area of Cells A and B, it will try to get a channel in Cell B first in order to reduce the handover requests. If no channel is available in both cells, the call will be blocked. When there is no channel available in Cell B but a channel in Cell A is available, it will be served by Cell A and it will initiate a handover request to Cell B in the meantime. As shown in Fig. 2.1, only ongoing calls in Cell A can possibly generate handover requests to Cell B. Therefore, the maximum queue length needed for handover requests in each cell is equal to the number of channels per cell. Different queuing schemes can be applied, depending on how the handover requests are ordered in the waiting queue of a cell. The most common queuing policy is the FIFO scheme, where handover requests are queued according to their arrival instants [6]. The other scheme, Last Useful Instant (LUI), has been proposed in [2]. LUI relies on the fact that a handover request is stored in a queue position before (after) all handover requests having a greater (lower) residual value of twmax. In such a way, the system tries to serve the most urgent handover request first. According to [2], the ideal LUI scheme represents the best scheduling strategy for handover requests. We consider both uniform and non-uniform traffic conditions in this work. For the uniform traffic model, the MTs are considered uniformly distributed over the simulation area. A new call can arrive at any point in the satellite cellular network with equal probability. For non-uniform traffic, the model for LEO-MSS proposed in [7] allows us to forecast a traffic distribution on the earth that has been divided in areas of 5° longitude x 5° latitude. The model in [7] is close to real traffic conditions, however, it is not very suitable for performance comparison between uniform and non-uniform traffic conditions nor for evaluating the performance of different resource management schemes under different traffic conditions. We propose here another model that makes the performance comparisons between uniform and non-uniform.conditions more clear 28 and consistent. The main idea for our new traffic model is that, corresponding to cells, we have the same number of traffic centers as cells in the simulated network. Corresponding to new call arrivals in each cell with the same traffic intensity, we consider in our simulations new call arrivals around each traffic center with uniformly distributed traffic intensity. The proposed non-uniform traffic model will function as follows. The number of traffic centers is the same as that of cells in our simulated mobile satellite network. As shown in Fig. 2.2, the distance Z), between the traffic centers / and (z'+l) is uniformly distributed between 0 and Dmax km. The MTs are distributed around each traffic center and the distribution of their distances from the traffic center is assumed to be Gaussian distribution with zero mean and variance 8. The value of 5 is assumed to be uniformly distributed between 0 and Smax, where dmax can be changed to reflect the degree of the non-uniformity of the traffic. The traffic intensity for each traffic center is also uniformly distributed between 0 and pmax. Dmax, pmax and 8max are system parameters that can be adjusted to study the performance of the resource management strategies under different traffic conditions. The traffic centers move across the simulated cellular network with satellite ground-track speed. The following are the most important system model parameters that have been adopted in our analysis and computer simulations: (1) the Iridium mobility model; (2) the new call arrival rate of each cell is A; (3) average call duration Tcau - 3 minutes; (4) the channel allocation scheme is F C A and each cell has C - 10 channels [1]; (5) the queuing policies for handover requests are FIFO and LUI; (6) a finite queue of C - 10 for handover requests in each cell; (7) the queuing policy for new call attempts is FIFO and the maximum queuing time is 10 seconds; (8) uniform and non-uniform traffic conditions. We also use the following well-known QoS parameters [l]-[4] to evaluate the performance of various resource management strategies: (1) Pbhck- the 29 blocking probability of new call attempts; (2) Pfaif. the handover failure probability; (3) Pdrop- the call dropping probability; (4) Pns: the unsuccessful call probability. A n unsuccessful call happens when the call is initially blocked, or is dropped due to the failure of subsequent handover requests. The simulated MSS network consists of rectangular-shaped cells that form a strip of coverage on the earth. We use both a 98-cell and 7-cell network model. The ninety-eight 425 km cells can each cover an approximate geographic area in strip form to offer continuous circular coverage of the earth's surface. In this model, the calls going out from the 98th cell will request a handover from the 1st cell. Nevertheless, extensive computer simulation runs have shown that, for uniform traffic, the 7-cell model can provide sufficient simulation accuracy and with significantly less computational complexity. Thus, we have obtained the simulation results for uniform traffic by the 7-cell model. On the other hand, in order to simulate the randomness of a non-uniform traffic distribution, we used the 98-cell model to evaluate the performance of the proposed TDDCR scheme. 2.3 Analysis of the FCR Scheme with FIFO-QH Policy In this section, we analyze and evaluate the performance of FCR with FIFO-QH in the LEO mobile satellite network, based on the methodology proposed in [6]. This approach allows us to verify our simulation results for FCR with FIFO-QH and also to partially verify our system model and computer simulation methods. Similar work has been presented in [l]-[3] for the analysis of queuing policies with the LEO mobility model, but in our work we focus on channel reservation. Furthermore, in [l]-[3], only the Pns curve is shown, which is not error sensitive because of its relatively much larger value than Pfau or Pdrop- Our analysis provides a more 30 accurate and complete approach, since our performance evaluation includes all four QoS parameters. The obtained results show very good agreement between analysis and simulation. The analysis is based on the following system assumptions and approximations: (1) the new call origination rate is uniformly distributed over the mobile service area; (2) a very large population of mobile users is assumed, thus the average call origination rate is independent of the number of calls in progress; (3) the new call arrivals and handover requests are generated as two independent Poisson processes, with mean rates X and Xh per cell, respectively. The traffic intensity of new call arrivals per cell is given by where Tcau is the mean call duration time. According to [4], the user mobility in LEO-MSS is characterized by a dimensionless parameter y defined as the ratio between Tcau and the user sojourn time in a cell Tc, i.e., y = Tcai/Tc, where Tc = L/Vtrk- Let tsrc be the dwell time of a M T in its source cell, i.e., the time interval from the instant when the M T originates a new call to the instant when the M T enters the handover area and requests a handover to the destination cell for the first time. Based on the Iridium mobility model described in Section 2.2, tsrc is a random variable uniformly distributed between 0 and L/Vtrk- The random call duration time td is exponentially distributed with mean Tcaii. A M T with a call in progress in its source cell will request its first handover whenever td > tsrc. Therefore, the handover probability of M T in its source cell Ph\ can mathematically be expressed as After a successful handover, the M T will travel a deterministic distance L with velocity Vtrk in a transit cell and then send out a new handover request. Thus, tc = L/Vtrk is the deterministic P = M^iErlangs), (2.1) (2.2) 31 inter-arrival time for subsequent handovers. Clearly, the residual call duration time of a call after a successful handover request has the same Probability Density Function (pdf) as td. Thus, the handover probability of M T in a transit cell is given by Pki=Pr{td>te} = e^'r\ (2.3) A given call will have i handovers i f the newly arriving call has a call duration long enough to generate / handovers and the system can provide the channels to the new call and its subsequent i-l handovers. After the successful i-l handovers, the admitted call will continue to generate its ith handover request. Therefore, the mean value of the number of times that a newly arriving call is handed over during its lifetime nt, can be derived as = S K 1 - / ^ ) ^ ( 1 - ^ . (2.4) i = l 1 V 1 "fail)"h2 Considering a statistical equilibrium between MTs leaving and entering a cell, X can be related to A/, as follows: *0 - P»o* ) + K ( 1 " Pfan ) = K + W - Pblock )(1 -/>*,) + K (1 - Pfan )(1 -PHI)- (2-5) On the left-hand side (LHS) of the above equation, there are two sources of calls that enter the cell. The first corresponds to the newly admitted calls, and the other comes from the successful handover calls. On the right-hand side (RHS), there are three types of calls that leave the cell. The first part is for the calls that are handed over to the destination cell, while the other two are for the calls ending in the cell without requesting another handover. Thus, from Eqs. (2.4) and (2.5), we have (2.6) 32 Let L0 denote the average distance covered in the overlap area, then the average value of the maximum queuing time Twis equal to L0/Vtrk- The maximum queuing time can be approximated as a random variable exponentially distributed [2, 3], with expected value l/juw = Tw. A call in progress will hold the channel until the call is complete in the cell or is handed over to the destination cell. Therefore, the channel holding time for the source (transit) cell is equal to either the unencumbered (remaining) call duration time or the dwell time of a M T in the cell before it moves out, whichever is less. The channel holding time can be expressed as thi - min[td,tsrc] and th2 = min[td,tc]. The expected value of E[thi\ results in [2] E[thi] = Tcall(l-Phi),i = l,2. (2-7) In [2], the channel holding time in a cell for both new arrivals and handover calls has been approximated as an exponential distribution with mean, l/ju, 1 _ A(l - Pblock )E[thl ] + A , (1 - Pm )E[th2 ] ( 2 8 ) As described in Section 2.2, C channels are assigned per cell according to FCA, and the queue length for handover requests in each cell is equal to C. In FCR, priority is given to handover requests by assigning Q channels exclusively for handover requests among the C channels in a cell. The remaining C-Ch channels are shared by both new calls and handover calls. A new call is blocked i f the number of available channels in the cell is less than or equal to Ch when the call is originated. A handover request is unsuccessful or needs to be queued only i f no channel is available in the destination cell. In other words, a new call wil l be admitted only when there are more than Ch channels available in the cell while a handover request wil l be admitted whenever there is an idle channel in the cell. Each cell can be modeled as an M/M/C/S queuing system with non-homogeneous arrival rates [6]. The state of this queuing system is defined as the 33 sum of the number of calls in service and the handover requests in queue, and its state transition diagram is shown in Fig. 2.3. Let Pj denote the probability of state j, it can be derived from this figure that the "rate-up = rate-down" state equations are P, = JM A JM Cu + (j - C)jUM j-i> \ < j < C - C h C-Ch+\<j<C C +1 < j < 2C. Using the above equation recursively, along with the normalization condition 2C 7=0 the probability distribution, Pj, is found as follows: k=0 kl/jk • + z k=C-Ch+\ • + C-Ch jk-(C-Ch) •h -1-1 k-C k=C+\ O.MCYl(CM + ijUw) (2.9) (2.10) (2.11) P0, (/L + ^ ) ( c - ^ ) l f ( C - ^ ) j\ juj C\ncYl (Cju + ijuj i' = 0 \ < j < C - C h PQ, C - C h + \ < j<C P0, C + \ < j < 2 C . (2.12) The probability of blocking, Pbiock, is the sum of the state probabilities when the queuing system is in the state j > C-Cn, i.e., 2C Pbiock ~ ^Li^j j=C-C„ (2.13) 34 A given handover request in queue will fail i f its M T leaves the handover area before it comes to the first position in queue and a channel becomes available in the destination cell. For the FIFO queuing policy, i f the current state is greater than C, the handover request is queued and it fails only i f the waiting time exceeds a random time derived from an exponential distribution with mean \/juw. Let us express the handover failure probability as c Pfail=HPfail\jPC+j, (2.14) where P/aiiy is the handover failure probability conditioned on the system queuing state, i.e., the number of handover requests in queue. Similar to [6], we define the random variable t,• (1 <i <C) as the remaining dwell time for the handover request in the /th position of the queue. We define the random variable T as the elapsed time from the instant a handover request joins the queue to the instant that a channel is released in the destination cell, i.e., the minimum channel holding time in the fully occupied destination cell. When the system is in a state less than C, the handover request can be served immediately, and the handover failure probability will be zero. However, when a handover request comes in while the state is C, the handover request will enter in the first position of the queue and the state will then be C+ l . The handover failure probability in this situation can be calculated as: Pfaii\\ = Pr {MT does not get the channel before leaving the overlap area} = Pr{h<T) = — ^ — . (2.15) Therefore, the probability of the M T getting a channel when it is in the first position of the queue is 35 Pri = Pr (MT gets a channel in first position of the queue} = 1 - P/au\i = — — — . (2-16) CM + MW For a successful handover, the handover request entering the queue in position j > 2 (at this moment, the system state becomes C+j) should be able to move to the first position and get a channel before its M T moves out of the overlap area. Thus, we have 1=1 where P(i\i+1) represents the probability that a handover request in position i+l moves to position i of the queue before its M T leaves the handover area [6]. A handover request in position z'+l will move forward to the next position in queue i f the remaining dwell time of its M T exceeds either at least one of the remaining dwell times tk, k = 1, 2, ... i, for any request ahead of it in the queue, or the minimum remaining holding time T of those calls in progress in the destination cell. Otherwise, the request will be cleared from the system. Since the MTs move independently of each other and have independent channel-holding times, the random variables, T, tk, (k = 1, 2, ... i) are statistically independent. Thus P{i\i + \) = \-Pr{tM <T,tM <tk,k = 1,2,...,i} i = l,2 = \-Pr{tM<T).Pr{tM<t,}-Pr{tM<ti} = i - ( — & L _ ) ( I y . (2.i8) CM + MW 2 From the above analysis and Eqs. (2.16) - (2.18), for system states that are greater than C+l we can get the following expression for Pfauy'. 36 Finally, using Eqs. (2.14), (2.15) and (2.19), by summing over all fs, we can obtain the following handover failure probability: pm = I> " f i t 1 - , r ^ + 7 ^ p ™ • (2-2°) & /=i (Cju+Mw)2 CJU+JUW C{i+Mw From the user's point of view, the probability of a forced termination, Pdrop, c a n be more significant than P/au [6]. A call that is not blocked will eventually be forced into termination i f it succeeds in each of the first i-l handovers but fails on the ith. This call dropping probability can be calculated as Pdron ~ iPfail\Ph\Q- ~ Pfail) Ph2 ] = 1 _ p / i _ p \ - (2-21) i=l 1 "hl\\ -fail) The unsuccessful call probability, Pns, which results from either blocking or an unsuccessful handover, is also used as a major parameter for evaluating overall system performance. This probability is the sum of the new call blocking probability and the probability that a call is admitted but eventually dropped due to handover failure. Thus, Pns can be expressed as Pns=Pbl0ck+PdropV-Pblock)- (2-22) From Eqs. (2.13) and (2.20) - (2.22), it can be observed that in order to analyze system performance, i.e., to obtain Pbhck, Pfail, Pdrop and Pns, the probability of each system state is needed. Clearly, this state probability, Pj, is related to both the new call and the handover arrival rate, i.e., A and Ah, (see Eqs. (2.11) and (2.12)). In order to study system performance under different traffic intensities, we can specify or change the value of the system parameter p and calculate A using Eq. (2.1). However, it should be noted that from Eq. (2.5), Ah is related to Pbhck and Pfaii. Therefore, a recursive approach is needed to compute Pbhck and Pfau. The system parameters under consideration are: p, Vtrk, L, Tcaii, L0, C and Ch- We begin the iterations with 37 Pbiock= 0 and Pfau = 0 and compute the Ah and ju, using Eqs. (2.2) - (2.8). Then we substitute Ah and ju into Eqs. (2.11) - (2.12) and calculate the Pj for j = 0, 1, 2, ... 2C. From Eqs. (2.13) and (2.20), the new values of Pbiock and P/au can be obtained. These new values are then averaged with the Pbhck and Pfau values from the previous step, and a new iteration starts with these averaged values. The iteration ends when the relative difference between the Pbhck (Pfau) values computed in two consequent steps is less than 10"6. Fig. 2.4 shows the analytical and simulation results of the FCR scheme with FIFO-QH, where the reserved channel number Ch = 1. The results for the four QoS parameters, Pbhck, Pfau, Pdrop and Pns, are presented. From this figure, we can see that the performance simulation results for all four QoS parameters are in very good agreement with the analytical results. 2.4 Traffic-Dependent Dynamic Channel Reservation Scheme As the most primitive channel reservation approach, the FCR scheme has previously been applied to LEO mobile satellite communication systems [7]. In Section 2.3, we have developed the theoretical analysis for FCR with FIFO-QH in the LEO mobile satellite network. For FCR, the system reserves a fixed number of channels exclusively for the handover requests among all the channels in a cell, and the remaining channels are shared by both new calls and handover calls. However, such an approach is very inflexible, as it does not consider the changing traffic conditions. With the aim of improving the efficiency of channel reservations, in this section we wil l present a novel and efficient Traffic-Dependent Dynamic Channel Reservation (TDDCR) scheme for LEO-MSS, which exploits the deterministic movement property of LEO satellites. The novelty of the proposed TDDCR scheme is based on the following concepts. The TDDCR scheme makes the channel reservation based on the estimated number of handover requests, and 38 adjusts the number o f reserved channels f requent ly and according to the current t raf f ic condi t ions. Fur thermore, the T D D C R scheme considers the pos i t ion o f the M T to determine the urgency o f m a k i n g such a corresponding channel reservation. A s prev ious ly ment ioned, w e assume that the cal l durat ion is exponent ia l ly d ist r ibuted, w i t h an average value o f Tcaii- Therefore, the system makes use o f the p d f o f the exponent ia l d is t r ibut ion to estimate the handover probabi l i ty . The system keeps the o r ig in t ime o f each ongo ing cal l i n its coverage area so that i t can easi ly track the durat ion o f the ca l l . I n this est imat ion, the probab i l i t y that a cal l has long enough cal l durat ion to make a handover to the next cel l is calculated. I n order to present the operat ion o f the new reservat ion scheme, w e need to introduce a new system parameter, referred to as the Channel Reservat ion N u m b e r , Cr. Th is parameter w i l l be updated every t ime a related event happens, i.e., new ca l l arr ivals, handovers and cal l terminat ions. W h e n a new cal l arr ives, the system w i l l calculate and store its handover probabi l i ty . Th is p robab i l i t y w i l l be used to calculate the Cr and w i l l be updated w i t h the probab i l i t y o f pe r fo rm ing another handover to the next cel l after the corresponding M T per forms the handover. The p d f o f the random cal l durat ion t ime ta is f(t) = juce^, iort>0, (2 .23) where juc = \/Tcau. The cal l w i l l pe r fo rm a handover on ly w h e n i t can last suf f ic ient t ime to get to the boundary o f the next cel l . W e denote the t ime that a cal l i has already spent as tu and the t ime for i t to reach the boundary as tn- Therefore, the handover p robab i l i t y Pm can be calculated as fo l l ows : PHi=P{td>{t2i^-t,i)} = e^t^\ (2.24) I t should be noted that the system does not take into account the memoryless proper ty o f the exponent ia l d is t r ibut ion w h e n m a k i n g this est imat ion, and that other suitable pdfs can also be 3 9 applied to the system in order to get a more accurate estimation. The choice of suitable estimation pdf depends on a good distribution approximation of the actual traffic conditions. Moreover, in order to efficiently reserve the channels, each probability will be weighted by the position factor vv,-, which is the MT's position x, divided by the cell size L (see Fig. 2.5), i.e., This position factor is used to determine the urgency of making such a corresponding channel reservation. Contrary to Pm, w,- should be re-calculated frequently enough to reflect the exact position of the M T and its urgency to request a handover. In this TDDCR scheme, w, will be updated whenever an event happens. When the traffic becomes heavier, events will happen much more frequently; accordingly, wt will be updated more frequently. When the traffic intensity is low, the need to perform such an estimation and make the channel reservation is not so important in order to guarantee the channels for handover calls. Therefore, by re-calculating w, when an event happens, we obtain a dynamic position factor update frequency and thus increase overall system efficiency. If the M T with a call in progress is still far away from the boundary, the weighting factor will be small, and this call wil l not significantly influence the number of reservation channels in the next cell. The Cr of each cell is the sum of the weighted probabilities of all the active MTs in its coverage area and will determine the number of reservation channels in its neighboring cell. In other words, the destination cell will then make the channel reservation according to this Cr value, i.e., where C is the total number of channels in each cell. The Cr will be rounded to the closest integer to be the number for channel reservation in the destination cell. (2.25) c (2.26) 40 The above process is graphically illustrated in Fig. 2.5. As shown in this figure, with the generation of a new call, the traffic condition in Cell 1 changes so that the Cr of that cell needs to be re-calculated and Cell 2 will try its best to reserve the number of channels according to the new Cr value. When a handover is performed from Cell B to Cell C, the traffic conditions of Cells B and C have changed so that the of Cells B and C need re-calculation. In the meantime, because there is a channel released in Cell B, it is possible that it can be reserved for the additional handovers from calls in Cell A as long as Cell A has not got enough reserved channels. Therefore, Cells B, C and D will adjust the number of their reserved channels according to the C rs. Similarly, when a call terminates in Cell I, the Cr of Cell I will then be calculated and Cells I and II will adjust the number of their reserved channels. In the proposed TDDCR scheme, because the channel reservations are made according to the current traffic condition, we can expect some improvement in the efficiency of channel reservation. We simulated the previously described algorithmic system and evaluated the performance of our proposed TDDCR scheme in conjunction with FIFO/LUI-QH schemes and the FIFO-QN scheme. In our simulations, we evaluated the performance of both FIFO and LUI-QH schemes. From our system model, the average length of the overlap area L0 is 50km, so that the average maximum queuing time Tw is less than 7 seconds [1]. For the LUI-QH scheme, the positions of the handover requests in queue need to be rearranged only when a new call is generated in the overlap area and will move out of this area earlier than some other ongoing calls within the overlap area. The probability of this occurring is very small. Moreover, even when the queuing position is rearranged, and after the more urgent handover request is served, the probability of the subsequent handover requests getting the channels before the expiration of their waiting time (so as to achieve improved performance) is also very small, due to the short maximum queuing 41 time and small difference among their last useful instants. As a result, the FIFO-QH scheme achieves almost the same performance as the ideal LUI-QH scheme. In Figs. 2.6 - 2.9, we have summarized the most important performance evaluation results, i.e., Pbhck, Pfau, Pdrop and Pns, for the following resource management schemes: FIFO/LUI-QH, TDDCR with/without QH, FCR with reserved channel number 1 or 2 (FCR, Ch = 1 or Ch = 2), non-priority scheme (NPS), and FIFO-QH with Q N and with/without TDDCR. It should be noted that since the performances of the FIFO and LUI-QH schemes were almost the same, in Figs. 2.6 - 2.9 only the FIFO-QH curve is shown. The proposed TDDCR approach provides an effective way to reduce the cost of channel reservation. For the conventional FCR scheme, however, because the fixed reservation does not take into account traffic conditions, it tends to waste valuable channel resources and/or results in a higher call dropping probability. As can be observed from Fig. 2.6, the Pbhck of the proposed TDDCR scheme is slightly higher than that for the FCR with one reserved channel (FCR, Ch = 1) . However, in Figs. 2.7 and 2.8 it can be seen that the proposed TDDCR scheme has achieved a significant improvement in Pfan and Pdrop, compared to FCR Ch = 1 . Furthermore, in Fig. 2.9 it can be seen that the unsuccessful probabilities (Pns) for both schemes are almost the same. This means that for a certain number of unsuccessful calls, we will encounter more forced terminations when using FCR than when using TDDCR. This performance improvement makes the proposed TDDCR scheme an excellent handover strategy candidate for LEO-MSS. The performance of TDDCR in combination with other resource management schemes, such as the FIFO-QN and FIFO/LUI-QH, also provided some interesting results. Queuing schemes make use of the user waiting time and the overlap area between two cells so that the performance is generally much better than that of the non-queuing schemes. They provide the most efficient 42 way to make use of the channels and can accommodate more calls in the system. As shown in Figs. 2.6 and 2.9, the FIFO-QH scheme with FIFO-QN achieves the best performance in terms of PMock and Pns. As expected, when the proposed TDDCR scheme is used in conjunction with the QH schemes, the performance greatly improves. For example, compared with the FCR with both one and two reserved channels (FCR Ch = 1 and Ch = 2), much lower Pfau and Pdrop are achieved (see Figs. 2.7 and 2.8) while low Pbhck and Pns are maintained (see Figs. 2.6 and 2.9). For non-uniform traffic, in our simulations the distance between two subsequent traffic centers is uniformly distributed between 0 and Dmax = 2L = 850 km, and the traffic intensity for each traffic center is uniformly distributed between 0 and pmax =14 Erlangs. The variance 8 is uniformly distributed between 0 and 8max, and we adjust the 8max to change the degree of traffic non-uniformity. If the value of <!w increases, the traffic condition wil l become close to that of uniform traffic. In Table 2.1, the performance of the proposed TDDCR in different non-uniform traffic conditions is presented. As can be observed, the proposed TDDCR scheme achieves better performance under non-uniform traffic conditions than under uniform traffic conditions: the Pblock and Pns are lower, while Pfau and Pdrop are a little bit higher. Overall, the simulation results show that the proposed TDDCR scheme can effectively limit the call dropping probability and can effectively handle various non-uniform traffic distributions over the coverage area of the LEO mobile satellite network. 2.5 Conclusions In this chapter, we first presented the mathematical analysis for the FCR scheme with FIFO-QH policy, the results of which are consistent with those obtained by means of computer simulation. In order to further improve the performance of FCR and exploit the high-speed 43 deterministic movement property of LEO satellites, based upon the analytical and simulation results of FCR, we proposed an efficient TDDCR scheme that is suitable for LEO mobile satellite networks. The proposed scheme outperforms the traditional FCR scheme while introducing only a little additional computational complexity. When there is an overlap area between the neighboring two cells, the proposed scheme can be combined with the FIFO-QH scheme to achieve much better performance. Moreover, the proposed TDDCR scheme can handle various non-uniform traffic conditions and maintain its efficiency. From the analysis and simulation results, we have established that the proposed scheme is an attractive scheme for LEO-MSS. 4 4 Table 2.1 Simulation performance results for various non-uniform traffic conditions and comparison with uniform traffic performance Pblock Pfail Pdrop Pns Uniform 0.2246 0.0028 0.0082 0.2310 Smax ~ 10km 0.2153 0.0029 0.0088 0.2222 S,nax = 30km 0.2168 0.0028 0.0081 0.2231 Smax = 60km 0.2178 0.0030 0.0091 0.2249 Smax = 100km 0.2212 0.0028 0.0083 0.2277 Smax = 150km 0.2239 0.0028 0.0085 0.2305 45 Re-enter Handover New Call Cell Generation length L Overlap Area L„ Out New Call Generation C e l 1 Cell Boundary Vtrk = 26600km/h F i g u r e 2.1 R e c t a n g u l a r cel l m o d e l f o r L E O - M S S . Tra f f i c Center i Gaussian Distribution Mean 0, Variance S. Traffic Intensity pt Uniform Distribution between 0 and pmax Uniform Distribution between 0 and Dmax Tra f f i c Center i +1 Gaussian Distribution Mean 0, Variance Sj+l Traffic Intensity pi+i Uniform Distribution between 0 and pmax Cell Vtrk = 26600km/h F i g u r e 2.2 P roposed n o n - u n i f o r m t r a f f i c m o d e l f o r L E O - M S S . 2// ( C - C J / / (C-Ch+l)M ' CM CH + H„ C;I + 2MW CJJ + CMW Figure 2.3 State-transition diagram for the FCR scheme with FIFO-QH. 10 10 -1 r" Simulation Analysis —t- Call blocking —(•- Call blocking -e- Handove failure - 6 - Handover failure -*- Call dropping -*- Call dropping - e - Unsuccessful call - a - Unsuccessful call 4.5 5.5 6 6.5 7 7.5 Traffic intensity per cell, new arrivals (erl) 8 8.5 Figure 2.4 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for FCR scheme with FIFO-QH. 47 2 A B Handover C New Call Generation Termination V,rk = 26600 km/h Figure 2.5 Traffic-dependent dynamic channel reservation procedures. 5.5 6 6.5 7 7.5 Traffic Intensity per Cell. New Arrivals (erl) Figure 2.6 Pbiock vs. traffic intensities for the different resource management schemes. 48 0.08 10.01 0.001 0.0003 4.5 5.5 6 6.5 7 7.5 Traffic Intensity per Cell, New Arrivals (erl) Figure 2.7 Pf„u vs. traffic intensities for the different resource management schemes. 0.3 Q O 0.001 —(— NPS - * - FIFO-QH • • o - FIFO-QH. QN - 0 - TD-DCR. FIFO-QH. QN TD-DCR —<a- TD-DCR, FIFO-QH -« - FCR.Ch = 1 —a - FCR,Ch = 2 4.5 5 5.5 6 6.5 7 Traffic Intensity per Cell, New Arrivals (erl) 7.5 8.5 Figure 2.8 Pdrop vs. traffic intensities for the different resource management schemes. 49 0.5 0.003 4.5 5.5 6 6.5 7 7.5 Traffic Intensity per Cell. New Arrivals (erl) Figure 2.9 P„s vs. traffic intensities for the different resource management schemes. 2.6 References [1] E. Del Re, R. Fantacci and G. Giambene, "Different Queuing Policies for Handover Requests in Low Earth Orbit Mobile Satellite Systems", IEEE Trans. Veh. Technol, vol. 48, no. 2, pp. 448-458, 1999. [2] E. Del Re, R. Fantacci and G. Giambene, "Handover Queuing Strategies with Dynamic and Fixed Channel Allocation Techniques in Low Earth Orbit Mobile Satellite Systems", IEEE Trans. Commun., vol. 47, no. 1, pp. 89-102, 1999. [3] E. Del Re, R. Fantacci and G. Giambene, "Efficient Dynamic Channel Allocation Techniques with Handover Queuing for Mobile Satellite Networks", IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 397-405, 1995. [4] G. Maral, J. Restrepo, E. D. Re, R. Fantacci and G. Giambene, "Performance Analysis for a Guaranteed Handover Service in an LEO Constellation with a 'Satellite-Fixed Cell ' System", IEEE Trans. Veh. Technol, vol. 47, no. 4, pp. 1200-1213, 1998. [5] E. Del Re, R. Fantacci and G. Giambene, "An Efficient Technique for Dynamically Allocating Channels in Satellite Cellular Networks", Proc.of IEEE GLOBECOM '95, pp. 1624-1628,1995. [6] D. Hong and S. S. Rappaport, "Traffic Model and Performance Analysis for Cellular Mobile Radio Telephone Systems with Prioritized and Nonprioritized Handoff Procedures", IEEE Trans. Veh. Technol, vol. VT-35, no. 3, pp. 77-92, 1986. [7] R. E. Sheriff, H . F. Hu, E. Del Re, R. Fantacci, and G. Giambene, "Satellite -UMTS traffic dimensioning and resource management technique analysis", IEEE Trans. Veh. Technol, vol. 4, pp. 131-134, 1998. 51 3 Performance Analysis of Dynamic Channel Allocation with FIFO Handover Queuing1 3.1 Introduction In the recent past there have been a number of papers investigating techniques that give priority to handover requests in dealing with the very high frequency of handovers during a call's lifetime in Low Earth Orbit Mobile Satellite Systems (LEO-MSS). Among the proposed techniques, the Dynamic Channel Allocation (DCA) technique, combined with FIFO Queuing of Handover (FIFO-QH) requests, achieved the best performance in terms of call dropping probability {Pdrop), while maintaining a very low new call blocking probability {Pbhck)- The performance of this technique has been evaluated by means of computer simulations in [1] and [2]. While a few approximate analytical methods estimating D C A performance for terrestrial mobile telecommunication systems have been investigated in the literature (e.g., [3], [4]), there has been no analysis proposed for evaluating the performance of D C A with FIFO-QH in conjunction with LEO-MSS. Motivated by the above, in this chapter we develop an approximate analytical method that efficiently and accurately evaluates the performance of the D C A technique with FIFO-QH for LEO-MSS networks. 1 A version of this work has been published. Z. Wang and P. T. Mathiopoulos, "On the performance analysis of dynamic channel allocation with FIFO handover queuing in LEO-MSS", IEEE Trans. Commun., vol. 53, issue. 9, pp. 1443-1446, 2005. 52 3.2 System Model Similar to [1] and [2], Iridium is adopted as the LEO-MSS model for this analysis. The satellite ground-track speed Vtrk is about 26,600 km/h, which is much greater than the moderate speed of Mobile Terminals (MTs). Thus the relative satellite-MT motion can be approximated by Vtrk. The usual assumptions are made that the new call arrival process is an independent Poisson process with an arrival rate of Xceu for each cell, and that the call duration time ta is exponentially distributed with an average value of Tcan. Moreover, uniform traffic with ideal propagation links is assumed. As illustrated in Fig. 3.1, the simulated MSS network consists of 49 identical rectangular-shaped cells. Each cell's length is L, and the overlap area between two subsequent cells has an average length of L0. The network will repeat itself in all four directions so that the border cell is adjacent to the cell(s) at the other side of the network (see the "dummy adjacent cells" in Fig. 3.1). Thus, a continuous cellular network has been considered and each cell has a complete set of interference cells (two tiers). The calls going out from the right-hand side border cell will request a handover from the 1s t cell at the left-hand side. For this network, the following well-known Quality of Service (QoS) performance parameters [1,2] are investigated: (1) Pbhck-blocking probability of new call attempts; (2) Pfair. handover failure probability; (3) Pdrop'- call dropping probability; (4) Pns: unsuccessful call probability. 3.3 Performance Analysis The cluster of cells is considered as a whole and has a size of R = 7, i.e., 7 cells per cluster. Each cluster is assigned a total of C channels and a queue length of C. Let y = Tcau/Tc, where Tc 53 is the user sojourn time in a cell and Tc = L/Vtrk. The handover probabilities of the M T in its source/transit cell {Ph\IPhi) can be expressed as [1]: Phi=r(l-e-iUr)) andPh2=e-^\ (3.1) while the traffic intensity of new call arrivals per cell is: P = KM TCM (Erlangs). (3.2) The channel holding time in a cell for both new arrivals and handovers can be approximated as an exponential distribution with mean, [2], 1 _ KeU fl - Pblock W h l ] + Kcell (1 ~ Pfail )E{th2 ] 3 M Kelt fl _ Pblock ) + Kcell fl _ Pfail ) where A/, c e//is the arrival rate of handover requests for each cell, and E[thi] and E[thi\ are the expected values of channel holding time in source and transit cell, respectively: E[thi] = Tcall(\-Phi),i-\,2. (3.4) Denoting Xin I Aoul as the calls that enter/leave the cluster, then Kn = H e « ( l - ^ ) + 3 / l A c e / ; ( l - J P / a , ) , (3.5) K u , = 3Kcel, + 7Ke„ fl - Pblock ) f l " ^ 1 ) + ^Kell fl " Pblock ) ^ , 7 + *Kell (1 " Pblock )P„l fl " Pfai, )(1 " P,2 ) + Kell fl " Pblock ) f l " ^ , 7 ) ^ 2 ^ , 7 + " - PfitfPMl ~ P»2) + 3 ^ , ( 1 " ^ , , ) f l " P « ) + 3 ^ ( 1 " ^ , 7 ) ^ , 7 + 3AAce„ (1 - Pfail f Ph2{\- Ph2) + Ahcell (1 - Pfail f Ph22Pfail + Ahcell (1 - Pfail fPh2\l-Ph2 )• (3.6) In Eq. (3.5), Ain consists of the newly admitted calls and the successful handover calls coming from outside of the cluster. Furthermore, in Eq. (3.6), Xout includes the calls handed over to the destination cells out of the cluster, the calls ended in the cluster and the dropped calls due to handover failures between the cells within the same cluster. Considering a statistical equilibrium 54 between MTs leaving and entering a cluster, Xcen can be related to Ahceii using Eqs. (3.5) and (3.6), with the condition Xin = Xout. For D C A , the possible existence of Co-Channel Interference (CCI) needs to be considered. Let b{i) denote the probability that arrival calls cannot find any available channel when the cluster has / < C channels in use. In other words, due to the possible presence of CCI, even i f there are (C-i) non-occupied channels, new calls may still be blocked. While it appears extremely difficult to calculate the exact value of b(i), an approximation similar to those proposed in [3] and [4] wil l be also adopted here, i.e.: b(i) = [1 - (1 - w)g ] C ' * -L"*J ,0<i<C, (3.7) where (l-Pblock)Dl + (l-Pfail)D2 w = : - , (g + l)C Di=*™L(l + i.)t.D2 = ^ ( l + £-). (3.8) ju R p R In the above equations, DI and D2 represent the traffic for the new arrival calls and the handover arrival calls in the cluster, while w represents the traffic carried by an individual channel and g is the number of interference cells for a given cell. Clearly, for the case where there are two tiers of interference cell for a given cell, g = 3(i?-l). In Eq. (3.7), \_i/R] is the floor integer value ofi/R. The handover requests will not be blocked immediately, but wil l be queued for some time waiting for channels to become available. The maximum queuing time is assumed to be exponentially distributed, with mean \/piw = Tw, where Tw is the average value of the maximum queuing time and Tw = L0 /Vtrk- The effective handover requests arrival rate, which will advance the system state one step by occupying one more channel, can be approximated by considering 55 which is the probability that handover calls in the queue wil l fail because of CCI in system state i.flj) can be approximated as fij) = ( - ^ ) ^ ( 3 K e \ ), 0 < i < C, (3.9) lu + uw Acdl + Ahcell where P, is the probability of state i. In the above equation, uw/(iu + uw) is used to approximate the probability that a handover call will fail to obtain a channel before leaving the handover area when i channels are in use. Clearly, it is derived from the probability that a handover request in the first position of the queue cannot find a channel before its maximum queuing time expires. Moreover, the call will be queued with probability Xhceul{Xhceii + Xceu), because only handover requests will be queued among all the calls that cannot find an available channel. Thus, the new and handover call arrival rates for state i (0<i< Q, Xt and Xu, can be expressed as 4 =Mcell(l-b(i)) and Xhi =RXhcell(1-/(0), 0<i<C. (3.10) For /' > C, only handover requests will be admitted, so that the arrival rate is simply Xh = RXhceii-The cluster can be modeled as an M/M/C/S queuing system with non-homogeneous arrival rates [3]. The state of this queuing system is defined as the sum of the number of calls in service and the handover requests in queue. The state-transition diagram is illustrated in Fig. 3.2. Using the "rate-up = rate-down" state equations recursively, along with the normalization condition, Pj (0 <j < 2 Q is found as follows: k=° , j Po, l^J^C f-MJ jfl P0, C + \<j<2C, OficYl(Cfi + kfjw) 56 7-1 C-l C ^ + , c ! / i c n ( c / / + i / / w ) 1=1 Having obtained Pj, the four QoS performance parameters, Pbhck, Pfau, Pdrop and Pns can then be computed as follows: Pblock is the sum of the blocking probabilities in states j < C and the probabilities of the states j > C,i.e., Pbiock=fJPjbU)^fJPj. (3-12) j=0 7=C+1 Similar to [5], Pfau for this queuing system in states j> C can be analytically derived as p~'$w?"'-- (3'13) For D C A , Pfau should also take into account the failure probability when the channels in the system are not fully occupied, i.e., in states under C. Thus, for such a system Pfau = tPjfU)\4lrJ/!W. pc,r (3-14) Pdrop and Pns can be derived as follows: Pdrop = Z^[A.d-PfaiiY-'Pi?] = , _ v (3-15) 1=1 1 "h2\l "fail) Pns=Pblock+Pdr0p(\-Pblock)- . (3-16) From Eqs. (3.12) and (3.14), it can be observed that in order to obtain Pbhck and Pfau, the probability of each system state Pj is needed. Clearly, from Eq. (3.11), Pj is related to Xk, Xhk and Xh. As Xhceii is related to Pbhck and Pfau (see Eqs. (3.5) and (3.6)), a recursive approach is needed to compute Pblock and Pfau. The iteration starts with Pu0ck= 0 and Pfau = 0, computes Xhceii and /u 57 using Eqs. (3.3) - (3.6), and calculates Ai and AM using Eqs. (3.7) - (3.10). Then it substitutes At, AM, Ah and JU into Eq. (3.11) and calculates Pj for j = 0, 1, 2C. From Eqs. (3.12) and (3.14), the new values of Pbhck and PfaU can be obtained. A new iteration starts with these new Pbhck and Pfau values. The iteration ends when the relative difference between the PbhcklPfaii values computed in two consequent steps is less than 10"8. Fig. 3.3 shows the analytical and equivalent simulation results for D C A with FIFO-QH in LEO-MSS with C = 70 and Tcau - 3 minutes where various performance evaluation results for the four QoS parameters (Pbhck, Pfaiu Pdrop and Pns) are presented. In general terms, it can be clearly seen that the simulation performance results for all four QoS parameters are in very good agreement with the equivalent analytical performance results. Nevertheless, for Pbhck and Pns, some small differences can be observed for both low- and high-traffic intensities. Although it is very difficult to theoretically justify this difference, it is believed that this happens because of the approximation used for b(i) (see Eq. (3.7)), which cannot accurately approximate the blocking behaviour. However, on the other hand, for Pfan and PdroP, the proposed algorithm provides such an accurate estimation that there is almost no difference between analytical and simulation performance results even though the values of Pfan and Pdrop are very low and error-sensitive. 3.4 Conclusions In this chapter, we have proposed an approximate analytical method for evaluating the performance of D C A in conjunction with FIFO-QH employed in LEO-MSS. This method provides an excellent alternative approach for obtaining the performance of D C A with sufficient accuracy, thus avoiding the well-known problem of very time consuming and computational intensive computer simulations. 58 Out Vtrk = 2 6 6 0 0 km/h Figure 3.1 Rectangular cell system model for the considered LEO-MSS network. Figure 3.2 State-transition diagram for DCA with FIFO-QH and co-channel interference. 59 10 10' Simulation Analysis ^block —<--- s - Pfail — e - p fail P drop — * - P. drop — B — P ns — B - P ns 7.5 8 8 . 5 ' 9 Traffic intensity per cell, new arrivals (erl) 9.5 10 Figure 3.3 Analytical and computer simulation evaluation results of the various performance probabilities for DCA with FIFO-QH technique in conjunction with LEO-MSS network. 6 0 3.5 References [1] E. Del Re, R. Fantacci and G. Giambene, "Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks," IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 397-405, 1995. [2] E. Del Re, R. Fantacci and G. Giambene, "Handover queuing strategies with dynamic and fixed channel allocation techniques in low earth orbit mobile satellite systems," IEEE Trans. Commun., vol. 47, no. 1, pp. 89-102, 1999. [3] Z. H. Zheng and W. H . Lam, "Performance analysis of dynamic channel assignment with queuing and guard channel combined scheme for handoff prioritisation", IEE Electronics Letters, vol. 38, no. 25, pp. 1728-1729, 2002. [4] S. S. Kuek and W. C. Wong, "Approximate analysis of a dynamic-channel-assignment scheme with handoffs", IEE Proc.-Commun., vol. 141, no. 2, pp. 89-92, 1994. [5] Z. Wang and P. T. Mathiopoulos, "Analysis and performance evaluation of dynamic channel reservation techniques for LEO mobile satellite systems," IEEE VTC2001 Spring, 53rd VTS, vol. 4, pp. 2985 -2989, 2001. 61 4 A Novel Traffic-Dependent Dynamic Channel Allocation and Reservation Technique1 4.1 Introduction The third generation (3G) mobile communication systems are aiming to provide seamless personal wireless communication services with global coverage. It is well known that the terrestrial cellular networks cannot provide services to under-populated areas or undeveloped rural areas economically and conveniently [1]. To supplement the existing terrestrial systems, Mobile Satellite Systems (MSS) are playing an important role by offering promising solutions for the areas where the terrestrial cellular networks are not competitive, not applicable or not developed. Among the various MSS, Low Earth Orbit (LEO) constellations have attracted a considerable amount of attention due to their many advantages over Geostationary MSS, including relatively low transmit power and much shorter transmission delays. These features enable direct communication between the low-power handheld Mobile Terminals (MT) and the satellites. Moreover, they provide coverage with smaller cells, and can thus achieve higher overall traffic capacities [2]-[4]. However, because of their high satellite speed and relatively small spotbeams, the very high frequency of handovers during a call's lifetime will strongly influence the system's capacity and performance. Thus, in order to efficiently make use of the precious wireless channels, increase the network capacity, create a seamless mobile telecommunication network and guarantee the required Quality of Service (QoS), flexible and 1 A version of this work has been published. Z. Wang and P . T. Mathiopoulos, " A novel traffic dependent dynamic channel allocation and reservation technique for LEO mobile satellite systems", Proc. of IEEE VTC 2002 Fall, vol. 3, pp. 1652-1656,2002. 62 efficient resource management techniques for managing handover requests are of great importance. This chapter investigates the LEO-MSS with "satellite-fixed" cells, i.e., cells that move on the earth according to the satellite's motion [5]. In such LEO-MSS, inter-beam handover requests occur frequently during a call's lifetime. The occurrence of a call being forced to terminate is considerably less desirable than the occurrence of blocking. Therefore, techniques that prioritize the handover requests with respect to the new call attempts are essential in order to reduce the call dropping probability (Pdrop) and attain a satisfactory QoS [6]. In the recent past, there have been some papers investigating techniques that give priority to handover requests, such as Dynamic Channel Allocation (DCA) techniques (e.g., [7]), Dynamic Channel Reservation (DCR) techniques (e.g., [8]) and Queuing policies of Handover requests (QH) (e.g., [9]). A l l of these techniques can reduce the Pdrop and increase the system capacity, but also have their own limitations. It is well known that the satellite ground-track speed is much greater than the typical speed of the terrestrial MT, and that the relative satellite-MT motion can be approximated only by the satellite ground-track speed. This deterministic M T movement allows a more accurate estimation of the handover instant and handover probability (Ph0)- The D C A techniques have not fully exploited the deterministic movement of the MT, whereas the performance of QH techniques can be greatly influenced by satellite speed, as well as the size and signal propagation conditions of the overlap area. If the satellite has a very high speed and the overlap area is relatively small, the performance improvement of the QH techniques becomes insignificant. Previously, we proposed and evaluated a D C R technique [8], which makes use of the deterministic M T movement to estimate Pno. However, its flexibility in adjusting the reserved channels according to the changing traffic is limited because of the Fixed Channel Allocation 63 (FCA) technique it is based on. With the aim of further improving the performance of the DCR technique, in this chapter we propose a novel and more efficient Traffic-Dependent Dynamic Channel Allocation and Reservation (TDDCAR) technique for LEO-MSS and evaluate the performance improvements it offers, compared with the other known techniques. The remainder of this chapter is structured as follows. In Section 4.2, the system models and parameters are introduced. Section 4.3 presents the proposed T D D C A R technique, followed by a comparison of the performance evaluation results among several popular resource management techniques. Finally, we draw conclusions in Section 4.4. 4.2 System Model Although there are several LEO-MSS models available, Iridium is adopted as the LEO-MSS model in this chapter as other researchers have done [5], and in order to make fair comparisons. It should be noted, however, that the proposed methodology can be applied to any LEO-MSS with "satellite-fixed" cells. The Iridium system consists of 66 satellites that are equally distributed in six near-polar circular orbits at about 780 km of altitude. The satellite ground-track speed V^k is around 26,600 km/h and the relative satellite-MT motion can be approximated only by the satellite ground-track speed. Therefore, the handover destination cell will always be the neighbouring cell in the direction of the relative satellite-MT motion. The usual assumptions are made that the new call arrival process is an independent Poisson process and that the call duration time tj is exponentially distributed with an average value of Tcaii. Similar to [5]-[10], uniform traffic with ideal propagation links are considered. Moreover, a positioning system is assumed to be integrated into the satellite system [5, 10] so that the user's position can be 64 determined with sufficient accuracy at the beginning of the call setup and the M T position can be tracked easily during the call's lifetime. In our study, each cell is modeled as a rectangular bounded cell. The cell length L is about 425 km. There is an overlap area between two subsequent cells with an average length L0 of approximately 50 km. As illustrated in Fig. 4.1, the simulated MSS network consists of 49 rectangular-shaped cells. Each cell has two tiers of interference cells. The network wil l repeat itself in all four directions so that the border cell is adjacent to the cell(s) at the other side of the network (see the "dummy adjacent cells" in Fig. 4.1). Thus, a continuous cellular network has been simulated and each cell has a complete set of interference cells. The calls going out from the right-hand side border cell will request a handover from the 1 s t cell at the left-hand side. The following are the most important parameters adopted in our computer simulations: (1) the new call arrival rate of each cell is X; (2) average call duration is 3 minutes; (3) for D C A , the total number of channels in the system is 70, and for F C A , each cell has 10 channels [9]; (4) queuing policy for handover requests is FIFO. As shown in Fig. 4.1, only ongoing calls in Cell A can possibly generate handover requests to Cell B. Therefore, the maximum queue length needed for handover requests in each cell is equal to the number of channels a cell can be assigned to, i.e., 10 for F C A and 70 for D C A ; (5) uniform traffic conditions is assumed. Furthermore, we use the following well-known QoS parameters [5]-[10] to evaluate the performance of various techniques: (1) Pblock'- the new call blocking probability; (2) Pfail: the handover failure probability; (3) Pdrop- the call dropping probability; (4) Pns: the unsuccessful call probability. 65 4.3 Traffic-Dependent Dynamic Channel Allocation and Reservation Technique Exploiting the deterministic M T movements, the proposed T D D C A R technique optimizes channel reservation by implementing an efficient DCR technique [8] based on the estimation of possible handovers with changing traffic conditions. The system stores the origin time of each ongoing call in its coverage area so that it can easily track the duration of the call. When a new call i arrives, the system will calculate and store its handover probability, Pm- i ^ w i H be updated with the probability of performing another handover to the next cell after the corresponding M T performs the handover. The Probability Density Function (pdf) of the random call duration time U is /(0 = / / c e-^ , fo r />0 , (4.1) where /uc = "\.ITcau. The call will perform a handover only when it can last sufficient time to get to the boundary of the next cell. Denoting the time that a call i has already spent as tu and the time for it to reach the boundary as tn, PHI can be calculated as follows: P H I = e'M'2i+'u). (4.2) It should be noted that the system doesn't take into account the memoryless property of the exponential distribution when making this estimation, and that other suitable pdfs can also be applied to the system in order to get a more accurate estimation. The choice of suitable estimation pdf depends on a good distribution approximation of the actual traffic conditions. Moreover, in order to efficiently reserve the channels, each probability wil l be weighted by the position factor w„ which is the MT's position x{ divided by the cell size L (see Fig. 4.1), i.e., 66 (4.3) This position factor is used to determine the urgency of making such a corresponding channel reservation. Contrary to Pm, vv,- should be re-calculated frequently enough to reflect the exact position of the M T and its urgency in requesting a handover. In this technique, w, will be updated whenever an event happens, i.e., new call arrivals, handovers and call terminations. When the traffic becomes heavier, events will happen much more frequently and w, wil l be updated more frequently. When traffic intensity is low, the need to perform such estimation and make the channel reservation is not so important in order to guarantee the channels for handover calls. Therefore, by re-calculating w, when an event happens, we obtain a dynamic position factor update frequency and thus increase overall system efficiency. If the M T with a call in progress is still far away from the boundary, the weighting factor wil l be small and this call will not significantly influence the number of reservation channels in the next cell. The Channel Reservation Number Cr of each cell is the sum of the weighted probabilities of all the active MTs in its coverage area. The destination cell will then make the channel reservation according to this Cr value, i.e., where C is the total number of active calls in each cell. Cr will be rounded to the closest integer to be the number for channel reservation in the destination cell. Cr wil l be updated after every event. The above-described process is graphically illustrated in Fig. 4.2. With the generation of a new call, the traffic condition in Cell 1 has changed so that the Cr of cell 1 needs to be re-calculated; Cell 2 will try its best to reserve the number of channels according to the new Cr c C r = ^ P f (4.4) 67 value. When a handover is performed from Cell B to Cell C, the traffic conditions of Cells B and C have changed so that the Crs of Cells B and C need re-calculation. In the meantime, it is possible that the channel released in Cell B can be reserved for additional handovers from calls in Cell A as long as Cell A has not got enough reserved channels. Therefore, Cells B, C and D will adjust the number of their reserved channels according to the CVs. Similarly, when a call terminates in Cell I, the Cr of Cell I will be calculated and Cells I and II will adjust the number of their reserved channels. In the meantime, T D D C A R utilizes the D C A algorithm to select the channel with minimum costs to be reserved for the possible handover call or to be assigned to the incoming new/handover call. With DCA, not only the channel allocation efficiency will be improved but also the flexibility of dynamic channel reservation will be greatly extended. Thus, the channel reservation cost will be reduced and the overall system performance will be improved. Compared to the equivalent D C A algorithm proposed in [7], the D C A algorithm in T D D C A R considers not only the cost of the idle channels, but also the cost of the reserved channels to incorporate with the DCR technique and improve reservation efficiency. Considering cell x, let I(x) be the set of cells interfering with cell x, A(x) the set of idle channels, R(x) the set of reserved channels, B(x) the set of busy channels and F(x) the set of channels assigned to x by FCA. Following [7], the cost of channel i in cell x can be calculated as kef(x) where Cx(k, i) is the cost contribution for channel i e A(x) due to the interfering cell k e I(x), given by cx (0 = ^(0+ £ c * ( * » o , (4.5) Cx (k, i) = vk (i) + 2(1 - qk (i)), Vkel(x). (4.6) 68 In order for the reserved channels to be included in the calculation of cost contribution, vk(i) and qk(i) can be expressed as v,(0 = 1, i e A(k) a, i e R(k) 0, otherwise, qk(i) = l0> i e F ( k ) (4.7) * 1, otherwise. a e [0, 1] is a variable, for which we derived the best value by the following: a was varied in small increments and the overall system performance {Pblock and Pdrop) was evaluated by means of computer simulation. We found that for a = 0.375, Pbhck is minimized, whereas for a= 0.125, Pdrop is minimized. Therefore, we adopted a = 0.25 as the best value for the considered cost function. From Eq. (4.7), it should be noted that idle channels have a higher cost value, as they can be allocated to either new or handover calls. The T D D C A R scheme allocates the channel /* that has the minimum cost to the new arrival call by evaluating the cost functions of all the channels, / e A(x): C x ( i ' ) = m p C x ( / ) . (4.8) To select channel i' for the handover call, the cost functions of all the channels i e A(x)uR(x) need to be evaluated, i.e., c*(f,) = -A>c*(°- (4-9) When a call is terminated or being handed over to a destination cell, T D D C A R releases the channel i" that has the maximum cost by evaluating the cost functions of all the channels / e R(X)KJB(X): 69 CM") = max CM). (4.10) ieR(x)l)B(x) X Whenever there is a need to reserve a channel for possible handovers or to serve an incoming new/handover call in cell x, the system wil l first allocate an available channel belonging to F(x) and then select the channel that has the least impact on the available channels of I(x). When a call has terminated in cell x or been handed over to the next cell, the system wil l release the channel in x with maximum cost, i.e., the channel that can be assigned to the maximum number of cells in I(x). The released channel will first serve the handover requests in queue. When there is no queuing handover request, it will be reserved for possible handovers i f the previous reservation requests have not been fulfilled. Lastly, it will be available for new calls. The T D D C A R technique also employs FIFO queuing for handover requests. As illustrated in Fig. 4.1, when a M T with a call in progress leaves Cell 8 and enters an adjacent/destination Cell 9, there is an overlap area where it can receive a signal with an acceptable power level from both Cells 8 and 9. The time the M T spends crossing the overlap area can be used to queue the related handover request i f no channel is available in Cell 9. We used computer simulation to evaluate the performance of the proposed T D D C A R technique. Figs. 4.3-4.6 summarize the most important performance evaluation results, i.e., Pbhck, Pfan, Pdwp and Pns for Non-Priority Scheme (NPS), Fixed Channel Reservation (FCR) technique with 1 or 2 reserved channel(s) (FCR1, FCR2), DCR with/without QH (DCR, DCR+FIFO), D C A and the proposed T D D C A R technique. Among these resource management strategies, it is clear that T D D C A R provides an effective means of reducing the cost of channel reservation. Conventional FCR tends to waste valuable channel resources or results in a higher Pdrop because of the fixed reservation, which does not take into account traffic conditions. For DCR, although it estimates the number of handover calls and frequently adjusts the number of reserved channels 70 according to current traffic conditions, its flexibility in dynamically allocating channels for reservation is greatly limited by the F C A technique it works with. As can be observed from Fig. 4.3, the Pbhck for the proposed T D D C A R technique is lower than for other channel reservation techniques (FCR1, FCR2, DCR) for most traffic intensities investigated, although it is higher than for D C A because of the channel reservation. However, as illustrated in Figs. 4.4 and 4.5, T D D C A R has achieved a significant improvement in Pfau and Pdrop, compared with all the other techniques considered, including DCA. Particularly, in heavy traffic conditions, T D D C A R continues to maintain very good Pdrop performance, while the performance of all other techniques deteriorates considerably. Furthermore, as Fig. 4.6 shows, Pns for T D D C A R is lower than for any other techniques based on FCA. This means that the proposed T D D C A R technique can accommodate more calls with reduced forced terminations during the conversation. Although T D D C A R significantly reduces the cost of channel reservation, there is still some inevitable waste in terms of channel utilization. New calls cannot make use of the reserved channels and those channels will stay idle waiting for possible incoming handover calls. Therefore, when using the same channel allocation algorithm, D C A performs better in terms of Pns. However, for increased traffic intensities, this advantage is significantly reduced (See Fig. 4.6). 4.4 Conclusions This chapter presents a novel and efficient T D D C A R technique that fully exploits the deterministic movement of the MTs and the efficiency of the D C A technique to provide an effective way of reducing the cost of channel reservation in LEO-MSS. Performance evaluation results obtained by means of computer simulation have shown that, when compared with other known channel resource management strategies (e.g., D C A , DCR, QH), the proposed T D D C A R 71 technique has demonstrated much better performance for LEO-MSS. It can effectively reduce the Pdrop and significantly improve overall system performance, especially under heavy traffic, where the performance of other techniques decreases considerably. All of our performance evaluation results obtained so far have clearly shown that the proposed TDDCAR technique is an excellent handover strategy candidate for LEO-MSS. 72 Vtrk = 26600 km/h Figure 4.1 Rectangular cell model for LEO-MSS. B Handovar C < > New Call Generation Termination V,rk = 26600 km/h Figure 4.2 Traffic-dependent dynamic channel reservation procedures. NPS -e FCR1 —*• FCR2 -e- DCR -o-DCR, FIFO —<t— DCA TDDCAR 10 6 6.5 7 7.5 8 8.5 9 9.5 10 Traffic intensity per eel (erl) Figure 4.4 Pfan vs. traffic intensities for different resource management techniques. 10 = 10 _ -e - - — <P -+- • NPS —e FCR1 —»- FCR2 - s - • CR -o- DCR, FIFO —«— DCA - 0 - TDDCAR 7 7.5 8 8.5 Traffic intensity per eel (erl) 10 Figure 4.5 Parop vs. traffic intensities for different resource management techniques. 7.5 8 8.5 Traffic intensity per eel (erl) 10 Figure 4.6 Pns vs. traffic intensities for different resource management techniques. 75 4.5 References [1] M . Werner, A . Jahn, E. Lutz and A. Bottcher, "Analysis of system parameters for LEO/ICO-satellite communication networks", IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 371-381, 1995. [2] A . Ganz, Y . Gong and B. L i , "Performance study of low earth orbit satellite systems", IEEE Trans. Commun., vol. 42, pp. 1866-1871, 1994. [3] C. E. Fossa, R. A . Raines, G. H . Gunsch and M . A . Temple, "An overview of the miDIUM® low earth orbit (LEO) satellite system", Proc. of IEEE 1998 National Aerospace and Electronics Conf, pp. 152-159, 1998. [4] E. Del Re, " A coordinated European effort for the definition of a satellite integrated environment for future mobile communications", IEEE Commun. Mag., pp. 98-104, 1996. [5] G. Maral, J. Restrepo, E. D. Re, R. Fantacci and G. Giambene, "Performance analysis for a guaranteed handover service in an LEO constellation with a 'satellite-fixed cell' system", IEEE Trans. Veh. Technol, vol. 47, no. 4, pp. 1200-1213, 1998. [6] D. Hong and S.S. Rappaport, "Traffic model and performance analysis for cellular mobile radio telephone systems with prioritized and nonprioritized handoff procedures," IEEE Trans. Veh. Technol., vol. VT-35, no. 3, pp. 77-92, 1986. [7] E. Del Re, R. Fantacci and G. Giambene, "Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks," IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 397-405, 1995. 76 [8] Z. Wang and P. T. Mathiopoulos, "Analysis and performance evaluation of dynamic channel reservation techniques for LEO mobile satellite systems," IEEE VTC2001 Spring, VTS 53rd, vol. 4, pp. 2985 -2989, 2001. [9] E. Del Re, R. Fantacci and G. Giambene, "Different queuing policies for handover requests in low earth orbit mobile satellite systems," IEEE Trans. Veh. Technol., vol. 48, no. 2, pp. 448-458, 1999. [10] E. Del Re, R. Fantacci and G. Giambene, "Handover queuing strategies with dynamic and fixed channel allocation techniques in low earth orbit Mobile Satellite Systems," IEEE Trans. Commun., vol. 47, no. 1, pp. 89-102, 1999. 77 5 Channel Resource Management Strategies for Multi-party Traffic: Performance Analysis and Improvements1 5.1 Introduction Third generation (3G) wireless systems are expected to provide wide-area wireless multimedia services, including multi-party conferencing such as voice and video conferences, distance learning, online multiplayer gaming, etc. [1]. With their inherent broadcast capability and large geographical coverage, satellite networks are ideally suited for long-distance and large-scale multi-party applications, especially for conference participants with broad geographical dispersion or in remote isolated areas where no terrestrial network access is available [2]. Low Earth Orbit Mobile Satellite Systems (LEO-MSS) have some additional properties that are very promising in providing global-coverage real-time mobile communication services to small handheld Mobile Terminals (MTs). Compared with Geostationary Satellite Systems (GEO-MSS), LEO-MSS require lower transmit power and have shorter transmission delay, thus allow direct communication with low-power handheld MTs. Furthermore, LEO-MSS provide for earth coverage with smaller cells than those obtained by GEO-MSS, and thus have higher traffic capacities [3, 4]. 1 A version of this work has been submitted to IEEE Trans. Veh. Technol. Z. Wang, P. T. Mathiopoulos and R. Schober, "Channel resource management strategies for LEO-MSS with multi-party traffic: Performance analysis and improvements". 78 However, there are also obstacles and challenges that LEO-MSS face in order to provide multi-party services with satisfying quality. For example, in multi-party calls each party will be simultaneously requesting channels from different cells in LEO-MSS and will hold all channels during the conference. Such calls fail if one party is blocked or encounters a handover failure. Moreover, the duration time of a multi-party conference call is usually longer than that of a two-party call, which results in more handovers during the call's lifetime. However, for LEO-MSS with "satellite-fixed" cells2, inter-beam handovers occur rather frequently during a call's lifetime due to the high-speed movement of the satellites and their relatively small size spotbeams. Thus, channel resource management techniques are necessary and of great importance so that the call blocking probability, Pb, and handover failure probability, Pf, can be reduced as much as possible. Some studies have considered various resource management techniques for LEO-MSS, including Fixed Channel Reservation (FCR) and Dynamic Channel Reservation (DCR) techniques [6], Dynamic Channel Allocation (DCA) schemes [7]-[9], and queuing policies of handover requests (QH) [10, 11]. However, all of these works considered traffic that requires only a single LEO-MSS channel, e.g., the traffic generated between a user of a landline telephone and another user of a satellite phone. Although there have been publications addressing routing issues for multicast applications in LEO-MSS (e.g., [12, 13]), the problem of effective use of resource management techniques to efficiently support multi-party traffic has not yet been 2 Such cells can be thought of as "moving" around the earth according to the satellite motion (e.g., Iridium system [5])-79 thoroughly studied in the open technical literature. Motivated by the above observations, in this chapter we propose an analytical framework suitable for evaluating the performance of the various channel resource management strategies used in LEO-MSS supporting multi-party traffic, which require simultaneous use of more than one LEO-MSS channel. In order to improve overall performance, we also introduce a new Adaptive Channel Reservation (ACR) scheme and a New Call Queuing (NCQ) policy complementary to the A C R scheme. This NCQ scheme significantly reduces the blocking probability while having little impact on the other performance measures of overall system performance. The remainder of this chapter is structured as follows. In Section 5.2 the system model and parameters are introduced. Section 5.3 presents the theoretical performance analysis of LEO-MSS supporting multi-party traffic with FCR. In Section 5.4, the proposed A C R and NCQ schemes are described. Analytical and simulation results are presented and discussed in Section 5.5. Finally, our conclusions are drawn in Section 5.6. 5.2 System Model and Parameters Although there are several LEO-MSS models available, we adopt Iridium as our model as in [7]-[ll]. It should be noted, however, that the methodology proposed in this chapter can be applied to any LEO-MSS with satellite-fixed cells. The Iridium system consists of 66 satellites that are equally distributed in six near-polar circular orbits at an altitude of about 780 km. The 80 satellite ground-track speed Vtrk is approximately 26,600 km/h, which is much greater than the typical speed of a terrestrial MT. Thus, the relative satellite-MT motion can be approximated by Vtrk, i.e., the MTs will cross the cellular network with a constant speed of 26,600 km/h in a single direction [10, 11]. Each multi-spotbeam antenna from a single satellite irradiates on the earth a regular honeycomb cellular network, and many regular-shaped footprints of the L E O satellites cover the entire earth's surface [10]. In our study, following [14], each cell is modeled as a rectangular bounded cell of length L = 425 km. As illustrated in Fig. 5.1, the considered complete MSS network consists of 49 rectangular-shaped cells. The network wil l repeat itself in all four directions so that the border cell is adjacent to the cell(s) at the other side of the network (see the "dummy adjacent cells" in Fig. 5.1). Thus, a continuous cellular network wil l be considered, so that the calls going out from the right-hand side border cell will request a handover from the 1 s t cell at its left-hand side. Consistent with other publications (e.g., see [6]-[l 1]) relevant to the system model adopted in this chapter, the following typical assumptions are made: (1) uniform traffic: MTs are considered uniformly distributed over the simulation area so that a new call or each party of a multi-party call arrives anywhere within the satellite cellular network with equal probability; (2) the new and handover call arrival processes are modeled as independent Poisson processes; (3) the call duration time is exponentially distributed; (4) the average call origination rate is independent of the number of calls in progress; (5) an ideal propagation link exists. 81 The following are the most important system model parameters that we have adopted in our analysis and computer simulations: (1) the new call arrival rates of each cell are Ak for the &-party traffic; (2) the call duration time for &-party calls is tdk with an average value of 7^; (3) a Fixed Channel Allocation (FCA) technique is adopted and a set of C channels is permanently assigned to each cell; a M T in a cell can only be served by the channels belonging to that cell; (4) for each multi-party call, the maximum number of parties allowed is K. Furthermore, we use the following well-known Quality of Service (QoS) parameters [7]-[l 1] to evaluate the performance of various resource management strategies: (1) P0k'- the blocking probability of the &-party new call attempts, which represents average fraction of new &-party calls that are not admitted into LEO-MSS because of unavailability of channels; (2) P#: the handover failure probability of the &-party calls, which represents the average fraction of handover attempts of the &-party call that are unsuccessful; (3) Pdk- the call dropping probability of the &-party calls, which represents the average fraction of new A>party calls that are not blocked but eventually forced into termination due to the handover failure; (4) PUSk. the unsuccessful call probability of the A:-party traffic, which represents the fraction of new &-party calls that are not completed because of either being blocked initially or being dropped due to the failure of subsequent handover requests. 5.3 Performance Analysis Since in LEO-MSS handovers occur frequently in a call's lifetime, there will be more 82 handover failures, which clearly are much more undesirable than new calls being blocked. Thus, techniques that prioritize handover requests over new calls, such as FCR, are of great importance. When FCR is employed, the system reserves a fixed number of channels exclusively for the handover requests among all the channels in a cell and the remaining channels are shared by both new calls and handover calls. Based on the system model, as well as the assumptions and parameters described previously, we present an analytical methodology in this section that enables the fast and accurate evaluation of the performance of LEO-MSS supporting multi-party traffic with FCR. Let Ak' be the arrival rate of the calls with k (1 <k<K) parties. Since there are k parties, the total traffic generated by the &-party calls is \ = k A k \ (5.1) Let Yk = Tdk/ Tc, where Tc = L / Vtrk denotes the user sojourn time in a cell and Tdk is the mean call duration time for the &-party calls. As in [11], the source cell is the cell where the new call starts, and transit cells are subsequent cells reached by the M T while the call is in progress. The handover probability of a M T in its source/transit cell (Ph\\JPh2\d can be expressed, respectively, as [11] ^ u = r * ( l - e " ( I / r t ) ) ^ d PhU=e-lUn). (5.2) The new call arrivals and handover requests of &-party traffic are assumed to be two independent Poisson processes with mean rates of Ak and Ahk per cell, respectively. Considering a statistical equilibrium between MTs leaving and entering a cell, Ak can be related to Ahk as 83 / L t ( l - ^ ) + A M ( l - ^ ) = A M + A t ( l - P M ) ( l - / > u ) + A M ( l - P # ) ( l - P A 2 J . (5.3) On the left-hand side (LHS) of the above equation, there are two sources of &-party calls that enter the cell: the newly admitted calls and the successful handover calls. On the right-hand (RHS) side, there are three types of A>party calls that leave the cell: the calls handed over to the destination cell and the calls ending in the cell without requesting another handover. Thus, from Eq. (5.3), we have (5.4) K \-(\-Pjk)Phlk The expected value of the channel holding time in a source/transit cell (Ek[th\kVEk[th2k]) can be expressed as [11] Ek[thik} = Tdk(\-Phik), ,-=1,2. (5.5) The total new call arrival rate X and the handover request arrival rate Xh can be expressed as K K X = Y^K and Xh = J^Xhk. (5.6) k=\ k=\ Similar to [11], the channel holding time in a cell for both new arrivals and handovers can be approximated with an exponential distribution of mean, l/ju, as follows: iK(i-^)^K , j+^a-^)^[^]} - = ^  ? , (5.7) k=\ ^{Ui-p^ + x^i-p^} Let Ch denote the channels reserved exclusively for handover requests. Each cell can be modeled as an M/M/C/S queuing system with non-homogeneous arrival rates. The state of this queuing system is defined as the sum of the number of calls in service, and its state transition ' 84 diagram is illustrated in Fig. 5.2. Denoting Pj as the probability of state j (1 < j < Q , the following "rate-up = rate-down" state equations can be obtained (see Fig. 5;2): 1 + 1, JM I JM hPj_x, \<j<C-Ch C - C h + \ < j < C (5.8) The recursive use of the above equation, along with the following normalization condition c results in the following probability distribution, Pf. (5.9) c-a k=0 + C-Ch jk-iC-C,,) k=C-Ch+\ -r1, (5.10) PJ = (X + Xh)J 0 ' (C-Ch) ^j-(C-C„) 1 < j < C - C, (5.11) -P0, C - C h + l < j < C The probability of blocking a single-party call, Pb\, is the sum of the state probabilities when the queuing system is in state j > C-Cn, i.e., j=c-ch The handover failure probability, Pj\, can be calculated as (5.12) (5.13) Without giving any priority to multi-party handover requests, the handover failure probability of &-party calls, P/k, equals Pj\, i.e., 85 Pr l<k<K. (5.14) For a &-party call, other parties will have the same statistical properties. Thus, when taking other parties into account, Pbk can be expressed as Similar to [6] - [11], in the above analysis it has been assumed that the new and handover call arrivals are independent Poisson processes. In order to be more accurate, their dependency is taken into account in the following analysis. Handover traffic is generated only when the newly arriving call has a duration longer than its dwell time in one cell. A given call will have i handovers i f the newly arriving call has a call duration long enough to generate i handovers and the system can provide the channels to the new call and its subsequent i-l handovers. After the successful i-l handovers, the admitted call will continue to generate its rth handover request. Therefore, the mean value of the number of times that a newly arriving A:-party call is handed over during its lifetime, tihk, can be derived as />* = 1 - ( 1 - / > „ ) * . • (5.15) n Z « i - ^ ) ^ ( i - ^ ) w ^ } = (5.16) hk ~ 1=1 The mean number of handovers for all incoming calls, nf,e, can then be calculated as K K (5.17) 4=1 i=l Similarly, the mean values of Pbk, Phik, and Phik can be derived as K K (5.18) 86 ^ i . = £^u(VZ4) . (5-19) 4=1 and ^ e = Z^(V2><)' (5-20) 4=1 i=l respectively. For failure during the w t^h handover, the call must be admitted into the system and be able to generate and successfully perform («/,e - 1) handovers. Therefore, in order to model the dependency between these handovers, the following correction factor, 5, is being introduced S = (1" PJPkl.(l ~ Pfi)*--2PZfe'2, (5-21) so that the handover failure probability in Eq. (5.14) can be revised as Pfk=PcS, \<k<K. (5.22) Without considering the dropping probabilities of other parties in this &-party call, a call that is not blocked will eventually be forced into termination if it succeeds in each of the first (z'-l)th handovers but fails in the /th. Therefore, denoting Pdk' as the dropping probability when considering only one party of the &-party call, Pdk and the handover failure probability Pjk, satisfies OO P P Pdk'=I^d-V^lK p 7 p V <5-2 3> 1=1 1 "h2k\l "jk) For a &-party call, when taking other parties of the call into account, the call dropping probability Pdk can be expressed as Pdk=\-(\-Pdk<f. (5.24) 87 The unsuccessful call probability, Pusk, resulting from either blocking or unsuccessful handover, is also used as a major parameter for evaluating overall system performance. This probability is the sum of the new call blocking probability and the probability that a call is admitted but eventually dropped due to handover failure, i.e., Pusk=Pbk+Pdk(y-Pbk)- (5-25) Denoting pk as the traffic intensity of new &-party call arrivals, Pk = KTdkiErlanSs) • (5-26) To study the system performance under different traffic intensities, we can specify or change the value of pk and calculate Xk using Eqs. (5.26) and (5.1). From Eqs. (5.12) - (5.15) and (5.23) -(5.25), it can be observed that in order to obtain Pbk, Pjk, Pdk, and Pusk, the probability of each system state Pj is needed. Clearly, Pj is related to A and Xh (see Eqs. (5.10) and (5.11)). As Xhk relates to Pbk and Pjk via Eq. (5.4), a recursive approach is needed to compute Pbk. We start the iterations with Pbk= 0 and compute X, Xh and JU using Eqs. (5.2) - (5.7), then use X, Xh and ju to calculate Pj for j = 1, 2, C. From Eqs. (5.12) - (5.15), the new value of Pbk can be obtained. A new iteration starts with this new value of Pok. The iteration ends when the relative difference o between the Pbk values computed in two consequent steps is less than 10" . 5.4 Performance Improvement Techniques Compared to single-party traffic, multi-party traffic has much higher new call blocking and handover call dropping probabilities. With the aim of reducing these probabilities, in this section, 88 w e introduce t w o ef f ic ient channel resource management techniques. The f i rst is based on an Adap t i ve Channel Reservat ion ( A C R ) scheme, whereas the second uses a N e w Ca l l Queuing ( N C Q ) approach. The effectiveness o f these two techniques i n i m p r o v i n g overal l system performance w i l l be presented i n Sect ion 5 .5 . 5.4.1 Adaptive Channel Reservation (ACR) Scheme Resource management strategies are essential fo r reduc ing handover fai lures i n L E O - M S S and par t icu lar ly cr i t ica l fo r mu l t i -par ty t ra f f ic , as these calls need more channel resources and usual ly last longer, generat ing more handovers. F C R is very in f lex ib le as i t does not consider the actual t ra f f ic condi t ions. I n order to improve the ef f ic iency o f channel reservations, w e explo i t the high-speed determinist ic movement proper ty o f L E O satellites and propose a nove l and ef f ic ient A C R scheme to ef fect ive ly reduce handover fai lures fo r mu l t i -par ty calls i n L E O - M S S and thereby achieve ext remely l o w handover fa i lure probabi l i ty . Based on the number o f channels i n use b y mu l t i -pa r ty cal ls, the A C R scheme reserves a number o f channels fo r mu l t i -par ty t ra f f ic and adjusts this number w i t h a dynamic f requency corresponding to the current t raf f ic condi t ions. The number o f reserved channels is updated f requent ly and after every related event, i.e., new cal l arr ivals, handovers, and ca l l terminat ions. O n the one hand, w h e n t ra f f ic becomes heavier, events w i l l happen more of ten, and accord ing ly the number o f reserved channels w i l l be updated more f requent ly f o l l o w i n g the changing needs o f channel reservation. O n the other hand, w h e n t raf f ic intensi ty is l o w , the need to adjust the 89 number of reserved channels is not critical for guaranteeing the availability of channels for handover calls. Therefore, by updating the number of reserved channels every time such an event happens, a "real-time" monitoring of the channel reservation requirements is taking place that should improve the overall LEO-MSS system efficiency and performance. This A C R process is graphically illustrated in Fig. 5.3 and will be explained next. With the arrival of a new multi-party call in Cell 1, the system will calculate how many channels in Cell 1 are used by multi-party calls and Cell 2 will reserve the same number of channels if it has a sufficient number of channels available. When a handover is performed from Cell B to Cell C, their traffic conditions have changed, so their reservation needs must be re-calculated. In the meantime, because there is a channel released in Cell B, it can be reserved for the additional handovers from multi-party calls in Cell A if Cell A does not have enough reserved channels. Therefore, Cells B, C, and D will adjust the number of their reserved channels according to the changing traffic conditions. Similarly, when a call terminates in Cell I, Cells I and II will adjust the number of their available reserved channels accordingly. 5.4.2 New Call Queuing (NCQ) Scheme Although the call dropping probability Pdk can be greatly reduced by employing the proposed A C R scheme, in practice, for multi-party applications the high blocking probability Pbk remains an unresolved issue. Each party of a multi-party call arrives at different cells and simultaneously requests channels from its host cell. Assuming the entire multi-party call will fail if one party is 9 0 blocked, obviously the blocking probability will be much higher as compared to a single-party call. Due to the high-speed movement of LEO satellites, the MTs currently occupying the channels in a certain cell will move to the next cell within a maximum time frame of LIVtrh We can take advantage of this characteristic and implement NCQ to greatly improve performance for multi-party call admission. Instead of being blocked, the newly arrived multi-party call can be queued to wait for channels to become available within a certain time frame (i.e., maximum queuing time Tqmax) if there is no channel available immediately. Every time a channel is released in a cell, it will first serve the new multi-party call requests waiting in the queue. While one or more parties of the multi-party call are waiting for channels to become available, the admitted parties will hold their channels and wait for every party to be admitted into the system. If Tqmax expires before the multi-party requests in queue manage to get channels, the requests in the queue will be cleared, and those that have already received the channels will release them. In this case, the multi-party call is blocked. If all parties get the channels before Tqmax expires, the multi-party call is established. For the current cell, the channels are released frequently due to the handover of MTs. Moreover, the MTs with requests in queue are also moving fast from one cell to the next. Therefore, by implementing NCQ, the multi-party requests are expected to have a much better chance of getting channels within a short waiting time. 91 5.5 Performance Evaluation Results and Discussion In this section, various analytical and computer simulation results obtained using the analysis and methodology presented in Sections 5.3 and 5.4 are summarized and discussed. Performance evaluation results for a LEO-MSS with incoming traffic from two sources, i.e., single-party and 7-party traffic, are obtained through both theoretical analysis and computer simulations. To maintain the necessary accuracy for the performance results obtained by computer simulation, we have chosen the following numerical values for the different system parameters: C = 20, pj = 0.06/?i, Td\ = 180 sec, and Tai = 37^1. The methodology proposed in Section 5.3 is verified by Figs. 5.4 and 5.5. With a fixed number of reserved channels Cn=2, Fig. 5.4 shows the performance of the four QoS parameters for the 7-party traffic. For comparison purposes, equivalent performance results for single-party traffic are also presented in Fig. 5.5. The analytical results illustrated in Figs. 5.4 and 5.5 were obtained using Eqs. (5.15), (5.22), (5.24) and (5.25), and clearly match well with the equivalent ones obtained by means of computer simulation. Nevertheless, some small differences between analytical and simulation results can be observed in Figs. 5.4 and 5.5. Although it is very difficult to justify this difference theoretically, it is believed that this happens because of the assumptions of independent new and handover call arrivals and the approximation used for the channel holding time in a cell for both new arrivals and handovers (see Eq. (5.7)), which were adopted to ensure mathematical tractability. 92 Next, the performance of the proposed schemes in Section 5.4 wil l be presented. In Figs. 5.6-5.8, we have summarized the most important performance results, i.e., Pbk, Pjk, Pdk, and Pusk, for the proposed A C R and FCR schemes. As can be observed from these figures, A C R achieves significantly better performance compared with FCR in every QoS performance measure. It should be also noted that in Figs. 5.6-5.8 there is no curve for P-p and P^due to their extremely low values. For example, extensive simulations have shown that no more than 4 failures in more than 6x l0 5 handovers occurred. Nonetheless, our simulation data shows that A C R has a lower number of handover failures and dropped calls than FCR with Ch = 4. Therefore, it is clear that the proposed A C R approach provides an effective way to reduce the cost of channel reservation. The conventional FCR scheme, however, because of the fixed reservation, which obviously does not take into account the traffic conditions, tends to waste the channel resources by unnecessarily reserving too many channels when traffic is light, or results in higher handover failures when traffic is heavy. This performance improvement makes the proposed A C R scheme an excellent handover strategy candidate for LEO-MSS. Figs. 5.9 and 5.10 show the performance improvement when implementing NCQ for multi-party traffic. An infinite queue is adopted and Tqmax of different values have been investigated. When setting Tqmax > 30 sees in our simulations, Pbk was reduced to zero. Thus, in Fig. 5.9 there is no Pbi curve for cases with NCQ. Furthermore, from Figs. 5.9 and 5.10 it can be seen that the impact NCQ has on other QoS parameters is almost negligible. For single-party calls, Pb\IPu\ and Pj\l Pd\ are only slightly higher when employing NCQ. For multi-party calls, while PplPdi increases slightly, the total number of unsuccessful calls is less with NCQ because 93 there is no multi-party call being blocked. In addition, N C Q can be used in conjunction with A C R to achieve highly guaranteed admission and handover success for multi-party calls. In our simulation results, when combining the two techniques, there is no multi-party call being blocked or forced to terminate. This is the reason why in Figs. 5.9 and 5.10, there are no PplPdilPbi curves for cases with NCQ in conjunction with ACR. 5.6 Conclusions In this chapter, an analytical methodology for evaluating the system performance for LEO-MSS supporting multi-party traffic with FCR has been developed. Our analytical results are in good agreement with the results obtained by computer simulation, which verifies our system model and analytical methods. Furthermore, in order to improve the performance of FCR and to exploit the high-speed deterministic movement property of LEO satellites, we have proposed an efficient A C R scheme for multi-party traffic. The proposed scheme outperforms the traditional FCR scheme while introducing little additional computational complexity. When used in conjunction with NCQ, extremely low blocking and handover failure probabilities can be achieved for multi-party traffic in LEO-MSS. The performance evaluation results provided show that the proposed A C R and NCQ schemes are attractive channel resource management techniques for LEO-MSS with multi-party traffic. 94 Out Vtrk = 26600 k m / h Figure 5.1 Rectangular cell system model for the LEO-MSS under consideration. X + X^ X + Xh X + Xh X + Xh X + Xh Xh Xh Xh Figure 5.2 State-transition diagram for FCR in LEO-MSS. • ] L 2 A B Han( C —> over D I II t New Multiparty Termination call generation < Vtrk= 26600 km/h Figure 5.3 Adaptive channel reservation procedures. Figure 5.4 Performance evaluation results for the various probabilities as a function of traffic intensity for 7-party traffic with FCR (Ch = 2). 96 IO' 7 I I I I I I I I I I I 4 4.5 5 5.5 6 6.5 7 7.5 B B.5 9 Traffic intensity per cell (erl, single-party traffic) Figure 5.5 Performance evaluation results for the various probabilities as a function of traffic intensity for single-party traffic with FCR (C/, = 2). Figure 5.6 Pb\ and Pus\ as a function of traffic intensity for FCR and ACR. 6.5 7 7.5 8 Traffic intensity per cell (erl, single-party traffic) • — *•-» P U 3 7 8.5 Figure 5.9 Pbk and PUSk as a function of traffic intensity for NCQ and ACR. Figure 5.10 Pjk and Pdk as a function of traffic intensity for NCQ and ACR. 5.7 References [1] V. Y. H . Kueh, N . Wang and B. Evans, "Performance evaluation of conference creation signalling over satellite UMTS" , Proc. of IEEE VTC 2005 Spring, vol. 4, pp. 2663-2667, 2005. [2] T. Asfour-Block and A. Serhrouchni, "Reliable multicast over satellite networks", Proc. 10th IEEE Symposium on Compu. and Commun., pp. 698-703, 2005. [3] C. E. Fossa, R. A. Raines, G . H . Gunsch and M . A . Temple, "An overview of the IRIDIUM ® low earth orbit (LEO) satellite system", Proc. of IEEE National Aero, and Electro. Conf, pp. 152-159, 1998. [4] E. Del Re, " A coordinated European effort for the definition of a satellite integrated environment for future mobile communications", IEEE Commun. Mag., pp. 98-104, 1996. [5] http ://www.iridium .com [6] Z. Wang and P. T. Mathiopoulos, "Analysis and performance evaluation of dynamic channel reservation techniques for LEO mobile satellite systems", Proc. of IEEE VTC 2001 Spring, vol. 4, pp. 2985-2989, 2001. [7] E. Del Re, R. Fantacci and G. Giambene, "An efficient technique for dynamically allocating channels in satellite cellular networks", Proc.of IEEE GLOBECOM '95, pp. 1624-1628, 1995. 100 [8] E. Del Re, R. Fantacci and G. Giambene, "Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks", IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 397-405, 1995. [9] Z. Wang and R T. Mathiopoulos, "On the performance analysis of dynamic channel allocation with FIFO handover queuing in LEO-MSS", IEEE Trans. Commun., vol. 53, issue. 9, pp. 1443-1446, 2005. [10] E. Del Re, R. Fantacci and G. Giambene, "Handover queuing strategies with dynamic and fixed channel allocation techniques in low earth orbit mobile satellite systems", IEEE Trans. Commun., vol. 47, no. 1, pp. 89-102, 1999. [11] E. Del Re, R. Fantacci and G. Giambene, "Different queuing policies for handover requests in low earth orbit mobile satellite systems", IEEE Trans. Veh. Technol., vol. 48, no. 2, pp. 448-458, 1999. [12] E. Ekici, I. F. Akyildiz and M . D. Bender, " A multicast routing algorithm for LEO satellite IP networks", IEEE/ACM Trans. Networking, vol. 10, issue 2, pp.183-192, 2002. [13] D. Yang and W. Liao, "On multicast routing using rectilinear Steiner trees for LEO satellite networks", Proc. of IEEE GLOBECOM '04, vol. 5, pp. 2712-2716, 2004. [14] G. Maral, J. Restrepo, E. Del Re, R. Fantacci and G. Giambene, "Performance analysis for a guaranteed handover service in an LEO constellation with a 'satellite-fixed cell' system", IEEE Trans. Veh. Technol, vol. 47, no. 4, pp. 1200-1213, 1998. 101 6 Analysis and Performance Evaluation of Channel Partitioning Policies for Multi-class Traffic1 6.1 Introduction With the advancement of modern communication technologies and the rapid growth of the wireless communication market, third generation (3G) wireless systems are expected to provide wide-area wireless multimedia services, such as multimedia games, video-on-demand, music/video broadcasting, etc. With their broadcast capability and large geographical coverage, satellite networks are ideally suited for long-distance and large-scale multimedia applications. In particular, Low Earth Orbit Mobile Satellite Systems (LEO-MSS) have some additional properties that are very promising in providing global coverage real-time mobile communication services to small handheld Mobile Terminals (MTs). Compared with Geostationary Mobile Satellite Systems (GEO-MSS), LEO-MSS require relatively low transmit power and short transmission delay, which permit reliable communication between low-power handheld MTs and satellites. In addition, LEO-MSS allow for earth coverage with smaller cells than those with GEO-MSS, thus resulting in higher traffic capacities [1, 2]. In order to provide multimedia services with acceptable quality using LEO-MSS, channel resource partitioning policies that determine the number of channels to be assigned to different 1 A version of this work will be submitted in the near future for possible publication in an IEEE Transactions journal. Z. Wang, P. T. Mathiopoulos, and R. Schober, "Performance analysis and evaluation of channel partitioning policies for multi-class traffic in LEO mobile satellite systems". 102 classes of traffic need to be studied. Different classes of traffic (e.g., voice, video) have different bandwidth requirements and call duration time. In addition, for those LEO-MSS with "satellite-fixed" cells, i.e., cells move on the earth according to the satellite motion (e.g., the Iridium system [3]), inter-beam handovers occur rather frequently during a call's lifetime. This happens because of the high-speed movement of LEO satellites and their relatively small size spotbeams. Thus, channel reservation techniques, which can effectively reduce the probability of handover failure, Pf, are necessary to support reliable communications. In the past, there has been some work investigating Fixed Channel Reservation (FCR) and Dynamic Channel Reservation (DCR) techniques [4], Dynamic Channel Allocation (DCA) schemes [5]-[7] and Queuing policies of Handover requests (QH) [8, 9] in LEO-MSS. However, all of these works considered traffic requiring only one channel (per call) from the employed LEO-MSS. Although there have been publications dealing with multi-class traffic in LEO-MSS (e.g., [10]-[13]), to the best of our knowledge, the subject of effective channel partitioning techniques for supporting multi-class communications in LEO-MSS has not been thoroughly investigated. In this chapter, we develop an analytical framework for theoretically evaluating the performance of multi-class traffic when employing the most primary but important channel resource partitioning schemes, namely, Complete Sharing (CS) and Complete Partitioning (CP) [14], in LEO-MSS. Our approach is general enough to also include Fixed Channel Reservation (FCR). Using a multi-dimensional Markov chain, the proposed analytical models are solved and explicit expressions for QoS parameters such as call blocking and dropping probabilities, 103 Handover failure probability and unsuccessful call probability are derived. The accuracy of the proposed theoretical analysis methods has been verified by means of computer simulations. To further improve system performance, a novel Threshold Call Admission (TCA) policy is introduced. The proposed T C A scheme overcomes the unfairness of CS, improves the channel utilization of CP and offers more flexibility for allocating channel resources to different classes of traffic according to their needs. The remainder of this chapter is structured as follows. After this introduction, in Section 6.2, the LEO-MSS model and different system parameters are introduced. Sections 6.3 and 6.4 present the theoretical analysis for multi-class traffic under the CS and CP policies, respectively. The T C A policy is described in Section 6.5, and the analytical and computer simulation performance evaluation results are presented and discussed in Section 6.6. Finally, conclusions are drawn in Section 6.7. 6.2 System Model and Parameters Although there are several LEO-MSS models available, we have adopted the Iridium model as have other publications (e.g., [5]-[9]). It should be noted, however, that the methodology proposed in this chapter can be applied to any LEO-MSS with satellite-fixed cells. The Iridium system consists of 66 satellites that are equally distributed in six near-polar circular orbits at about 780 km of altitude. The satellite ground-track speed Vtrk is approximately 26,600 km/h, which is much greater than the typical speed of the MTs on earth. Thus, the relative satellite-MT 104 motion can be approximated by Vtrk, i.e., the MTs will cross the cellular network with a constant speed of 26,600 km/h in a single direction [8, 9]. Each multi-spotbeam antenna from a satellite irradiates a regular honeycomb cellular network on the earth and many regular-shaped footprints of LEO satellites cover the entire earth's surface [8]. In our study, following [15], each cell is modeled as a rectangular bounded cell with cell length L = 425 km. As illustrated in Fig. 6.1, the considered MSS network consists of N rectangular-shaped cells. The network will repeat itself in the opposite direction of Vtrk so that the border cell is adjacent to the cell(s) at the other side of the network, thus simulating a continuous cellular network. The calls going out from the right-hand side border cell wil l request a handover from the first cell at the left-hand side. We further make the following typical and well-accepted assumptions2: (1) uniform traffic: MTs are considered uniformly distributed over the simulation area so that a new call arrives anywhere within the satellite cellular network with equal probability; (2) the new and handover call arrival processes are modeled as independent Poisson processes; (3) the call duration time is exponentially distributed; (4) the average call origination rate is independent of the number of calls in progress; and (5) an ideal propagation link exists. The following are the most important system model parameters we have adopted in our analysis and computer simulations: (1) the new call arrival rates of each cell are Xk for class k traffic; (2) the call duration time for class k traffic is tdk, with an average value of Tdk\ (3) the Fixed Channel Allocation (FCA) technique is adopted, and a set of C channels is permanently 2 Similar assumptions have also been made in other publications, including [4] - [9]. 105 assigned to each cell; a M T in a cell can only be served by the channels belonging to that cell; (4) the maximum number of classes allowed is K. Furthermore, we use the following well-known Quality of Service (QoS) parameters [4]-[9] to evaluate the performance of various resource management strategies under investigation: (1) Pbk'- blocking probability of the class k new call attempts; (2) P#: handover failure probability of the class k calls; (3) Pdk'- call dropping probability of the class k calls; and (4) Pusk\ unsuccessful call probability of the class k traffic. We note that an unsuccessful call happens when the call is initially blocked or is dropped due to the failure of subsequent handover requests. 6.3 Performance Analysis of Complete Sharing Based on the system model, assumptions and parameters described previously, in this section we develop an analytical methodology to evaluate the performance of multi-class traffic in the LEO mobile satellite network under the CS policy. As will be shown later, this approach enables fast and accurate performance evaluation of multi-class traffic in LEO-MSS. Note that with the CS channel partitioning policy, all of the channels in a cell can be equally accessed by any multi-class traffic, and that an arriving call is admitted by the LEO-MSS whenever there are sufficient channels available. Let ft = Tdk/ Tc, where Tc - L / Vtrk denotes the user sojourn time in a cell and Tdk is the mean call duration time for the class k calls. As in [9], the source cell is the cell where the new call starts; transit cells are subsequent cells reached by the M T while the call is in progress. The 106 handover probability of a M T in its source/transit cell {Ph\klPhi\d can be expressed, respectively, as [9] Phu =y4(l-e" ( , / r t )) and J>M4 = e - ° ' » ) . (6.1) The new call arrivals and handover requests of class k traffic are assumed to be two independent Poisson processes with mean rates Ak and Ank per cell, respectively. Considering a statistical equilibrium between MTs leaving and entering a cell, A,k can be related to Ank as ^ ( l - ^ J + A M ( l - P ^ ) = A M + ^ ( l - P s J ( l - P M J + A M ( l - P ^ ) ( l - P A 2 t ) . (6.2) On the left-hand side (LHS) of the above equation, there are two sources of class k calls that enter the cell: the newly admitted calls and the successful handover calls. On the right-hand side (RHS), there are three types of class k calls that leave the cell: the calls handed over to the destination cell and the calls ending in the cell without requesting another handover. Eq. (6.2) can be rewritten as Kk _ 0-~pbk)ph\k (6.3) \ \-(\-Pjk)Ph2k Furthermore, the expected value of channel holding time in a source/transit cell {Ek[th\k]IEk[th2k\) can be expressed as [9] Ek[thik]-Tdk(l-Phik), , = 1,2. (6.4) Thus, the channel holding time in a cell for both new arrivals and handovers can be approximated as an exponential distribution with mean, \/juk, [9] (6.5) 1 _ Xk (1 - Pbk )Ek [thn ] + Xhk (1 - Pfi )Ek [th2k) Mk W-P^ + ^il-Pfi) 107 6.3.1 CS without FCR Let bk denote the number of channels required by class k traffic in one cell and iik denote the number of ongoing new and handover calls of class k traffic in the cell, then the total number of channels used by the MTs is b»n, where b = (bj, b2, hx), n = (ni, 112, .... TIK), and K b*n = Y,bknk- (6-6) k=\ The cell will accept the arriving call when there is a sufficient number of channels, i.e., it admits an arriving class k call i f bk < C - b»n; otherwise, it blocks the call. Such a procedure can be modeled by a Markov process with state space S, S = {neI:b»n<C}, (6.7) where / is the set of non-negative integers. The current state of the system can be represented by the vector n. Fig. 6.2 shows the Markov state diagram for two incoming traffic with b\ = 1 and 62 = 2, assuming C is an even number. For each n e S, denote P(n) as the probability that the system is in state n. The equilibrium distribution for this stochastic process can be derived as [15] P , , n I rQ k=l nk • where 1 = <69> neS k=\ nk • and pk is the traffic intensity of class k arrivals including both new and handover calls, given by 108 Pk = (\ +Kk)! Mk (Erlangs). (6.10) Let Skhe the subset of states in which the system admits an arriving class k call, then Sk = {neS:b*n<C-bk}. (6.11) The arrival process is Poisson, thus the probability that one arriving class k call gets blocked is ^=l-2>("). . (6.12) Substituting Eq. (6.8) into Eq. (6.12), the following expression for the blocking probability is obtained: ^M=1-^T • (6-13) neS j=\ Because the number of discrete state spaces S and Sk is large, even for moderate values of C and K, it is very difficult to directly calculate Pbk from the above equation. Therefore, a recursive method is used to calculate the blocking probabilities [15]. For this procedure, let S(c) = {neS:b*n = c}, (6.14) q(c)= X P(n), (6.15) neS(c) where S(c) is the set of states when c channels are in use and q(c) is the probability of this event. Let q(c) = 0 for c < 0. q(c), c = 1, 2, ..., C satisfies the recursive equation K cq(c) = YJbkpkq(c-bk), c = l , 2 , C. (6.16) k=\ 109 Exploiting the above equation, the channel occupancy probabilities q(c) can be calculated recursively. Blocking occurs when there are insufficient channels for the incoming call, so that the blocking probability of the class k call can be expressed as p»= i>(c>- (6-i?) c=C-bk+\ Without implementing any channel resource management techniques to give handover calls higher priority, the handover failure probability is equal to Pbk, i.e., Pjk=Pbk- (6-18) Similar to [4] - [9] and to make the mathematical formulation tractable, in the above analysis the new and handover call arrivals are assumed to be independent Poisson processes. However, in order to be more accurate, their dependency is taken into account in the following manner. Handover traffic is generated only when the newly arriving call has a call duration longer than its dwell time in one cell. A given call will have i handovers i f the newly arriving call has a call duration long enough to generate i handovers and the system can provide the channels to the new call and its subsequent z'-l handovers. After the successful z'-l handovers, the admitted call will continue to generate its z'th handover request. Therefore, the mean value of the number of times that a newly arriving class k call is handed over during its lifetime, nnk, can be derived as nhk =1 A O ~ p b k f l - V P & > = i In"- P ^ P • ( 6 " 1 9 ) 1=1 1 yl ~jk)"h2k For failure during the w^th handover, the call must be admitted into the system and be able to generate and successfully perform (nnk - 1) handovers. Therefore, in order to model the dependency between these handovers, the following correction factor, b\, is introduced, 110 (6.20) so that the handover failure probability in Eq. (6.18) can be revised as P. (6.21) A call that is not blocked will eventually be forced into termination i f it succeeds in each of the first (i-l) handovers but fails on the r'th. Therefore, Pdk and P# satisfy The unsuccessful call probability, Pusk, resulting from either blocking or an unsuccessful handover, is also used as a major parameter for evaluating overall system performance. This probability is the sum of the probability of a new call being blocked and the probability that a call is admitted but eventually dropped due to the handover failure, i.e., From Eq. (6.17), it can be observed that in order to obtain Pbk, q(c) is needed, as it is related to Xk and Xhk (see Eqs. (6.10) and (6.16)). In order to study the system performance under different traffic intensities, we can specify or change the value of Xk. However, as Xhk is related to Pbk via Eq. (6.3), a recursive approach is needed to compute Pbk- The iteration starts with Pbk- 0 and Xhk and jUk are computed using Eqs. (6.3) and (6.5). Then Xhk and /& are used to calculate q(c) for c = 1, 2, ... C. From Eq. (6.17), the new value of Pbk can be obtained, and then another iteration starts using this new value for Pbk- The iteration ends when the relative difference between the Pbk values computed in two consequent steps is less than 10"6. OO p p (6.22) Pusk ~ Pbk + Pdk 0 Pbk ) • (6.23) 111 6.3.2 CS with FCR In LEO-MSS, handovers occur rather frequently in a call's lifetime, which in consequence results in many more handover failures. It is well known that handover failures are much more undesirable than new calls being blocked. Thus, techniques prioritizing handover requests over new calls are of great importance. As the simplest but rather effective channel reservation approach, the FCR scheme has been applied in the past to LEO mobile satellite communication systems [16]. For FCR, the system reserves a fixed number of channels exclusively for the handover requests among all the channels in a cell and the remaining channels are shared by both new calls and handover calls. In the previous section, we analyzed the CS policy without giving priority to handover calls. Although this policy is simple to describe and administer, it suffers from high handover failure/call dropping probability because of the very high handover frequency taking place in LEO-MSS. Therefore, we are motivated to study the system performance when reserving some channels exclusively for handover requests. Let Ak(ri) denote the arrival rate and /&(») the departure rate of class k calls, respectively, when the system is in state n. When the system is in state n, whether or not the call is admitted into the system is determined by the FCR scheme. Letting C M represent the number of reserved channels for class k traffic, we have - , , [K + \k> b*n + bk<C-Chk [ hk' C-Chk + l<b»n + bk <C Mk(n) = nkMk- b*n<C. (6.25) 112 Furthermore, letting P(n) be the probability that the system is in state «, then the global balance equations for this ^-dimensional Markov process for multi-class traffic with FCR are as follows: £ [\ (n) + uk (n)]P(n) = £ Xk (« - ek )P(n - ek) + £ Mk (n + ek )P(n + ek), n E S (6.26) k=\ k=\ k=\ where ek is a .KT-dimensional vector of all Os except for a 1 in the Ath place. Using the following normalization condition, ZP(W) = 1> (6-27) neS Eq. (6.26) can be solved using any linear equation procedure, such as Gauss-Siedel iteration. Once the equations are solved, the QoS parameters such as PDk and Pjk can be directly calculated. A new class k is blocked i f there are less than bk channels available in addition to the Chk reserved channels. Let SDk be the subset of states in which the system blocks an arriving class k call. Then Sbk = {« e S: b • n > C-Chk -bk), (6.28) so that Pbk can be derived as Pbk = Z • (6-29) Similarly, by letting Shk be the subset of states in which the system blocks a handover class k call, i.e., Shk = {neS:b»n>C-bk}, . (6.30) Pjk can be written as ^ = ! > ( « ) . (6.31) 113 We then need to consider the relationship between new call arrival and handover traffic by multiplying by the correction factor b\, i.e., Pdk and Pusk can be calculated using Eqs. (6.22) and (6.23), respectively. A recursive algorithm similar to the one described in Section 6.3.1 can be used to solve the above equations. 6.4 Performance Analysis of Complete Parti t ioning The CP channel allocation scheme divides the available bandwidth into separate subsets according to the different user types and their requirements. Although CP can ensure the QoS of certain types of traffic, because of complete partitioning different classes of traffic cannot make use of the idle channels in other subsets, resulting in lower channel resource utilization. In the following two subsections, we analyze system performance employing CP with and without FCR. We denote bk as the number of channels required by each class k call in one cell, nk the number of ongoing new and handover calls of class k traffic in the cell and Ck the number of channels allocated for class k traffic. 6.4.1 CP without FCR Since all of the C channels in a cell have been completely partitioned into K subsets with Ck (1 < k< K) channels allocated to class k traffic, each partition can be modeled as an M/M/C/S queuing system. The state of this queuing system is defined as the sum of the number of class k (6.32) 114 calls in service, and its state transition diagram is shown in Fig. 6.3. By letting nik = Cjjbk and Pj denote the probability of state j, it can be derived from this figure that the "rate-up = rate-down" state equations are Juk Using the above equation recursively, along with the normalization condition mk ZP; = 1> (6.34) the probability distribution, Pj, is found as PO=[2J ~—}—] > (6.35) P j = ( Z k * ^ ' P o . l*j*mk. (6.36) The probability of blocking the class k calls, Pbk, is the sum of the state probabilities when the queuing system is in the state j = nik, i.e., Pbk=Pmk. (6-37) Without giving priority to handover calls, we can derive the handover failure probability of class k calls as Pjt = p m k • (6-38) Similar to the analysis in Section 6.3.1, factor b\ needs to be taken into account, so that Pjk = PbA • (6-39) Using Eqs. (6.22) and (6.23), Pdk and Pusk can then be calculated respectively. 115 6.4.2 CP with FCR Similar to Section 6.4.1, each partition is now modeled as an M/M/C/S queuing system with non-homogeneous arrival rates. The state of this queuing system is defined as the sum of the number of class k calls in service, and its state transition diagram is shown in Fig. 6.4. If Chk is the number of reserved channels for class k traffic and Pj the probability of state j, it can be derived from this figure that the "rate-up = rate-down" state equations are Chk JMk JMk [hk_ p c - c c (6.40) Using the above equation recursively, along with the normalization condition 5^ = i, J=o (6.41) the probability distribution, Pj, is found as Clr -C),l Ck-Chk j_(Ck-Chk) 7=0 f-Mk • c>~c> b Kk Dk r . ? (Ak + xhk) /! II.  J j \<j< C - c ^ k ^ hk (Ck-Chk ) (C-C„) (Ak + A,hk) Ahk. b ^ k ^ h k C t c " + j < y- < c (6.42) (6.43) The probability of class k calls being blocked, Pbk, is the sum of the state probabilities when the queuing system is in the state j > (Ck-Chk)lbk, i.e., 116 Pbk= TPJ. (6.44) The handover failure probability, Pjk, can be expressed as (6.45) Taking again the factor 8k into account, PT Cklbk"k- (6.46) Pdk and Pusk can be calculated using Eqs. (6.22) and (6.23), respectively. In addition, for both cases (i.e., with and without FCR), a recursive algorithm similar to the one described in Section 6.3.1 is used to solve these equations. 6.5 Threshold Call Admission Policy It is clear that CS offers the best channel utilization, since it allows all channels to be shared by all incoming calls. As long as the number of available channels meets the bandwidth requirement of the incoming call, the call can be admitted into the system. However, this sharing policy could result in different classes of traffic being handled unfairly. With CS, broadband traffic has a very high blocking and handover failure probability, and narrowband traffic tends to monopolize the channel resource. By contrast, CP ensures QoS by assigning an exclusive set of channels to each class of traffic. However, channel utilization is poor; when no channels are available within the set of channels assigned to its class, an incoming call cannot make use of the idle channels belonging to another class. In addition, CP is very inflexible in dealing with 117 changing traffic conditions, which require channel re-partitioning to optimize system performance. Motivated by the above, a novel Threshold Call Admission policy (TCA) is presented that we will apply to a LEO-MSS to improve its system performance. The idea behind the T C A scheme is to limit the number of new calls of each class that can be admitted into LEO-MSS. Therefore, instead of dividing the total number of C channels into K subsets, in T C A each class is assigned a nominal capacity, Nk, as the threshold of class k traffic. However, in contrast to the CP policy, T C A does not require YJKk_lbkNk < C. A newly arriving class k call is admitted in state n i f and only i f nk+l<Nk and b»n + bk< C. To deal with the very frequent handovers and reduce their failures in LEO-MSS, handover requests are given higher priority by accepting them whenever there are sufficient channels in the destination cell without any threshold restrictions. The nominal capacity threshold, Nk, is derived from traffic forecasts in conjunction with QoS requirements for each class of traffic. If the arrival rate of a class is higher while the overall traffic is light, the T C A scheme acts like CS and can achieve a high multiplexing gain. On the other hand, when a class of traffic exceeds the threshold, T C A will act as CP, which isolates this class of traffic by limiting the maximum number of new calls accepted into the cell. Furthermore, without the restriction of bkNk < C, the values of Nk can be more flexibly adjusted. Thus, T C A provides a better and more flexible means of precisely allocating the available channel resources so that channel utilization can be optimized. Compared with CS, since T C A imposes more restrictions, it prevents one class of traffic from overwhelming other classes. Compared with CP, T C A offers more flexibility and can better handle the statistical fluctuation of traffic 118 with its multiplexing properties. In addition, priority is given to handover requests when limiting new call arrivals from each class of traffic. For all of these reasons, it is expected that improved handover performance will be achieved when T C A is employed. 6.6 Performance Evaluation Results and Discussion In this section, we summarize and discuss the various analytical and computer simulation results we obtained using the analysis and methodology presented in Sections 6.3 - 6.5. Performance evaluation results for a LEO-MSS with two incoming traffic, i.e., single-channel and 2-channel traffic, were obtained by means of theoretical analysis and computer simulations. To maintain the necessary accuracy for the performance results obtained by computer simulations, we chose the following numerical values for the different system parameters: C = 10, b\ = \,b2 = 2, Td\ = 180 sees, Tdl = 540 sees and pi = 0.02/?i. The performance results illustrated in Figs. 6.5 - 6.8 verify the methodology proposed in Section 6.3. More specifically, Figs. 6.5 and 6.6 show good agreement between the analytical and simulation results for multi-class traffic under the CS policy. It can be also seen that the CS policy results in unfair access to the available channel resources. For example, although 2-channel traffic has much lower traffic intensity, its new call blocking and handover failure probabilities are rather high. Figs. 6.7 and 6.8 compare the analytical and simulation results for multi-class traffic under CS policy with FCR = Cm = 2). Again, it can be seen that the analytical results are in good agreement with the equivalent computer simulation results. 119 Although FCR effectively reduces Pj\ and Pd\, the unfairness of channel access for different classes of traffic remains unchanged. . In Figs. 6.9 and 6.10, the analytical results were obtained using the methods described in Section 6.4. In particular, we have analyzed and simulated the performance of LEO-MSS systems under the CP policy with and without FCR using the following parameters: C = 10, Cn = 2 and Tdl = 180 sec. From these figures, it can be seen again that the analytical results are very close to the equivalent computer simulation results. For .the CP channel resource allocation scheme, each traffic class has its own subset of channels, and mk = C]jbk should be an integer to make the most efficient use of the available channels. Therefore, the analysis and simulation results for other values of Ck and bk are equivalent to those obtained by using b\ = 1 and C\ = Cklbk. It is also observed that Pj\ and Pd\ are reduced considerably, at the expense of a minor increase in Pa\ and Pus\, which makes FCR attractive for use in conjunction with LEO-MSS. Finally, a comment regarding the small differences in the obtained performances between the analytical and computer simulated results shown in Figs. 6.5 to 6.10 is in order. Although it is very difficult to theoretically justify these differences, we believed that they are a result of the assumptions of independent new and handover call arrivals and the approximation used for the channel holding time in a cell for both new arrivals and handovers (see Eq. (6.5)). As previously explained, we adopted these assumptions to ensure mathematical tractability. We now turn to the performances of CS, CP and T C A shown in Figs. 6.11 - 6.14. For the convenience of the presentation, TH«i«2 will denote the case N\ = n\ and N2 = «2, and CP«i«2 the case C\ = «i and C2 = «2- Figs. 6.11 and 6.12 compare various performance evaluation results for 120 CS and T C A schemes. Fig. 6.11 clearly shows that T C A significantly reduces probability of handover failure. Fig. 6.12 shows that the CS policy results in unfair access to the available channel resources. Although 2-channel traffic has much lower traffic intensity, its unsuccessful call probability is rather high; with TCA, this unfairness can be alleviated. For TH82, the difference between Pus\ and Pusj has been significantly reduced. For TH63, PUS2 becomes even lower than Pusj, which is highly desirable considering the much lower traffic intensity of 2-channel traffic. Figs. 6.13 and 6.14 compare the performance differences between T C A and CP. Compared with CP82, TH63 performs much better in terms of P/\, Pp and PUS2. Only Pus\ is higher than CP82, which is acceptable considering that 1-channel traffic has very high intensity. With light traffic, T C A performs much better than the equivalent CP schemes due to the multiplexing gain. When p\ < 5 erls, TH73 has similar Pus\ as CP82 but achieves better Pj\, Pfi and PUS2. When p\ < 4 erls, all four performance measures of TH82 are lower than those of CP82. While Pf\ of TH82 is significantly lower than that of CP82, the difference in Pfi between TH82 and CP82 is reduced as traffic intensity decreases. It should be noted that no CP64 performance curves are shown; although Pfi and PUS2 of CP64 are low, its very high Pj\ and P ^ i make this partitioning scheme impractical. These high values occur because the number of channels allocated to 2-channel traffic greatly exceeds requirements, while an insufficient number of channels are allocated to 1-channel traffic. Without the restriction X*LA^* TCA schemes have more combinations to fine-tune the threshold for each class of traffic and can adapt better to the needs of different traffic conditions, thus achieving better performance. In this case, considering 0 < pi < p\ and 121 denoting (C\, Ci) and (N\, Ni) as possible partitioning schemes for CP and TCA, respectively, we have (8, 2) and (6, 4) for CP whereas we have (8, 1), (8, 2), (8, 3), (7, 3), (7,2), (6, 2), (6, 3), (6, 4), etc. for TCA. 6.7 Conclusions In this chapter, we developed analytical methods to evaluate the performance of CS and CP channel resource partitioning policies for multi-class traffic in LEO-MSS. In both cases, we studied the performance with and without FCR. Our analytical results are in good agreement with results obtained by computer simulation, which verifies our system model and analytical methods. To improve system performance, an efficient T C A policy is proposed. Compared with CS, T C A prevents certain traffic monopolizing channel resources by setting proper thresholds for incoming traffic. Compared with CP, T C A achieves better efficiency of channel utilization because of its multiplexing gain, especially under light traffic conditions. Moreover, T C A offers more flexibility by adjusting the thresholds to cope with changing traffic conditions. It can also effectively reduce handover failures in LEO-MSS. 122 Re-enter * 1 Handover New Call C e n L e n gth L Generation New Call Generation Out C e l 1 Cell Boundary Vtrk = 26600km/h Figure 6.1 Rectangular cell system model for the considered LEO-MSS network. Figure 6.2 Markov state diagram for two incoming traffic with b\ = 1 and bi = 2. 123 Figure 6.3 State-transition diagram for CP without FCR in LEO-MSS. Figure 6.4 State-transition diagram for CP with FCR in LEO-MSS. — e — P b 1 , simulation * P ( 1 , simulation * P d 1 , simulation e ^us1' s ' m L J ' a t ' o n e- — P M , analysis * P ( 1 , analysis * P c f 1 , analysis —e— p u s i • a n a | y s i s 5.5 6 6.5 7 7.5 Traffic Intensity per cell, (erl, single-channel traffic) 8.5 Figure 6.5 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for l-channel traffic under CS policy. 5.5 6 6.5 7 7.5 Traffic Intensity per cell. (erl. single-channel traffic) Figure 6.6 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 2-channel traffic under CS policy. 125 5.5 6 8.5 7 7.5 Traffic Intensity per Cell, (erl, single-channel traffic) Figure 6.7 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CS with FCR. - P ( 2 , simulation - P ^ . simulation , simulation us2' — — P b 2 , analysis # p f 2 , analysis 4, P d 2 , analysis 0 P ^ , analysis 5.5 6 6.5 7 i Traffic Intensity per Cell, (erl, single-channel traffic) Figure 6.8 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 2-channel traffic under CS policy with FCR. 126 e P b 1 , simulation « P ( 1 . simulation P d l , simulation 9 P u s l , simulation e— P b 1 , analysis — — P t 1 , analysis 1 P d 1 , analysis — e — P r e a n a l y s i s 5.5 6 6.5 7 7.5 Traffic Intensity per cell, (erl; single-channel traffic) Figure 6.9 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CP policy. Traffic Intensity per cell, (ed, single-channel traffic) Figure 6.10 Analytical and simulation results for the various performance probabilities as a function of traffic intensity for 1-channel traffic under CP policy with FCR. 127 T 1 r _] L 5.5 6 6.5 7 7.5 Traffic intensity per cell (erl. single-channel traffic) Figure 6.11 Pfk as a function of traffic intensity for CS and TCA. 4.5 5.5 6 6.5 7 7.5 Traffic intensity per cell (erf. single-channel traffic) 8.5 Figure 6.12 PUSk as a function of traffic intensity for CS and TCA. 128 Figure 6.13 Pjk as a function of traffic intensity for CP and TCA. —a— — $ — P u s 2 . C P 8 2 — — P u s V ™ 2 — - * — P U S 1 . T H 6 3 » - — - - - f r — P u = r T H 7 3 P u S 2 .T H ? 3 5 B Traffic intensity per cell (erl, single-channel traffic) Figure 6.14 PUSk as a function of traffic intensity for CP and TCA. 129 6.8 References [1] C. E. Fossa, R. A. Raines, G . H . Gunsch and M . A . Temple, "An overview of the IRIDrUM Low Earth Orbit (LEO) Satellite System", Proc. of IEEE National Aero, and Electro. Conf, pp. 152-159, 1998. [2] E. Del Re, " A coordinated European effort for the definition of a satellite integrated environment for future mobile communications", IEEE Commun. Mag., pp. 98-104, 1996. [3] http://www.iridium.com [4] Z. Wang and R T. Mathiopoulos, "Analysis and performance evaluation of dynamic channel reservation techniques for LEO mobile satellite systems", Proc. of IEEE VTC 2001 Spring, vol. 4, pp. 2985-2989, 2001. [5] E. Del Re, R. Fantacci and G. Giambene, "An efficient technique for dynamically allocating channels in satellite cellular networks", Proc.of IEEE GLOBECOM'95, pp. 1624-1628, 1995. [6] E. Del Re, R. Fantacci and G. Giambene, "Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks", IEEE J. Select. Areas Commun., vol. 13, no. 2, pp. 397-405, 1995. [7] Z. Wang and R T. Mathiopoulos, "On the performance analysis of dynamic channel allocation with FIFO handover queuing in LEO-MSS", IEEE Trans. Commun., vol. 53, issue. 9, pp. 1443-1446, 2005. 130 [8] E. Del Re, R. Fantacci and G. Giambene, "Handover queuing strategies with dynamic and fixed channel allocation techniques in low earth orbit mobile satellite systems", IEEE Trans. Commun., vol. 47, no. 1, pp. 89-102, 1999. [9] E. Del Re, R. Fantacci and G. Giambene, "Different queuing policies for handover requests in low earth orbit mobile satellite systems", IEEE Trans. Veh. Technol, vol. 48, no. 2, pp. 448-458, 1999. [10] S. Karapantazis and F.-N. Pavlidou, "Design issues and QoS handover management for broadband LEO satellite systems", IEE Proceedings Commun., Vol. 152, Issue 6, pp. 1006 -1014,2005. [11] C. Tzeng, K . Ke and H . Wu, "Resource allocation and adaptive routing in multimedia low earth orbit satellite mobile networks", IEEEICME'04, Vol. 3, pp. 1795-1798, 2004. [12] B. S. Yeo and L . F. Turner, " A multi-class LEO satellite network", IEEE VTC 2001 Fall, Vol. 4, pp. 2202-2205,2001. [13] F. Huang, S. Wu, H. Xu , J. Liu and B. Xiao, "Probability Based Dynamic Channel Reservation Strategy for Reliable Handoff in Multimedia LEO Satellite Communications", IEEE International Symposium on MAPE, Vol. 2, pp. 1567 - 1570, 2005. [14] K W. Ross, Multiservice Loss Models for Broadband Telecommunication Networks, Springer-Verlag, 1995. [15] G. Maral, J. Restrepo, E. Del Re, R. Fantacci and G. Giambene, "Performance analysis for a guaranteed handover service in an LEO constellation with a 'satellite-fixed cell' system", IEEE Trans. Veh. Technol, vol. 47, no. 4, pp. 1200-1213, 1998. [16] B. Bjelajac, "Performance analysis of mobile satellite systems with dynamic channel assignment and handover resource reservation", European Personal Mobile Commun. Conf,pp. 601-606, 1997. 132 7 Conclusions and Topics for Future Research 7.1 Conclusions In this thesis, different channel resource management strategies for LEO-MSS have been investigated through a combination of analytical and computer simulation methods. The performance and limitations of known and newly proposed resource management strategies have been evaluated. The achievable QoS with different channel resource management strategies in supporting multi-class and multi-party traffic has also been studied. The proposed channel resource management strategies can be applied to the current and new LEO-MSS to improve the overall system performance. In addition, the theoretical methodology developed in this thesis improves the understanding of various resource management strategies for LEO-MSS and provides efficient and accurate evaluation of the system performance. The major contributions of the thesis are summarized as follows. 7.1.1 Analysis and Performance Evaluation of FCR and TDDCR Techniques We first proposed a mathematical analysis for the FCR scheme with FIFO-QH policy in LEO-MSS, the results of which are consistent with those obtained by means of computer simulation. In order to further improve the performance of FCR and exploit the high-speed deterministic movement of LEO satellites, we proposed an efficient TDDCR scheme based on the analytical and simulation results of FCR. The proposed scheme outperforms the traditional FCR scheme while only marginally increasing the computation complexity. We also studied the performance of the TDDCR in conjunction with other resource management schemes, such as 133 QH and FIFO-QN. In addition, we introduced a non-uniform traffic model to evaluate the performance of the proposed TDDCR scheme for LEO-MSS in more realistic traffic environments. The computer simulation results have shown that the TDDCR scheme can effectively limit the handover failure probability and handle various non-uniform traffic distributions over the coverage area of the LEO-MSS network. 7.1.2 Performance Analysis of DCA with FIFO-QH Compared with QH, FCR and DCR, the D C A technique combined with FIFO-QH achieves much better performance in terms of Pdrop while maintaining a very low Pbhck- hi order to better understand D C A in LEO-MSS, we have developed an approximate analytical method for evaluating the performance of D C A in conjunction with FIFO-QH in LEO-MSS. This method provides an excellent alternative approach for obtaining the performance of D C A with sufficient accuracy, thus avoiding the well-known problem of very time-consuming and computationally intensive computer simulations. 7.1.3 A Novel TDDCAR Technique Based on our study of D C A , we furthered our research to improve the handover performance of D C A techniques. In particular, we have proposed a novel and efficient T D D C A R technique that fully exploits the deterministic movement of the M T and the efficiency of the D C A technique, thus providing an effective way to reduce the cost of channel reservation in LEO-MSS. Performance evaluation results obtained by means of computer simulation demonstrated that, compared with other known channel resource management techniques (e.g., D C A , DCR, QH), the proposed T D D C A R technique has much better performance for LEO-MSS. The 134 T D D C A R technique can effectively reduce the handover dropping probability Pdrop and significantly improve overall system performance, especially under heavy traffic where the performance of other techniques decreases considerably. 7.1.4 Analysis and Performance Evaluation of Channel Resource Management Strategies for Multi-party Traffic We developed analytical methods in this thesis to evaluate the performance of multi-party traffic with FCR in LEO-MSS. Also, to improve the performance of FCR and exploit the high-speed deterministic movement property of LEO satellites, we have proposed an efficient A C R scheme. The proposed scheme outperforms the traditional FCR scheme while introducing little additional computational complexity. When used in conjunction with the NCQ technique, extremely low blocking and handover failure probabilities can be achieved for multi-party traffic in LEO-MSS. Performance results have shown that the proposed A C R and NCQ schemes are attractive channel resource management techniques for LEO-MSS in supporting multi-party traffic. 7.1.5 Analysis and Performance Evaluation of Channel Resource Management Strategies for Multi-class Traffic In order to support multimedia traffic in LEO-MSS, channel resource partitioning techniques need to be investigated. In our research, we have developed analytical methods to evaluate the performance of the well-known CS and CP channel resource partitioning schemes for multi-class traffic in LEO-MSS, both with and without FCR. Our analytical results are in good agreement with results obtained by computer simulation. To improve the performance of CS and CP, we 135 have introduced a more efficient T C A technique and evaluated and compared its performance with those ofCS and CP. 7.2 Topics for Future Research From the contributions of this thesis summarized in the previous section, it is clear that channel resource management strategies are complex issues for LEO-MSS, opening up a multitude of avenues for additional research. In the following sections, we suggest several topics for future research. 7.2.1 Optimization of Channel Reservation Mechanism As described in Chapter 2, the proposed TDDCR scheme can efficiently reduce the probability of handover failure. However, although this newly proposed TDDCR scheme makes reservations according to current traffic conditions and therefore partly reduces the cost of channel reservation, it cannot avoid the waste of channel resources caused by the reservation. For example, when some of the channels in the cell are reserved for handover requests, they will remain idle unless occupied by future handover calls. Therefore, in future research, from both theoretical analysis and computer simulation point of view, the possibility of obtaining a nearly optimal solution for channel reservation is worth investigating. Research efforts may focus on finding appropriate cost functions to analyze and decrease the cost of channel reservations. A more efficient channel reservation scheme, however, will most likely introduce complex control mechanisms, thereby increasing system requirements for computation and large data storage. Therefore, it is meaningful and interesting to investigate the tradeoff between control/computational complexity and performance improvement. 136 7.2.2 Efficient Channel Resource Management Techniques for Multi-class Traffic In Chapter 6, we studied the performance of multi-class traffic in LEO-MSS under CP and CS channel partitioning policies and introduced a T C A technique that is more suitable for LEO-MSS in supporting multi-class traffic to improve system performance. However, other channel resource management schemes (e.g., QH, DCA) should also be investigated to deal with the high frequency of handovers in LEO-MSS and further improve system performance. 7.2.3 Optimization of Channel Resource Management Strategies In practice, the ultimate goal is to maximize the revenue generated by LEO-MSS. Therefore, it is meaningful to investigate resource management strategies that maximize revenue under appropriate revenue functions. However, particularly for LEO-MSS supporting multimedia and multi-party applications, it may be quite complicated to develop appropriate revenue functions and determine the optimal channel resource management strategies. 7.2.4 System Utilization and Packet Level System Performance In this thesis, the simulation and analytical work has focused on the connection level. We have used four QoS parameters (i.e., new call blocking probability Pbhck, handover failure probability Pfan, call dropping probability P„rop and unsuccessful call probability Pns) to evaluate system performance. To present a more detailed performance evaluation, it is meaningful to further calculate system utilization and extend the research to include packet level analysis and simulation. 137 7.2.5 Resource Management Strategies for Integrated Satellite-Terrestrial Mobile Telecommunication Networks The deployment of broadband mobile satellite communication systems and their integration with the terrestrial cellular networks will pave the way for the global roaming envisaged in future mobile telecommunication systems. The UMTS is an integrated network consisting of. all LEO/MEO/GEO satellite systems as well as terrestrial UMTS systems. The terrestrial microcells in densely populated areas and macrocells in rural areas, together with the spot beams of LEO/MEO and GEO MSS, will be overlap and create a multi-layered hierarchical cellular system. Efficient and flexible resource management strategies are of great importance in such a mobile network environment in order to increase network capacity and reduce service costs. 7.2.6 Resource Management over Fading and Shadowing Channels The research results we have obtained so far are all under the assumption that signals are not impaired by interference and/or fading, which is not the case in practice. Therefore, another interesting topic for future research is to include additional critical system parameters in the LEO-MSS communication environment, such as elevation angle, interference and fading, etc. Adopting a more realistic channel model in conjunction with appropriate network configurations wil l allow for studying the channel statistical properties, including channel availability. The performance of queuing schemes, channel allocation schemes, and channel reservation schemes may be evaluated subsequently. Furthermore, more efficient channel resource management strategies appropriate for the above system configuration may be proposed and their performance may be studied. 138 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0064902/manifest

Comment

Related Items