Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Connection architecture and protocols to support user terminal mobiltity over an ATM/B-ISDN personal… Yu, Oliver T.W. 1997

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

Item Metadata

Download

Media
831-ubc_1997-19681X.pdf [ 5.4MB ]
Metadata
JSON: 831-1.0065156.json
JSON-LD: 831-1.0065156-ld.json
RDF/XML (Pretty): 831-1.0065156-rdf.xml
RDF/JSON: 831-1.0065156-rdf.json
Turtle: 831-1.0065156-turtle.txt
N-Triples: 831-1.0065156-rdf-ntriples.txt
Original Record: 831-1.0065156-source.json
Full Text
831-1.0065156-fulltext.txt
Citation
831-1.0065156.ris

Full Text

Connection Architecture and Protocols to Support User Terminal Mobility over an ATM/B-ISDN Personal Communications Network by  OLIVER T.W. Y U  B.A.Sc,  The University of British Columbia,  1981  M.A.Sc,  The University of British Columbia,  1991  A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF T H E REQUIREMENTS FOR T H E D E G R E E OF DOCTOR OF PHILOSOPHY in T H E FACULTY OF GRADUATE STUDIES DEPARTMENT OF E L E C T R I C A L ENGINEERING  We accept this thesis as conforming to the required standard  The UNIVERSITY OF BRITISH COLUMBIA ©Oliver T.W. Yu, March 1997  In  presenting  degree  at the  this  thesis  in  University of  partial  fulfilment  of  of  department  this thesis for or  by  his  or  requirements  British Columbia, I agree that the  freely available for reference and study. I further copying  the  representatives.  an advanced  Library shall make it  agree that permission for extensive  scholarly purposes may be her  for  It  is  granted  by the  understood  that  head of copying  my or  publication of this thesis for financial gain shall not be allowed without my written permission.  Department The University of British Columbia Vancouver, Canada  DE-6 (2/88)  Abstract Future personal communications networks (PCNs) will support personal and terminal mobility based on the B-ISDN platform, and will likely employ ATM-based backbone networks to interconnect wireless mobile networks. This thesis addresses the problem of providing robust, fast and seamless network-wide handoffs over an ATM-based P C N to support non-restrictive terminal mobility and to maintain connection-oriented performances during calls in order to enable handoff transparency to users. A novel "mobile virtual circuit (MVC) connection architecture" is proposed in the A T M layer to facilitate fast handoffs and to allow mobile and non-mobile traffic to share connection resources. The proposed M V C maintains the current and the potential handoff connections terminated at the neighbouring base stations as a dynamic connection tree, which converges at a root node in the ATM-based backbone to enable connection reuse during handoff. To minimize resource overhead and to enable fast handoff, the proposed M V C predetermines routes for potential handoff connections during call setup and after each successive handoff, and employs the proposed robust and fast scheme of generic bandwidth reservation along the predetermined route to complete the establishment during handoff. A novel "general real-time connection rerouting service" in ATM/B-ISDN is proposed with capabilities that can be configured to support robust, fast and seamless call rerouting over fixed connections in general and to support call handoffs over ATM-based PCN in particular. This service includes: (1) a novel packet ordering synchronization transaction protocol (POSTP) to prevent call disruptions due to transient data loss and sequencing errors during the traffic rerouting phase; (2) a novel robust and fast resource reservation transaction protocol (RFRTP) to enable fast connection establishment over a predetermined route with optional localized rerouting over congested or failed links. To support general real-time connection control services (e.g., the proposed rerouting ii  service), extensions to the SS7-based signaling network are proposed with source-routing associated-CCS and inband signaling to provide fast distributed signaling transport and synchronization data transport respectively. To facilitate robust handoffs or to maintain the probability of forced call termination due to handoff blocking under non-stationary call arrival condition, a novel "dynamic guard channel resource management architecture" is proposed to adapt the number of guard channels in each cell according to the current estimate of the handoff arrival rate derived from the current number of ongoing calls in neighbouring cells and the mobility pattern, so as to keep the handoff blocking probability close to the targeted objective while constraining the new call blocking probability to be below a given level.  The  proposed scheme is applicable to channel allocation over cellular mobile networks, and is extended to bandwidth allocation over the backbone network.  It is demonstrated  that fast, robust and seamless network-wide handoffs over an ATM-based P C N can be achieved by integrating the proposed service and architectures. Handoff performance is analyzed in terms of handoff processing delay and forced call termination probability.  iii  Contents Abstract  ii  List of Tables  viii  List of Figures  ix  Acknowledgment  xii  Chapter 1  Introduction  1  Section 1.1  Motivations  1  Section 1.2  Objectives  4  Section 1.3  Background and Approach  5  1.3.1  Connection Architecture  5  1.3.2  Connection Rerouting Service  7  1.3.3  Connection Resource Management Architecture  12  Organization of Thesis  15  Section 1.4 Chapter 2  ATM Connection Architecture to Support Fast and Efficient Handoffs  16  Section 2.1  M V C Connection Configuration  18  Section 2.2  M V C Connection Establishment  21  Section 2.3  M V C Rerouting During Handoff  23  Section 2.4  M V C Reconfiguration After Handoff  25  Section 2.5  Effect of TP Availability On Connection Reuse  28  iv  Chapter 3  Signaling Network Architecture and Transaction Protocols to Support Real-time Connection Rerouting in  Section 3.1 3.1.1  ATM/B-ISDNs  31  Signaling Transaction Protocols  37  Robust Fast Reservation Transaction Protocol (RFRTP) . 3 7  3.1.1.1  Fast Reservation Transaction  40  3.1.1.2  Robust Localized Retry Transaction  45  3.1.1.3  Error Recovery  46  3.1.2  Packet-Ordering Synchronization Transaction Protocols (POSTPs)  47  3.1.2.1  Intra-network P O S T P  48  3.1.2.2  Inter-network P O S T P  49  Section 3.2  Signalling Transports  51  3.2.1  Associated-CCS Transport  51  3.2.2  Inband Signaling Transport  54  Performance Analysis  56  3.3.1  Reservation Transaction Delay with R F R T P  56  3.3.2  Reservation Transaction Robustness with R F R T P  59  Section 3.3  Chapter 4  Fast, Robust and Seamless Handoffs over ATM-Based PCNs  62  Example Scheme  62  4.1.1  Fast Handoff Connection Establishment  65  4.1.2  Seamless Synchronized Traffic Rerouting  66  Section 4.1  Section 4.2 4.2.1 4.2.1.1 4.2.2  Performance Analysis  68  Processing Delays  68  Factors Affecting Handoff Processing Delay  70  Effect of Handoff Processing Delay on Unnecessary Handoff Probability  Chapter 5  73  Adaptive Connection Resource Management Architecture to Provision Call Level Performances over an ATM-based PCN  Section 5.1  78  Dynamic Guard Channel Scheme in Cellular Mobile Networks  79  5.1.1  Instantaneous Handoff Call Arrival Rate Estimation . . . . 82  5.1.2  Capacity Limit or Guard Bandwidth Adaptation  83  5.1.3  Performance Analysis  86  5.1.3.1  Call Blocking Probabilities Under Stationary Traffic  ....87  5.1.3.2  Call Blocking Probabilities Under Non-stationary Traffic . . 92  5.1.3.3  Forced Call Termination and Overall Call Failure Probabilities  Section 5.2  95  Bandwidth Allocation over ATM-Based Backbone Network  98  Dynamic Guard Channel Extension  98  5.2.1.1  Extension via Mobile Virtual Circuit (MVC)  99  5.2.1.2  Dynamic Guard Bandwidth Management Functions with  5.2.1  MVC  100 vi  5.2.2  Performance Analysis  103  5.2.2.1  Reservation Failure Probabilities  103  5.2.2.2  Forced Call Termination Probability  104  Conclusions  108  Possible Future Works  110  Chapter 6 Section 6.1 Bibliography Appendix A  112 List of Acronyms and Abbreviations  vii  117  List of Tables Table 1  Performance Requirements for Real-Time Connection Rerouting  35  Table 2  R F R T P Signaling Messages  38  Table 3  ATM Cell Header Fields Encoding and ATM Switch  Table 4  Functions  53  Probability of Unnecessary Handoff  76  viii  List of Figures Figure 1  Extensions to ATM/B-ISDN Protocol Model to Support Network-Wide Seamless Handoffs  Figure 2  10  Location Independent Transport over Backbone ATM Network  17  Figure 3"  Connection Tree Configurations  18  Figure 4  M V C Connection Architecture to Support Mobile Communications  20  Figure 5  M V C Mobile Links Rerouting During Handoff  24  Figure 6  M V C Reconfiguration After Handoff  27  Figure 7  Effect of TP Availability on Connection Reuse  30  Figure 8  Connection Dynamics due to Real-Time Operational Requirements on Connection Rerouting  Figure 9  33  Proposed Signalling Network Architecture and Transaction Protocols to Support Real-Time Connection Rerouting . . 36  Figure 10  Localized Rerouting  40  Figure 11  Example of R F R T P Operations with Fast Reservation Transaction  Figure 12  42  Example of R F R T P Operations with Fast Reservation and Robust Retry Transactions  43  Figure 13  Operational Environment for Inter-Network P O S T P  Figure 14  Associated-CCS Transport: Source-Routing Unicast and  Figure 15  . . . . 48  Multicast  52  Signaling Cells Format  52  ix  Figure 16  ATM Switch Functions  55  Figure 17  R F R T P Performance Analysis: Reception Cycles  59  Figure 18  R F R T P Performance Analysis: Reservation Robustness .61  Figure 19  Example of Fast Seamless Handoff Processing with M V C Connection  Figure 20  64  Effect on Handoff Processing Delay due to M V C Connection Configuration  Figure 21  Effect on Handoff Processing Delay due to Targeted per-link Handoff Call Blocking Probability  Figure 22  71  72  Effect on Handoff Processing Delay due to Reservation over Alternative Routes  73  Figure 23  Handoff Initiation Margin  74  Figure 24  Dynamic Guard Channel Scheme  81  Figure 25  Estimation of Instantaneous Handoff Call Arrival Rate at a BS  Figure 26  84  Number of Guard Channels Required for Targeted Blocking Probabilities  85  Figure 27  Simulation Model  86  Figure 28  Effect of Handoff Probability on New Call Blocking Probability  Figure 29  88  Effect of Handoff Probability on Handoff Call Blocking Probability  Figure 30  89  Effect of New Call Arrival Rate on. New Call Blocking Probability  89 X  Figure 31  Effect of New Call Arrival Rate on Handoff Call Blocking Probability  90  Figure 32  Adaptive Gain  91  Figure 33  Effect of Channel Loading Factor on Blocking Probabilities  91  Figure 34  Per Connection Blocking Probabilities  93  Figure 35  An example of Non-Stationary Call Traffic Condition . . . . 94  Figure 36  Effect of Non-Stationary Condition on Call Blocking Probabilities  Figure 37  '.  95  Forced Call Termination and Overall Call Failure Probabilities  Figure 38  97  Estimation of Instantaneous Handoff Call Arrival Rate at a Wired Link with M V C  Figure 39  101  Effect on Forced Call Termination Probability due to M V C Connection Configuration  Figure 40  Effect on Forced Call Termination Probability due to Targeted per-link Handoff Call Blocking Probability  Figure 41  105  ...  106  Effect on Forced Call Termination Probability due to Reservation over Alternative Routes  xi  107  Acknowledgment I would like to express my sincere gratitude to my research supervisor, Dr. V . C . M . Leung for his valuable suggestions, critiques and proof readings.  Part of this research  work was supported by a grant from the Canadian Institute for Telecommunications Research under the N C E Program of the Government of Canada.  xii  Chapter 1 Introduction 1.1 Motivations Personal communication services (PCS) support ubiquitous personal and terminal mobility. The broadband integrated services digital network (B-ISDN) employing asynchronous transfer mode (ATM)  [1, -2, 3] is a telecommunication network architecture that  provides broadband and multimedia communication services. A s air interface technologies are being developed to support broadband wireless access [4, 5, 6], it is expected that the next generation personal communication network (PCN) will include the capabilities of B - I S D N and call upon an ATM-based backbone to interconnect broadband wireless networks [7, 8, 9]. This motivates us to examine the issues and problems of extending PCS  capability to the A T M / B - I S D N to support user terminal mobility during calls, i.e.,  call handoff management [10], and of provisioning connection resources management over an ATM-based P C N . The  architecture of a first generation P C N consists of separate narrowband mobile  access networks, generally configured as regional hierarchical structures with mobile terminals (MTs) in radio cells accessing base stations (BSs) connected to mobile switching centers (MSCs), which in turn interconnect with the backbone public switched network (PSN)  via gateway M S C s [11, 12, 13].  Handoff management is required when M T s  cross cell boundaries while engaged in calls, and involves reassignment of radio channels and  possibly rerouting of network connections.  First generation P C N s restrict handoff  management to cell areas associated with each individual mobile access network.  As  P C N s evolve to accommodate network-wide terminal mobility, the call handoff process is required to support non-restrictive cell transition of an active M T . Furthermore, if the  1  adjacent cells are connected to different parts of the fixed backbone network, then the handoff process is required to operate over both the mobile wireless and fixed backbone networks. In [10], a cell transition of a M T is classified as either intra-cell-cluster or intercell-cluster, and the procedure for intra-cell-cluster handoff is much more simplified and efficient. It is preferable to have one uniform call handoff process that can be applicable to every cell transition of any active M T because an heterogeneous call handoff process may cause large variance in handoff performance and implementation complexity.  Connection rerouting to support mobile call handoff can cause connection-oriented service performance lapses resulting in transient data loss, sequencing errors and delay jitter at the connection level, and forced call termination at the call level. Data loss can occur due to momentary unavailability of the old or new connection, or premature traffic rerouting from the sender node via the new connection before the receiver node is ready. Mismatch of data-unit transit times between the old and the new connections during the handoff instance could result in transient data-unit delay jitter or disordering (sequencing error). For synchronous traffic, if the inaugural data-unit of the new connection arrives later than the expected time at the destination, transient data-unit delay jitter occurs due to the traffic lagging-gap.  For both synchronous and asynchronous traffic, if the  inaugural data-unit of the new connection overtakes data-units of the old connection at the destination, transient data-unit disordering occurs. Forced call termination occurs when the handoff connection cannot be established due to network resources unavailability, signaling failure or the lack of mobility support.  A s P C N s evolve to accommodate broadband and multimedia traffic transported over an ATM-based B - I S D N backbone (with uniform size of data-unit or A T M cell), applications will have diverse quality of service (QOS) requirements which translate into 2  network performance requirements in the A T M network. A T M bearer or connectionoriented service classes are defined to provide different levels of network performances. Consequently, the levels of tolerance to network performance lapses due to mobile call handoff depend on the application and bearer service classes. With broadband traffic, even very short duration of cell loss could result in significant and unacceptable loss of user information for most data applications. Cell disordering would be unacceptable for reliable data applications. Excessive cell delay jitter may not be acceptable for real-time data applications such as voice and it could destroy temporal synchronization among multimedia traffic components. Forced call termination would be unacceptable for most applications. Current research on network-wide handoff over an ATM-based P C N has been limited to hard or disruptive handoff scheme with connection rerouting function results in call disruptions due to cell loss and sequencing errors. To enable handoff transparency to the users, connection-oriented service performance lapses due to connection rerouting must be minimized or eliminated if possible. To eliminate data loss and sequencing errors, a traffic rerouting synchronization service needs to be developed. The probability of forced call termination due to resource unavailability during handoffs cannot be reduced to zero unless each connection resource is overallocated in anticipation of handoffs during call setup, which would drastically reduce resource utilization and is not practically applicable. Alternatively, forced call termination probability can be minimized by employing congestion recovery services (e.g., request retried or queued). The rerouting synchronization and congestion recovery services would incur delay overhead in handoff processing. Handoff performance requirements to enable handoff transparency to the user are as follows: (1) fast handoff to minimize the delay jitter and the probability of data loss due 3  to the loss of radio contact before handoff completes; (2) seamless handoff to eliminate data loss and sequencing errors due to the lack of traffic rerouting synchronization control; (3) robust handoff to minimize forced call termination probability.  1.2 Objectives The objective of this thesis is to maintain mobile connection with comparable fixed connection service performances when subjected to network-wide handoffs over an ATMbased P C N , which would result in connection rerouting over both the wireless mobile and the ATM-based backbone networks.  A novel integrated platform of connection  architecture, connection rerouting service, and resource management architecture is proposed to minimize or eliminate performance lapses due to connection rerouting during handoffs. The proposed platform is intended to support application services that have the most stringent requirements on connection performances and are most sensitive to performance lapses due to handoff rerouting. The proposed platform enables fast handoffs by minimizing delay jitter, seamless handoffs by eliminating data loss and sequencing errors, and robust handoffs by minimizing forced call termination probability. The connection architecture to facilitate fast handoff is developed with the provision that connection resources are shared between mobile and non-mobile traffic to optimize resource utilization, and that it is applicable for any cell transition of an active M T . The connection rerouting service to support robust, fast and seamless handoff needs to be generally applicable for MTs with or without the capability for multi-channel access during handoffs. The resource management architecture to facilitate robust handoffs is developed with the provision to enable resource utilization optimization under networkwide handoffs. Furthermore, it aims to prioritize resource access for handoffs over new calls with shared resource capacity for mobile traffic, and the priorities are adaptive to non4  stationary call arrivals with time-varying rates. The resource management architecture is intended to be applicable in both the wireless and wired mediums by extending wireless resource management scheme into the ATM-based backbone network. In the wireless medium, it can be employed by T D M A and F D M A air interfaces with fixed channels or C D M A air interface with the number of users limited by the received signal requirements. The integrated platform makes no assumption on the air interfaces. Call disruptions associated with hard handoffs over F D M A air interfaces may result in data loss. T D M A air interfaces may enable seamless handoffs by switching channels and time slots within a T D M A frame, while soft handoffs over C D M A air interfaces which coherently combine simultaneous transmissions over the new and the old cells are inherently seamless. In both cases, it would be desirable for the mobile bearer services over an ATM/B-ISDN to employ the integrated platform to preserve packet sequencing between the old and the new A T M connections, and to enable fast and robust handoffs over the A T M connections.  1.3 Background and Approach 1.3.1  Connection Architecture  A connection tree architecture linking BSs to a common root node could simplify and speed up handoff management by restricting handoff rerouting to the links between the M T and the root node, and by identifying these links as a group, eliminating routing table updates in nodes along the new connection. This is the basis of the permanently assigned connection tree architecture [10] which connects all BSs in a cell cluster to the root node using permanent virtual channels (PVCs) in the A T M layer. Within the geographical area or cell cluster covered by the BSs, MTs contend for the connections and intracell-cluster handoff is provided to support terminal mobility within a cell cluster. Intercell-cluster handoff to support terminal mobility over different cell clusters has not been  considered. Consequently, this connection architecture potentially limits mobility support to within a cell cluster and causes network resources to be partitioned between mobile and non-mobile traffic.  In contrast, we propose a new mobile virtual circuit (MVC)  connection architecture [14] consisting of dynamically configured per-call connection trees. Implemented as an extension to the A T M layer and illustrated in Fig. 1, each M V C connects the BS in the current cell serving an M T and the BSs in all the immediate neighboring cells to which the M T may potentially handoff its call, to the root node herein called the tethered-point (TP). After a handoff, the TP may shift to another node as the M V C is reconfigured. The M V C has all the desirable properties of the connection tree discussed above, while providing network-wide handoff management support across cells or mobile access networks connected to any point of the backbone network, and allowing complete sharing of network resources between mobile and non-mobile traffic.  Recent researches have dictated that handoff connection establishment, consisting of route determination followed by connection resource allocation, be executed in whole before or after connection rerouting initiation. In [10], a connection tree with connections terminated at the based stations of a cell cluster and at a root node in the ATM-based network are pre-established permanently to support handoffs within the cell cluster. Establishing potential handoff connections before handoffs removes the real-time requirement for handoff connection establishment during handoffs and minimizes the handoff processing delay at the expense of allocating connection resources which may not be used. In [15], handoff connection establishment during handoffs economizes on connection resources at the expense of possibly lengthy handoff processing delay, and distributed route determination schemes to minimize the call routing delay during handoffs are proposed. This thesis proposes a novel time-distributed handoff connection establishment scheme 6  which we have also presented in [16] and applied in [14, 17, 18]. The proposed scheme dictates that the routes for potential handoff connections are determined at call setup or after each call handoff (i.e. before connection rerouting initiation) but resource allocation is deferred until connection rerouting is actually initiated. This scheme balances handoff processing delay with efficient utilization of connection resources. It removes real-time requirement for route determination but dictates real-time requirement for resource allocation. To meet the real-time requirement for handoff connection resource reservation, this thesis newly develops a robust and fast resource reservation service, which we have also presented in [19, 20] and applied in [14, 17, 18].  1.3.2 Connection Rerouting Service Connection rerouting service to support call rerouting (e.g., mobile call handoff) consists of new connection establishment followed by traffic rerouting from the old to the new connection, and old connection release, supported by corresponding signaling protocols. Real-time and functional requirements for connection rerouting to minimize traffic level performance lapses are determined by the supporting connection dynamics, which is specified by the temporal relationship among the three phases of connection rerouting. The time interval from connection rerouting initiation to old connection takedown determines the upper bound of allowable completion time for connection rerouting to avoid call disruption or forced call termination. The nature of traffic level performance lapses is determined by the temporal relationship between old connection take-down and new connection establishment. A mobile call handoff is initiated when the received signal between the M T and the current BS falls below the receiver threshold level when a M T is transiting from a cell into a neighbouring cell. Connection rerouting to support mobile call handoff 7  involves rerouting traffic from the old connection terminated at the current BS to the new or handoff connection terminated at one of the neighbouring BSs. For mobile call handoff, since the air link portion of the old connection may be taken down at any time after connection rerouting initiation (this probability increases as time progresses) due to increasing signal attenuation and fading in the mobile cellular environment, fast or realtime connection rerouting becomes a necessary requirement to support seamless mobile call handoff, as well as to minimize transient delay jitter. Real-time requirement on connection rerouting translates into real-time requirement on new or handoff connection establishment which consists of route determination followed by connection resource allocation and routing establishment functions. Besides being fast to satisfy the real-time requirement, connection resource reservation must also be robust to avoid reservation failure due to congested or failed links. Researches on connection establishment in high speed networks have been limited to fast connection establishment schemes [21, 22]. This thesis proposes a robust and fast resource reservation service which we have also presented in [19, 20] and applied in [14, 17, 18] to support real-time handoff connection establishment. The problem of preventing call disruptions due to A T M cell loss and sequencing errors during traffic rerouting over an ATM-based P C N has received little attention in the literature. This thesis proposes a rerouting synchronization service, which we have also presented in [19, 20] and applied in [14, 17, 18] to support seamless or disruption-free handoffs. The basic synchronization protocol of the proposed service is applicable when the M T can maintain simultaneous access to the old and the handoff connections. The extended synchronization protocol of the proposed service dictates additional synchronization and buffering capabilities at the BS when the M T cannot 8  maintain simultaneous access to the old and the handoff connections. The novel disruption-free real-time connection rerouting service, presented in this thesis, consisting of the proposed fast robust resource reservation and traffic rerouting synchronization services, is realized by extensions to the B - I S D N signaling protocols and signaling network architecture. The extensions are backward compatible to current B I S D N standards and implementations, and are formulated with the following constraints: (1) to minimize the performance impact on existing connection control services supported in the signaling network; (2) to minimize the connection resource overhead in supporting handoff. The proposed A T M / B - I S D N extensions are illustrated in Fig. 1. Handoff signaling can be either implicit via stimulus monitoring or explicit via message passing.  In [10], the root switch of the virtual connection tree monitors the  cell transition of a M T by the corresponding change in the virtual channel and port identifiers of the incoming traffic from the M T . After a M T cell transition is detected, traffic directed to the M T would be rerouted to the handoff connection.  During the  interval from the M T cell transition to traffic rerouting completion, traffic directed to the M T via the old connection would be lost as the M T has already dropped the old connection and transmitted traffic via the handoff connection.  While stimulus based  signaling schemes avoids the overhead for transporting signaling messages, they would not be able to meet the real-time interactive signaling requirement. This thesis employs message based signaling protocols to implement real-time connection rerouting to support seamless handoff.  Message based signaling protocol can employ either centralized or distributed servicelogic paradigm. The B - I S D N implementation parallels the development of the advanced intelligent network (AIN) [23] which employs a SS7-based common channel signaling  9  Management Plane Control (Signaling) Plane Higher Layers Centralized Service Logic  User Plane Higher Layers  Higher Layers Distributed Service Logic Real-Time Connection Rerouttn  Route Determination I  POSTPs  RFRTP Robust Fmrt Reservation Tr.>ii£<i6tlolt Protocol  Packet-Ordering Synchiontzatlon Transaction Protocol  Signalling ATM Adaptation Layer  Assoctetedccs Intelligent Network  SourceRouting U nicest & WJutllcasst  Adaptation Layer  inband Signaling  c ATM Layer MVC (Mobile Virtual Circuit)  at  mm n ma 3  V C (Virtual Channel)  V P (Virtual Path)  Q  I a  Physical Layer  Figure 1 Extensions to ATM/B-ISDN Protocol Model to Support Network-Wide Seamless Handoffs  (CCS) network [24] and a centralized service-logic paradigm. The SS7 protocol model defines a transaction capabilities application part (TCAP) layer in which transaction protocols enable remote operations employing this service-logic paradigm. This paradigm is inappropriate for real-time connection rerouting services because of the potential bottlenecks at centralized service-logic providers. This thesis employs a distributed service10  logic paradigm to support the proposed real-time connecting rerouting service. The proposed fast resource reservation service is implemented by the robust fast reservation transaction protocol (RFRTP), and the proposed traffic rerouting synchronization service is implemented by the packet-ordering synchronization transaction protocols (POSTPs). These distributed signaling protocols in the T C A P layer dictate handshaking procedures among the M T , the BSs and the A T M switches of the old and the handoff connections, and the A T M switch where the connections converged. The orientation of these distributed signaling protocols within the B-ISDN protocol model is illustrated in Fig. 1. The real-time connection rerouting service logic is assumed to be implemented in each A T M switch. We have also presented these proposed signaling protocols in [20].  Signaling transport can be classified as either CCS or inband signaling. The SS7based signaling network can supports associated, non-associated and quasi-associated CCS signaling modes.  Signaling transport to enable the proposed reservation and  synchronization transaction protocols requires fast unicast and multicast associated-CCS transport services and an inband signaling transport service, respectively, realized by new signaling transport protocols in the signaling A T M adaptation layer (SAAL), as illustrated in Fig. 1. For the different CCS signaling modes, current implementations of SS7 provides either a connectionless or connection-oriented network service at the signal connection control part (SCCP) and message transfer part (MTP) which maps into the network layer of the OSI model. While connectionless services are deficient in reliable real-time transfer capability, connection-oriented services require prior explicit connection establishment and do not support efficient multicast transport. The proposed associated-CCS sourcerouting unicast and multicast protocols overcome these limitations. The proposed inband signaling protocol is employed to transport packet-ordering synchronization markers 11  during connection rerouting to preserve the timing relationship between the user data and the synchronization markers. We have also presented the proposed extensions to the SS7-based C C S network in [20].  1.3.3 Connection Resource Management Architecture To support network-wide handoffs over an ATM-based P C N , new and handoff call requests will compete for connection resources in both the mobile and the ATM-based backbone networks. Call level performance parameters due to resource congestion will include new and handoff call blocking probabilities.  Handoff calls require a lower  blocking probability, relative to new calls because forced terminations of ongoing calls due to handoff call blocking are generally more objectionable than new call blocking from the subscriber's perspective. Connection resource sharing policy for the two call classes ranges from complete partitioning which is more appropriate when the arrival rates for all classes are large, to complete sharing which is appropriate when the arrival rates for all classes are small. When the resource capacity is completely partitioned, differential resource congestion related performances or blocking probabilities can be provided simply by differential partition size.  In [25], given that the bandwidth resource is completely partitioned  between new calls and handoff calls, it examines different traffic class based sharing policies for realtime and non-realtime traffic sharing the bandwidth resource capacity partitioned for handoff calls. When the resource capacity for a given traffic class is shared among the new and handoff calls for better resource utilization, differential blocking probabilities can be supported by a resource access priority scheme [26].  The access priority can be  established via differential treatments of new and handoff calls by one or more of the 12  following methods: resource capacity limit, congestion control discipline, and admission criterion [27]. In the mobile networks, one common bandwidth resource access priority scheme is the guard channel (or bandwidth) scheme [28,  29, 30, 31] which gives  a higher access priority to handoff calls by assigning them a higher capacity limit. Resource access priority can also be established via differential congestion controls. For example, servicing handoff calls and new calls with blocked-calls-queued and blockedcalls-dropped disciplines, respectively, would further enhance access priority for handoff calls, or explicit queueing priority can be established between different call classes [31].  These schemes are developed under the assumption of stationary call arrivals or loading condition and may not be able to adapt to non-stationary loading while meeting the congestion related performance objectives. Future P C N s will employ microcells and picocells to support a higher capacity [32], thus increasing the frequency of call handoffs and dramatizing the effects of non-stationary traffic conditions due to fluctuations in new call arrivals and mobility pattern, and the achieved handoff call blocking probability may deviate significantly from the targeted objective. To minimize subscriber dissatisfaction, the requirement to keep the handoff call blocking probability at a targeted objective in the mobile and backbone networks under non-stationary loading condition is becoming even more important in future PCNs.  In [27], a new call admission control takes into  account of the projected future handoff call blocking probabilities in the originating and the neighbouring cells, which are kept to the target objective by blocking new calls even if capacity is currently available to service these new calls.  Since realtime congestion recovery during handoff call request can still result in forced call termination due to excessive recovery delay, non-realtime preventive congestion control could be provisioned to minimize the invocation of realtime congestion 13  recovery. This thesis proposes a novel dynamic guard bandwidth or channel scheme which adapts the guard bandwidth in each cell according to the number and mobility patterns of current active MTs in the neighbouring cells, so as to keep the current handoff call blocking probability within the targeted objective. The orientation of this scheme within the B-ISDN protocol model is illustrated in Fig. 1. We have also presented this proposal in [33, 34]. In the dynamic guard bandwidth scheme, a change in channel assignment status in any cell will trigger neighbouring cells to perform preventive congestion controls for handoff calls by increasing or decreasing the capacity limits for handoff calls. By changing the parameters of the adaptation control policy, the targeted objective can be maintained strictly, or loosely by ensuring that new calls are not denied resource access completely. The problem of maintaining differential congestion-related performances associated with setting up new and handoff calls in the backbone network has received little attention in the literature. Under non-stationary loading condition, this problem is compounded by the need to estimate instantaneous handoff call arrival rate for adaptation purpose. While handoffs into a radio cell can only originate from the neighbouring cells, handoffs to a wired link in the backbone network are dependent on network topology and call routing, estimation of instantaneous rerouting call arrival rate for each network link becomes a difficult proposition.  The problem is eased considerably in ATM-based PCNs by  employing the proposed M V C connection tree architecture to support handoff processing. The M V C of a M T links the base stations (BSs) of the current and neighboring (potential handoff) cells to a common root node in the network. Consequently, the number of MTs that can initiate handoff calls to a network link is determined by the number of potential handoff connections of the M V C s passing through it.  14  1.4 Organization of Thesis The thesis is organized as follows. Chapter 2 presents the M V C connection architecture in the A T M layer to support mobile call handoff, in terms of connection configuration, connection establishment, connection rerouting during handoffs and connection reconfiguration after handoffs. Chapter 3 presents the generic realtime connection rerouting service in A T M / B - I S D N with fast, robust and seamless capabilities that can be dimensioned to support call rerouting in general and mobile call handoff in particular, and the signaling network architecture with associated-CCS (source-routing unicast and multicast) and inband signaling in the A T M adaptation layer to support real-time connection rerouting. Chapter 4 presents an example scheme of implementing fast, robust and seamless handoffs over ATM-based P C N s by configuring the proposed generic realtime rerouting service to support mobile call handoff over the proposed M V C connection architecture. Chapter 5 presents the dynamic guard channel scheme and its extension to bandwidth allocation over the A T M backbone network. Chapter  15  6concludes  the thesis.  Chapter 2 ATM Connection Architecture to Support Fast and Efficient Handoffs The neighbouring cells of a given cell determine the potential handoff connections in the backbone network, associated with each ongoing call in this cell. For example, there are six potential handoff connections for a mobile terminal in a six-sided regular cell pattern (Fig. 2).  Our approach to enable fast handoffs over the A T M / B - I S D N backbone  while minimizing resource utilization is to predetermine the routes for all potential handoff connections at call setup via I N control, to employ a fast resource allocation scheme for completing the establishment of the selected handoff connection, and to enable fast handoff processing during call handoff without IN involvement. The proposed approach calls for a multipoint-to-point connection or connection tree configuration to link the current and the potential handoff connections between the M T and the network.  As  illustrated in Fig. 3, the connection tree topology can be either distinct-threaded, with individual links specific to each of the multiple endpoints (i.e., specific links) converging at a common endpoint at the other end, or common-threaded, with specific links from the multiple endpoints converging at a common intermediate node, namely the tethered point (TP), within the network before progressing to the endpoint at the other end over common links.  In the extreme case where the T P maps into the single endpoint, the common-  threaded multipoint-to-point connection reduces to the corresponding distinct-threaded multipoint-to-point connection.  In a later section, it will be shown that the proposed fast seamless handoff scheme produces better performance with the common-threaded multipoint-to-point connections than with the corresponding distinct-threaded multipoint-to-point connections. Currently, 16  ATM-Based Backbone  0—0-0—  0 T  •  Mobile Terminal  Mobile Switching Centre  Base Station Cell  0  Tethered-point Connecting Point (ATM Switch) (ATM Switch)  Figure 2 Location Independent Transport over Backbone A T M Network A T M does not support common-threaded multipoint-to-point connections. In the following section, we propose to extend the A T M architecture to include the M V C connection, a dual common-threaded multipoint-to-point connection (Fig. 3) to support mobile connection-oriented bearer service over B-ISDN. To support network-wide terminal mobility, the potential handoff connections of the M V C are reconfigured after each call handoff to enable continuous support of successive handoffs during a call session.  17  ^  Endpoint  ^——  O  Tethcrcd-Point  Common Links  O  Connecting Point  Specific L i n k s  Figure 3 Connection Tree Configurations  2.1 MVC Connection Configuration The proposed M V C ATM  connection is conceptually implemented at a new level of the  layer hierarchically above the VC and the V P levels (Fig.  4). A generalization of the  fixed connection tree in [10], this topology dynamically changes the multipoint-to-point connection and relocates the TP to another node after each handoff. As stated before, the multiple endpoints include the current BS and the BSs in neighboring potential handoff cells. The single endpoint at the other end represents a fixed terminal, or a BS associated with a peer MT.  The connecting points are ATM 18  switches. The original connection and  the potential handoff connections are each composed of specific connecting links for that connection and common connecting links for all these connections. The specific connecting links represent the "mobile" links which need to be reconfigured during call handoff and the common connecting links represent the "fixed" links that stay constant during call handoff.  The TP represents the connecting point where the mobile links  converge to a common link. Thus a M V C connection consists of multiple mobile specific link-sets jointed to a single fixed common link-set at a TP. Note that after each handoff, a new set of mobile links are chosen, and the TP is reconfigured to a possibly different connecting point in the network where the new set of mobile links converge. In the usual hexagonal cell grid, after a handoff the set of reconfigured potential handoff BSs includes the old BS, two previous potential handoff BSs, and three new BSs.  A l l logical links of the multipoint-to-point connection between the multiple endpoints and the TP are supported by the proposed group virtual channel (GVC). At the virtual channel (VC) level, a logical connection between two points is defined by a series of logical link identifiers associated with the concatenation of logical links. If the multipointto-point connection were to be implemented at the V C level, then the logical connections between the multiple endpoints and the TP are defined by different series of logical link identifiers and the incoming logical link identifiers at the TP will be all different. On the other hand, at the proposed G V C level, all logical connections between the multiple endpoints and the TP are defined by a single logical link identifier associated with all the concatenated links that is unique at the TP. Consequently, the employment of G V C allows a TP to associate a single G V C identifier (GVCI) with a group of endpoints associated with an M T . The G V C I is encoded in the standard V C identifier (VCI) field of the cell header. 19  _  M V C Connection  Mobile  •0  Virtual Circuit  o.  Level (MVC)  Tethered  Tethered  Point  Point  [Mobile' VUser  HooTle Specific  Links V /  1  A/Mobile\ Fixed Common Link  Mobile SpecificlinksWVUser  4 *  Group Virtual Channel  i m  Mpt-to-pt \ GVC Connection  ] 0*  VP Connections  <ja  D'  Level (GVC)  \  Mpt-to-pt / GVC Connection /  A  Virtual Channel Level (VC)  -j. VP Connections Virtual Path Level (VP)  Transmission Path  /'Transmission Patliv  ' Transmission Path i  Level  Connecting Link  % Endpoint H  * Potential Connecting Link  Potential Endoint  Q Connecting Point  Figure 4 M V C Connection Architecture to Support Mobile Communications  The G V C is supported over a group of virtual paths (VPs) which correspond to the set of mobile specific links of the original and potential handoff connections. The VPs are fixed by network topology and bind the BSs to the TP. The VCs are multiplexed within a V P which is in turn multiplexed with others within a physical link. Unlike a V C , a V P route can be established without assigning bandwidth along the path. Since VPs are 20  usually established permanently or on a long term basis, G V C s can be established over them on demand. Consequently, the support of the G V C over a group of V P s allows the T P to identify the M T via the G V C and to identify the mobile specific link-sets from the M T to the T P via the V P s . Each M V C requires reservation of the following logical link identifiers:  (1) the  incoming and outgoing VPIs at each connecting point of current or potential handoff mobile links, up to the T P ; (2) a common G V C I for all the mobile links reserved at the TP; and (3) the incoming and outgoing VCIs and VPIs reserved at each connecting point of common fixed links, starting from the T P . A T M switching nodes generally support V P crossconnection and V C / V P crossconnection, using the first bit of the 12-bit V P I field to indicate the crossconnection type. The G V C I is encoded in the 16-bit V C I field. Consequently, each node operating as a T P can maintain a local limit of 2  1 6  G V C I s to  identify M V C s supporting mobile calls, and each node operating as a mobile connection connecting point can maintain a local limit of 2  1 1  V P I to identify mobile links.  In  general, M V C s multiplexed into a common V P over a mobile link must have distinct GVCIs.  Therefore the G V C I space must be managed so that if the connection trees  emanating from two TPs share common VPs over some mobile links, then the G V C I s assigned at these TPs do not overlap. Alternately, different VPIs could be assigned to ensure that physically overlapping connection trees are logically distinct. Even though switches may not be able to process the full range of G V C I or V P I values, by proper management of the address space and reusing addresses where possible, a substantial number of M T s can be globally accommodated by the M V C connection architecture.  2.2 MVC Connection Establishment In a general connection-oriented packet-switched network, connection establishment 21  involves the following steps: (1) reservation of logical link identifier at each switch associated with the connection; (2) establishment of routing information for translating incoming logical link identifiers into outgoing logical link identifiers; (3) reservation of communication resources (buffer and physical link bandwidth) at each switch. The same procedures are employed in the A T M / B - I S D N which employ fast packet switching with IN support, where logical links are identified by virtual circuit identifiers (VCIs) or virtual path identifiers (VPIs) reserved via the IN. The new connection establishment scheme is based on reserving logical link identifiers via the IN for handoff connections, but delaying resource reservation until handoff processing (which does not involve the IN), when a fast resource reservation scheme is invoked to complete the setup of the handoff connection. The proposed scheme for establishment of the M V C connection during initial call processing is decomposed into the following tasks, with logical link identifiers ( G V C I , V C I and VPI), routing information, and resource requirements provided by the IN:  1.  Establishment of the fixed common connecting links shared by the original connection and the handoff connection —  a.  Standard V C connection establishment along the fixed common connecting links, i.e., VCIs and routing information entries in V C switch tables are reserved by the A T M switches along the fixed common connecting links;  b.  Communication resources are reserved to satisfy connection performance requirements.  2.  Establishment of the mobile specific link-sets between the multiple BSs and the T P —  a.  G V C connection establishment between the BSs group and the T P , a single unique logical link identifier in the form of G V C I is reserved by the T P ; 22  b.  Standard V P connection establishment between each B S and the T P (since the VPs are usually pre-established permanently by the network, this step may not be necessary), each B S stores its outgoing V P I accordingly;  c.  Bandwidth for the G V C connection is reserved from the V P associated with the mobile link-set specific to the original connection;  d.  For those V P s associated with the mobile link-sets specific to the potential handoff connections, bandwidth is not reserved immediately for the G V C connection; instead, a fast resource reservation scheme described later is employed to reserve bandwidth for the actual handoff connection during handoff processing.  After a handoff, the actual handoff connection becomes the original connection subject to a potential future handoff, and the T P may be mapped to a different connection point with greater hop separation from the B S to prevent the potential handoff connections to include the same links twice.  In anticipation of a future handoff, once handoff  processing is complete, the M V C needs to be reconfigured by repeating the above process. If T P remapping is required, the original T P and the reconfigured T P would engage in seamless redirection of the A T M cells into an alternative logical path within the same link-set via switching of routing entries. The M V C reconfiguration process after handoff will be explained in Section 2.4.  2.3 MVC Rerouting During Handoff Mobile link-set rerouting within the M V C connection in response to a M T handoff is illustrated in Fig. 5, where an M T has roamed from cell X to cell Y and is accessing a new B S controlled by a different M S C . Consider the initial situation when the M T was located in cell X . The A T M cells from the old M S C X arrives at port 1 of node X I bearing GVCI-j and VPI-la. Node X I 23  GVCI-j, VPI-llaT GVCI-j, VPI-l4 12 i 2 VP CrossConnect 3 4 NodoYll 3 4 Node X1 GVCI-k, VPI-lllb GVCI-k, VPI-IVb GVCI-j, VPI-lb GVCI-j, VPI-Hty 1 2 12 VP CrossConnect 3 , NodfiY? 3 4 Node X2 k GVCI-k, VPI-IMa GVCI-k, VPI-IVa GVCI-j, VPI-lc^ \ GVCI-j, VPI-lli 1 91 2 92 \ r, V l- ('n>v,r»iin«c-i 1 I i-rlo-riil I'.inin  VP CrossConnect Table (Node XI) Incoming Port VPI  la lllb  Outgoing Port  VPI  3  lb lllc  1 4  ?  i-X"  33  3  Before Handoff V C / V P CrossConnect Table (Tethered Point)  VCI-t, VPI-Vlb  VCI-m, VPI-Va  VCI  \'< VI 1 nisstoiiliiil 1  t  VCI-s, VPI-V1a  VCI-n, VPI-Vb  \ C \ 1" Cnjsxi.iDiHO ('I t-llii-ivd I* ti GVCI-p, VPI-Xa}  i (GVCI)  Outgoing Incoming Port VCI VPI VPI k Ilia 3 Vlb (GVCI) Ic  1  m  Va  Port 91 93  After Handoff  GVC-q, VPI-Xlb  V C / V P CrossConnect Table (Tethered Point)  VP CrossConnect  VCI  GVCI-p, VPI-Xb I " GVCI-q, VPI-XIa  Outgoing Incoming Port v a VPI VPI  t  Vlb  3  i (GVCI)  He  2  k IVa (GVCI) m  Va  Port 92 93  Dp VTrF] Figure 5 M V C Mobile Links Rerouting During Handoff  performs V P crossconnection: it translates VPI-la at port 1 to VPI-lb at port 3 with the outgoing A T M cells bearing GVCI-j and VPI-lb. Then the A T M cells arrive at port 1 of node X 2 bearing GVCI-j and VPI-lb. Node X 2 performs similar V P crossconnection. After passing through node X 2 , the A T M cells arrive at port 1 of the T P bearing GVCI-j and VPl-Ic. Since both the G V C I and the V C I are encoded in the same field of the A T M cell header, the T P performs standard V C crossconnection and it translates GVCI-j and VPI-Ic at port 1 to VCI-m and VPI-Va at port 93. O n the other hand, the reverse traffic from the fixed link-set would arrive at port 3 of the T P bearing VCI-t and VPI-Vlb . The  24  T P performs standard V C crossconnection and it translates VCl-t and VPI-VIb at port 3 to GVCI-k and VPI-IIIa at port 91. Now consider the handoff situation when the M T is about to enter cell Y . The old B S - X would signal the T P to indicate that the GVC-} connection will exit from V P - I and the new B S - Y would signal the T P to indicate that GVC-} connection will now be multiplexed instead within VP-II. The T P rerouting controller would update both the forward and reverse routes translation stored in the V C crossconnection table accordingly; i.e., the reverse group virtual circuit connection GVC-k will be moved from V P - I l l to VP-IV. After successful handoff signaling, the forward traffic from M S C - Y would arrive at port 2 of the T P bearing GVCI-] and VPl-llc.  The T P performs standard V C  crossconnection and it translates GVCI-j and VPl-llc at port 2 to VCI-m and VPI-Va at port 93. O n the other hand, the reverse traffic from the fixed link-set would arrive at port 3 of the T P bearing VCl-t and VPI-VIb. The T P performs standard V C crossconnection and it translates VCl-t and VPI-VIb at port 3 to GVCl-k and VPl-lVa at port 92.  2.4 MVC Reconfiguration After Handoff Given that the connection tree concept is employed to support user terminal mobility over the A T M / B - I S D N and that the M T has unrestricted movement across the mobile networks, it is necessary to address the problem of supporting multiple successive handoffs over the A T M / B - I S D N as an M T roams to adjacent mobile cellular regions. One solution is to assume that the range of movement of an M T can be predetermined during call setup, in which case a static connection tree [10] is sufficient to support user terminal mobility. Essentially, this solution only supports restricted successive handoffs and forced call terminations would occur when an M T roams outside the assumed mobility range. 25  In contrast, we allow an M T to have an unrestricted range of movement or network-wide terminal mobility. Under this non-limiting assumption, our solution is to employ the dynamic connection tree described above to provide unrestricted and continuous handoff support. The connection tree is dynamic because after each handoff, the connection tree is reconfigured to account for the new set of immediate neighbouring cells into which the M T can potentially enter. Since the connection tree is implemented by the proposed M V C , we now discuss how the M V C is reconfigured after each handoff to support successive handoffs. M V C reconfiguration after handoff is illustrated in Fig. 6, where an M T has roamed from cell 0 to cell 6 and is accessing a new BS controlled by a different M S C . Cell 0 is an immediate neighbour to cells 1 to 6; while cell 6 is an immediate neighbour to cells 0, 1,5, 7, 8, 9. Before the handoff, the M V C is characterized as follows: TP at node D; fixed common links DE, EF, FG and G H ; current ongoing connection associated with cell 0 has mobile specific links A B , B C and CD; and the potential handoff connection associated with cell 6 has mobile specific links X Y , Y Z and ZD. The IN-controlled M V C reconfiguration after each handoff (e.g. from cell 0 to cell 6) is described as follows: 1. The potential mobile endpoints are reconfigured to account for the new set of immediate neighbouring cells into which the M T can potentially enter. 2.  As illustrated in Case 2 of Fig. 6, the TP may be mapped to a different connecting node with increasing hop separation from the BS (i.e., nodes E, F, G, and H) to prevent the potential handoff connections to include the same links twice. The route for the mobile specific links from each potential mobile endpoint (i.e., those associated with cells 1, 0, 5, 7, 8, 9) to the reconfigured TP is determined via IN control. 26  a Mobile Terminal  Mobile Switching Centre  A  0  Base Station  Cell  m  Tethered-point ( A T M Switch)  o Connecting Point ( A T M Switch)  Figure 6 M V C Reconfiguration After Handoff  The reconfigured T P is requested to reserve the same current ongoing G V C identifier to support the multipoint-to-point connection between the new set of multiple endpoints and the reconfigured T P . If the particular G V C identifier is being employed by 27  another M V C at the reconfigured T P , then a new G V C identifier must be reserved. 4.  In the original T P , the routing entry translating the incoming mobile link into the outgoing fixed link will be complemented with an alternative routing entry translating the same incoming mobile link into the alternative outgoing mobile link.  5.  In the reconfigured T P , the routing entry translating the incoming fixed link into the outgoing fixed link will be complemented with an alternative routing entry translating the same incoming fixed link into the alternative outgoing mobile link.  6.  In each of the connecting nodes between the original T P and the reconfigured T P , the routing entry translating the incoming fixed link into the outgoing fixed link will be complemented with an alternative routing entry translating the alternative incoming mobile link into the alternative outgoing mobile link.  7.  After all the alternative routing entries have been established, the original T P would redirect the outgoing A T M cells to the reconfigured T P via the alternative entry. Likewise in the other direction, the reconfigured T P would redirect the outgoing A T M cells to the original T P via the alternative entry.  2.5 Effect of TP Availability On Connection Reuse Reuse of connecting links of the old connection by the handoff connection should be maximized to minimize handoff processing delay. With the M V C connection architecture, the degree of connection reuse is measured by the ratio of the number of shared fixed links to the number of total links. This section examines how the degree of connection reuse is affected by the degree of T P availability (as measured by the proportion of network nodes that can function as TPs) and the degree of node connectivity (defined as the ratio of the actual number of links to that required to fully interconnect all the network nodes). 28  1  Assuming a hexagonal cell grid with six potential handoff connections, a simulation program is written to see how the connection reuse ratio depends on the T P availability ratio for a given node connectivity ratio. The program performs a random experiment by generating a random network of 100 nodes based on a node connectivity ratio and a T P availability ratio, and determines the set of connection reuse ratios for all possible M V C connection trees that may be configured for this network. Each M V C connection tree is generated based on randomly selected mobile endpoints associated with a proximity parameter and a randomly selected destination endpoint; and the connection reuse ratio of each generated connection tree is recorded. For each specified combination of node connectivity ratio and T P availability ratio, the experiment is repeated 100 times and an overall average connection reuse ratio is obtained by averaging over all elements in the connection reuse ratio sets resulting from all repetitions of the experiment. Fig. 7 shows the effect of the T P availability ratio on the connection reuse ratio for a given node connectivity ratio. For a given T P availability ratio, increasing the node connectivity ratio tends to increase the connection reuse ratio. For a given node connectivity ratio, the connection reuse ratio increases as the T P availability ratio with decreasing increments. As the T P availability ratio increases. When the T P availability ratio approaches one, the connection reuse ratio approaches a maximum value of 0.7 regardless of node connectivity ratio.  It is desirable to maximize the T P availability ratio in order to facilitate connection reuse. This is possible if every network node can function as a T P , and the discussion in Section 3.2 show that this is not a particularly onerous requirement since the existing switch architecture can be augmented with add-on signaling components as illustrated in Fig. 16.  Under such conditions, the connection tree can be configured using the 29  1 .o No. of Potential Handoff Connections = 6  0.8  ca  ar  0.6  § CD  or  0.4  Q Node Connectivity Ratio = 0.01 •sfc- Node Connectivity Ratio = Q.1  0.2  OO  O.O  0.2  0.4 0.6 T P Availability Ratio  0.8  1 .O  Figure 7 Effect of T P Availability on Connection Reuse following algorithm:  1.  Determine the optimal route for the current connection between the current mobile endpoint and the destination endpoint.  The decision is based on minimizing the  number of total links of the current connection. 2.  Select one of the nodes along the determined route of the current connection to be the T P . The decision is based on minimizing the number of mobile specific links or maximizing the number of fixed common links of the current and the potential handoff connections.  3.  Determine the route for the mobile specific links from each potential mobile endpoint to the T P .  30  Chapter 3 Signaling Network Architecture and Transaction Protocols to Support Real-time Connection Rerouting in ATM/B-ISDNs The connection rerouting service consists of the following connection control services, i.e., new connection establishment followed by traffic rerouting from the old to the new connection and optional release of old connection, supported by corresponding signaling protocols. The signaling protocols can employ C C S over separate signaling connections, or ihband signaling over the user connections, to transport signaling messages.  The  user-perceived connection performance lapses due to connection rerouting include call disruptions, delay jitter and forced call termination. Call disruption due to transient data loss error can be caused by momentary unavailability of the (old or new) user connection, or premature information rerouting from the sender node via the new connection before the receiver node is ready. Call disruption due to transient data-unit sequencing error occurs if the inaugural packet or cell of the new connection overtakes final cells of the old connection at the destination node due to a mismatch of cell transit time between the old and the new connections.  For synchronous traffic, if the inaugural data-unit of  the new connection arrives later than the expected time at the destination, transient delay jitter occurs due to the traffic lagging-gap. Forced call termination occurs when a new connection cannot be established due to lack of network resources or signaling failure. A call rerouting is said to be seamless if it is completed without call disruptions and forced call termination, and with minimized delay jitter. This thesis proposes a real-time connection rerouting service to enable fast, robust and seamless call rerouting.  Connection dynamics are determined by the temporal relationship among connection rerouting initiation, old connection take-down and new connection establishment. 31  The  delay of old connection take-down from connection rerouting initiation determines the upper bound of allowable connection rerouting time. According to the temporal relationship between old connection take-down and new connection establishment, connection dynamics as illustrated in Fig. 8 can be classified as either sequential where the old connection is taken down before the new connection is established, or concurrent where both the old and the new connections are available upon the initiation of connection rerouting. Sequential connections dynamics is mandatory for rerouting necessitated by an unforeseen old connection take-down or if the nodes where the old and the new connections converge cannot access both connections simultaneously.  We will show how  concurrent connections dynamics, subject to early detection of rerouting initiation prior to old connection take-down, may employ real-time connection rerouting to achieve fast, robust and seamless call rerouting. We will also show how sequential connection dynamics employs real-time connection rerouting to achieve call rerouting with minimal call disruptions or errors.  Under concurrent connection dynamics, data loss error is preventable by employing fast connection rerouting to ensure connection rerouting is completed before the old connection is taken down and by employing connection rerouting synchronization at the sender node with a simple protocol dictating that the sender node must not reroute traffic via the new connection until the receiver node is ready for reception via the new connection. Under sequential connection dynamics, data loss error is non-preventable and fast connection rerouting can be employed to minimize the error interval. Regardless of the type of connection dynamics, robust new connection establishment is always required to minimize the chance of forced call terminations.  With inband signaling, since the  take-down of the old connection would also cause signaling failure, a fast connection 32  Connection _, , _ Reouting ™ Connection Initiation Take-down  New Connection New Connection Establishment Establishment Initiation Completion  Information Rerouting Initiation  New Connection Establishment Delay  Information Rerouting Completion  Information Rerouting Delay  Connection Rerouting Delay Call Disruption Delay  Sequential Connection Dynamics  Rpontlnn Reouting Initiation  ° ° Establishment |  N  e  w  C  n n e c t i  n  n i t i a t i o n  New Connection Establishment Completion r  Information Rerouting Initiation  Rerouting Completion  Old Connection Take-down  I.NUfUIUII  New Connection Establishment Delay  Information Rerouting Delay  Connection Rerouting Delay  *~\  Upper Bound of Connection Rerouting Delay -  Concurrent Connection Dynamics  Figure 8 Connection Dynamics due to Real-Time Operational Requirements on Connection Rerouting rerouting service that ensures signaling is completed before old connection is taken down is also required to prevent forced call termination. However, forced call termination is non-preventable under sequential connection dynamics with inband signaling. With sequential connection dynamics, data-unit sequencing error is non-preventable. With concurrent connection dynamics, data-unit sequencing error can be prevented by providing packet-ordering  synchronization at the receiver node, with a simple rule  dictating that incoming packets or A T M cells via the new connection must not be 33  presented to the user until the last cell via the old connection has been received. The receiver node must be notified of the following events to achieve packet-ordering synchronization:  (1) arrival of the last cell via the old connection; (2) arrival of the  inaugural cell via the new connection.  If the inaugural cell arrives before the last  cell, then synchronization waiting results. During the synchronization waiting interval, the receiver node buffers the incoming cells via the new connection until the last cell arrives and then presents the buffered cells to the user. To mark the events temporally and to identify the synchronization user cells (the last or inaugural user cells) without marking the user cells themselves, use of inband signaling messages is mandatory, as it is impossible to ascertain timing relationships between C C S messages and user cells. Disruptive packet-ordering synchronization [35] results if the receiver node would simply abort the old connection voluntarily before the sender node reroutes information via the new connection. The performance requirements on the necessary signaling protocols and network architecture to support real-time connection rerouting under different connection dynamics are summarized in Table 1.  Depending on the connection dynamics, real-time connection rerouting requires one or more of the following supporting services: fast connection rerouting service, robust new connection establishment and packet-ordering synchronization during information rerouting. A fast connection rerouting service necessitates fast connection establishments. As illustrated in Fig. 9, this thesis proposes the robust fast resource reservation transaction protocol (RFRTP) during new connection establishment over a predetermined route in Section 3.1.1, and packet-ordering synchronization transaction protocols (POSTPs) during information rerouting in the intra-network and interworking environments in Section 3.1.2. The associated-CCS and the inband signaling network mechanisms to support the R F R T P 34  User-Perceived Connection Performance  Real-Time Connection  Data Loss Error  Rerouting  Data-unit  Forced Call Termiantion  Sequencing Error  Operational Attributes Sequential  Non-Preventable  Non-Preventable  Dynamics  Preventable with Common Channal  Connection  Signaling  (delay minimized by fast connection  (by robust new  rerouting)  connection establishment) Non-Preventable with Inband Signaling Concurrent  Preventable  Connection Dynamics  (by fast and  Non-Preventable  Preventable with  with Common  Common Channel  Channel Signaling  Signaling  synchronized connection  (by robust new  rerouting)  connection establishment) Preventable with  Preventable with Inband  Inband Signaling  Signaling  (by packet-ordering  (by fast connection  synchronization)  rerouting and robust new connection establishment)  Table 1 Performance Requirements for Real-Time Connection Rerouting and the P O S T P respectively are described in Section 3.2. With the new connection being established during connection rerouting, the proposed R F R T P and the associated-CCS mechanism can be employed under sequential connection dynamics to minimize call disruption delay, and under concurrent connection dynamics to avoid call disruption. The proposed P O S T P and the inband signaling mechanism can be employed under concurrent connection dynamics to avoid packet-disordering. Seamless connection rerouting under concurrent connection dynamics can be achieved by employing the proposed R F R T P ,  SS7/BISDN Protocol Model BISDN-UP  MAP  Mobile Application Part  BISDN User Part  Self-Healing or Fast Restoration^ OSI Protocol Model  SS7/ISDN Protocol Model ISDN-UP  OMAP  ISDN User Part  Mobile Call Handoff  ASE ASE Real-Time Static Connection Rerouting  ASE  ASP  ASE Real-Time Mobile| Connection Rerouting  Application  TCAP Transactions Capabilities Applcation Pari  POSTPs  RFRTP  PacKft-OroVring Synchronization Transaction Protocols  Robust Fast Reservation Transaction Protocol  Presentation Session Transport  SCCP  SCCP  Signalling Connection Control Part  Network  MTP Level 3  MTP Level 3 M e s s a g e Transfer Part Level 3  Signalling ATM Adaptation Layer Inband Signaling  Associated-CCS Source-Routing Unicast & Multicast!  MTP Levels 2 and 1 ATM Physical  Figure 9 Proposed Signalling Network Architecture and Transaction Protocols to Support Real-Time Connection Rerouting P O S T P and the corresponding signaling mechanisms.  36  3.1 Signaling Transaction Protocols 3.1.1 Robust Fast Reservation Transaction Protocol (RFRTP) The R F R T P assumes that an end-to-end route has been selected prior to the reservation attempt. The route is known at the source node in the form of source-routing information, and  signaling transports via source-routing unicast and multicast are assumed.  The  R F R T P provides the fast transaction for reserving distributed resources (e.g., bandwidth, logical link identifiers) along a selected route, and the robust retry transaction for reserving alternative resources along a localized reroute between two neighbouring nodes of the original selected route. Each transaction defines the context within which distributed signaling operations that coordinate multiparty interactions are executed. The parties involved in each transaction include a reservation initiator in the source node, a set of distributed resource servers in the connecting nodes along the original given route, and an overall resource reservation responder in the end node. The R F R T P provides atomic commitment in executing the signaling operations of the reservation transaction, i.e., either all of the operations are performed, or none is performed. support the atomic commitment.  A fast two-phase commit scheme is proposed to The logical link identifiers and the link bandwidth  can be reserved together in both directions of a duplex connection via the two-phase commitment scheme, and the routing table entries can be established during the final commitment phase. Each node is assumed to be configured with a reservation initiator, a distributed resource server and a reservation responder. The signaling messages provided to support the R F R T P ' s signaling operations are described in Table 2. The fast reservation transaction comprises of the following signaling operations: 37  Table 2 R F R T P Signaling Messages R F R T P Signalling Message  From  To  Description  ROBUSTRERSERVE  Reservation Initiator  Remote Reservation Responder  Begin a reservation transaction with localized retry  NON-ROBUSTRESERVE  Reservation Initiator  Remote Reservation Responder  Begin a reservation transaction with no localized retry  INITIAL-COMMIT  Reservation Initiator  Remote Resource Servers  Initial commitment of distributed resource  ALLOCATIONSUCCESS  Resource Servers  Remote Reservation Responder  Indicate a successful status for the initial commitment based on the allocation control result  ALLOCATION-FAIL  Resource Servers  Remote Reservation Responder  Indicate a failed status for the initial commitment based on the allocation control result  FINAL-COMMIT  Reservation Responder  Remote Resource Servers  Final commitment of distributed resource  ABORT  Reservation Responder  Remote Resource Servers  Abort the reservation transaction by releasing the initial commitment of the distributed resource  LOCAL-RETRY  Resource Server  Colocated Reservation Initiator  Begin a local retry transaaction  RESERVESUCCESS  Rerservation Responder  Reservation Initiator  Indicate a successful status for the reservation transaction  RESERVE-FAIL  Rerservation Responder  Reservation Initiator  Indicate a failed status for the reservation transaction  REROUTINGREPORT  Resource  Reservation Responder  Report the local alternative route along which the distributed resource is reserved  38  1.  The reservation initiator signals the reservation responder to begin the transaction of reserving the distributed resources as one overall unit, and to specify the allocation policy and retry mode (retry or no retry).  2.  The reservation initiator signals the distributed resource servers for the initial commitment of a specified amount of the distributed resources according to a specified allocation control policy and retry mode.  3.  Each distributed resource server executes the allocation control processing.  If  reservation fails and retry is enabled, the resource server initiates the localized retry transaction as described later. 4.  Each distributed resource server signals the reservation responder to report the initial commitment result, and the localized rerouting information if necessary.  5.  The reservation responder signals the distributed resource servers for the final commitment or release of the distributed resources.  6.  The reservation responder signals the reservation initiator to report the result of the transaction.  If a distributed resource server determined that reservation fails along the link to the next connecting node along the predetermined or original selected route and retry is enabled, the resource server would execute a localized retry by selecting an alternative route (consists of one or more connecting links) to the next connecting node as illustrated in Fig. 10. The local rerouting algorithm avoids back-hauling by rejecting any alternative route that calls for preceding and succeeding nodes as connecting nodes. The identities of the preceding and succeeding nodes are encoded in the source-routing information field of the incoming signaling cell via the associated-CCS transport mechanism as described in Section 3.2.1. The source-routing information originated at the reservation initiator 39  Figure 10 Localized Rerouting node identifies the series of preassigned node addresses along the predetermined route to the reservation responder node, and it is updated after each local rerouting. After a connecting node has determined the local alternative route to the neighbouring connecting node that a message will follow, it expands the incoming source-routing information by inserting the partial-sequence of alternative node addresses after its own node address.. The localized retry transaction comprises of the following signaling operations:  1.  The resource server signals the collocated reservation initiator to coordinate the reservation along the alternative route to the next connecting node with no retry.  2.  The collocated resource initiator starts a reservation transaction as described above by signaling the reservation responder of the next connecting node and specifying no retry.  Fast Reservation Transaction  Fig. 11 illustrates the fast resource reservation transac-  tion of the R F R T P . A n application process in the source node invokes the collocated 40  reservation initiator with the following parameters. The first parameter is the series of globally known logical names of the distributed resource servers and the reservation responder (always at the end destination) along a predetermined route. The second parameter specifies the demand on the distributed resources. The third parameter specifies the allocation control policy or the manner under which the distributed resources are allocated (e.g. peak or average bandwidth allocation). These reservation parameters are based on the service parameters negotiated between user and network service provider during call setup.  The reservation initiator composes the I N I T I A L - C O M M I T messages to the resource servers and the R O B U S T - R E S E R V E message to the reservation responder into a single multicast message and sends this composite message via the source-routing multicast (copy-and-forward) signaling transport service described in Section 3.2.1. The I N I T I A L C O M M I T / R O B U S T - R E S E R V E multicast messages consist of the following parameters: reservation transaction id, set of translated source-routing addresses, allocation control policy, demand on resources and retry mode.  When the multicast message from the reservation initiator arrives at a node, the signaling transport entity first examines the next addressing sub-field in the source-routing address field. If the sub-field is empty, the node is the destination node and the signaling transport entity then passes the multicast message as a R O B U S T - R E S E R V E message to the reservation responder. Otherwise, the node is a connecting node. The signaling transport entity in each connecting node determines the next downstream node towards the destination and the corresponding connecting link by decoding the appropriate subfield of the source-routing address field, and executes the following tasks:  (1) copy  the multicast message and pass the copy as a I N I T I A L - C O M M I T message to the local 41  Figure 11 Example of R F R T P Operations with Fast Reservation Transaction resource server; and (2) forward the multicast message to the next downstream node along the predetermined route. When the resource server receives the I N I T I A L - C O M M I T message, it executes the resource allocation algorithm according to the specified allocation control policy.  If  allocation succeeds, an appropriate amount of resources would be set aside according to 42  Second Reception C y c l e Delay  f  I  Non-robust  RgaarvaV  I Transaction End  j  """transit Transit Delay  O End Node  ^verify  Verification Control Processing Delay  ® Forward Node  alloc  Allocation Control Processing Delay  T  £ Copy-and-forward Node  Figure 12 Example of R F R T P Operations with Fast Reservation and Robust Retry Transactions  the specified demand. Upon completion of allocation control processing, the resource server sends an A L L O C A T I O N - S U C C E S S or A L L O C A T I O N - F A I L U R E message to the 43  reservation responder to indicate the outcome of the allocation. Each of these allocation status messages is identified by the reservation transaction id parameter.  If localized  retry transaction has occurred, the resource server also sends a R E R O U T I N G - R E P O R T message to the reservation responder to report the local alternative route along which the distributed resources are reserved. When the R O B U S T - R E S E R V E message is forwarded by the signaling transport entity at the destination node to the local reservation responder, the reservation responder waits to receive all the allocation status messages from the resource servers and verifies the overall availability of the distributed resources by an "all-or-none" algorithm.  The  verification succeeds if all received allocation status messages are A L L O C A T I O N S U C C E S S , and fails if any A L L O C A T I O N - F A I L U R E message is received, or not all allocation status messages are received within a specified time window. Upon successful verification of resource allocation, the reservation responder sends the F I N A L - C O M M I T messages to the resource servers and the R E S E R V E - S U C C E S S message to the reservation initiator.  If the verification fails, the reservation respon-  der sends the A B O R T messages to the resource servers and the R E S E R V E - F A I L U R E message to the reservation initiator. The F L N A L - C O M M I T and R E S E R V E - S U C C E S S messages are composed in a single multicast message for transmission by the reservation responder using the source-routing multicast signaling transport service.  The A B O R T  and R E S E R V E - F A I L U R E messages are likewise composed into a single multicast message and transmitted. Each of these composite multicast messages is identified by the reservation transaction id parameter.  Processing and forwarding of the (return) multicast messages from the reservation responder via the connecting nodes to the reservation initiator follow the same procedures 44  as described above for multicast messaging in the opposite (forward) direction. A t each connecting node, the signaling transport entity passes the embedded F I N A L - C O M M I T or A B O R T message to the local resource server which performs the corresponding actions on the resources previously set aside in the initial commitment. A t the reservation initiator, the signaling transport entity passes the embedded R E S E R V E - S U C C E S S or R E S E R V E F A I L U R E message to the reservation initiator, which initiates the next phase of the rerouting operation or recovery actions (e.g., retry or report to network management system).  Robust Localized Retry Transaction  Fig.  12 illustrates  the robust localized retry trans-  action of the R F R T P when there is a congested link and the local alternative route for reservation retry consists of two links with one additional connecting node. Each distributed resource server along the predetermined route can initiate the localized retry transaction when it determines that resource reservation along the predetermined link to the next downstream connecting node has failed.  The resource server first selects an  alternative route (consists of one link, or one or more connecting nodes) to the next downstream connecting node. Then it sends the L O C A L - R E T R Y message to the local reservation initiator to reserve the distributed resources along the local alternative route with no further retry. The L O C A L - R E T R Y message contains the retry transaction id parameter. The local resource initiator starts a reservation transaction as described previously by signaling the reservation responder of the next connecting node and specifying no retry via the N O N - R O B U S T - R E S E R V E message. Upon completion of the localized retry transaction, the resource server sends an A L L O C A T I O N - S U C C E S S or A L L O C A T I O N - F A I L U R E message, identified by the reservation transaction id parameter, to the reservation responder to indicate whether the  45  allocation is a success or failure. If the localized retry transaction is successful, the resource server also sends a R E R O U T I N G - R E P O R T message to the reservation responder to report the local alternative route along which the distributed resources are reserved.  Error Recovery  During the transport of the  INITIAL-COMMIT/ROBUST-RESERVE  multicast message from the reservation initiator to the resource servers and reservation responder, the employment of robust source-routing via localized retry will bypass link failures (including congestion), but it will not overcome node failures. If node failures occur, the intended reservation responder will remain in the inactive state since it will never receive the I N I T I A L - C O M M I T message. The intended reservation responder will receive and discard unsolicited A L L O C A T I O N - S U C C E S S messages from those resource servers that have been activated before signaling is stopped by a node failure. Under this situation, the activated resource servers and the reservation initiator will never receive the F I N A L - C O M M I T / R E S E R V E - S U C C E S S multicast message from the intended reservation responder.  The activated resource servers will time out and release the  resource allocated during the initial commit phase.  The reservation initiator will time  out and enter the reset state (resulting in retry or aborted operation). If node or link failures occur during the transport of  ALLOCATION-SUCCESS  message from a resource server to the reservation responder, the reservation responder will time out and send a R E S E R V E - F A I L U R E message to the reservation initiator. Upon receiving the R E S E R V E - F A I L U R E message, the reservation initiator will enter the reset state. If  node  or  link  failures  COMMIT/RESERVE-SUCCESS  occur  during  the  transport  multicast message from  of  the  FINAL-  the reservation responder  to the resource servers and the reservation initiator, the recovery procedures are as 46  follows:  (1) each resource server will time out and release the allocated resource; (2)  the reservation initiator will time out and enters the reset state.  3.1.2 Packet-Ordering Synchronization Transaction Protocols (POSTPs) Packet-ordering synchronization transaction protocols (POSTPs) between the sender and the receiver nodes are proposed for rerouting traffic from the old connection to the new connection in the intra-network and interworking environments. The proposed intra-network P O S T P operates over a homogeneous network and assumes the followings: (1) the sender and the receiver nodes can engage simultaneously with the old and the new connections during connection rerouting; and (2) inband signaling capability exists between the sender and the receiver nodes. With inband signaling, the sender node inserts synchronization control cells (markers) into the user information stream to identify the preceding user cell as the last cell via the old connection or to identify the succeeding user cell as the inaugural cell via the new connection. The arrival of a marker at the receiver node would mark the arrival of the corresponding synchronization user cell (last or inaugural cell) which is identified implicitly by its immediate proximity to the marker. The proposed inter-network P O S T P operates over access networks interworking via an ATM-based backbone network. In the interworking environment, end-to-end inband signaling connections may not be possible since the backbone and the access networks may employ different signaling systems. Furthermore, if the old and the new connections do not converge to the same gateway node interworking between the backbone and the access networks, an end node in the access network may not be able to access simultaneously the two gateway nodes associated with the old and the new connections. Taking these operational limitations into account and illustrated in Fig. 13, the proposed inter-network P O S T P assumes the followings: 47  (1) the end node in the access network  is restricted to access the old and the new connections sequentially; (2) the end node in the backbone network can access the old and the new connections concurrently; and (3) inband signaling capability within the backbone network. End-to-End Connections  Sequential Access by End Node  Concurrent Access by End Node  New C o n n e c - j o n U ^ ^ ^  n  «  O -  .neOionUnK-Sets Old Com  Connection Link.sete  Access Network (without Inband Signaling)  [XI  Gateway Node  ^  O  Backbone Network (with Inband Signaling)  Network Node  End  Node  Figure 13 Operational Environment for Inter-Network P O S T P  The inter-network P O S T P dictates synchronization and buffering capabilities at the gateway nodes.  Acting as proxy agents for the end node in the access network, the  gateway nodes engage with the end node in the backbone network (with concurrent access) to reroute traffic over the backbone network. O n the other hand, acting as proxy agents for the end node in the backbone network, the gateway nodes engage with the end node in the access network (with sequential access) to reroute traffic over the access network.  Intra-network POSTP  The intra-network P O S T P transaction defines the following pro-  tocol entities: synchronization initiator at the sender node and synchronization responder at the receiver node.  For example, employing the intra-network P O S T P in network  self-healing, the sender and the receiver nodes can be the pair of nodes terminating a 48  path containing the failed links, or the pair of actual terminator nodes of the original connection. The transaction comprises of the following signaling operations: 1.  The synchronization initiator signals the responder that the incoming traffic via the old connection to the responder is now terminated by issuing a T E R M I N A T E - O L D marker cell via the proposed inband signaling transport mechanism described in Section 3.2.2.  2.  The synchronization initiator signals the responder that the incoming traffic via the new connection to the responder is now activated by issuing an A C T I V A T E - N E W marker cell via the proposed inband signaling transport mechanism.  3.  If the synchronization responder receives the A C T I V A T E - N E W  marker without  receiving the T E R M I N A T E - O L D marker cell, the synchronization responder buffers the incoming cells via the new connection. 4.  The above operations are repeated in the other direction for duplex connection.  Inter-network  POSTP  The inter-network P O S T P transaction defines the following  protocol entities: (1) synchronization initiator at the sender node; (2) old and new proxy agents for the synchronization initiator at the gateway nodes of the old and the new connections respectively; (3) synchronization responder at the receiver node; and (4) old and new proxy agents for the synchronization responder at the gateway nodes of the old and the new connections respectively.  For example, employing the inter-network  P O S T P in mobile call handoff across an ATM-based backbone network, the sender or the receiver node would be mapped into either a mobile-terminal or a network switching center within the backbone network anchoring the old and the new connections, and the gateway nodes would be mapped into the base stations. 49  The transaction comprises of the following signaling operations (initiator's old and new proxy agents are omitted to avoid overt details):  1.  The synchronization initiator signals the responder's old proxy agent that the responder's incoming traffic via the old connection is now terminated by issuing a T E R M I N A T E - O L D marker via the inband signaling transport mechanism proposed in Section 3.2.2.  2.  The synchronization initiator signals the responder's new proxy agent that the responder's incoming traffic via the new or handoff connection is now activated by issuing an A C T I V A T E - N E W marker via the inband signaling transport mechanism.  3.  Upon receiving the incoming T E R M I N A T E - O L D marker, the responder's old proxy agent signals the responder to switch to its new proxy agent.  4.  Upon receiving the incoming A C T I V A T E - N E W marker, the responder's new proxy agent buffers responder's incoming traffic.  5.  Before switching to the responder's new proxy agent, the responder signals the synchronization initiator via its old proxy agent that its outgoing traffic via the original connection is now terminated by issuing a T E R M I N A T E - O L D marker via the inband signaling transport mechanism.  6.  The responder dissociates from its old proxy agent and signals its new proxy agent to activate the new connection. The responder's new proxy agent then presents the incoming buffered data to the responder.  7.  The responder signals the synchronization initiator via its new proxy agent that its outgoing traffic via the new connection is now activated by issuing an A C T I V A T E N E W marker via the inband signaling transport mechanism. 50  8.  The synchronization initiator synchronizes the responder's outgoing traffic via the old and the new mobile links by employing the incoming T E R M I N A T E - O L D and A C T I V A T E - N E W markers, and buffering the outgoing traffic on the new link as necessary.  3.2 Signalling Transports 3.2.1 Associated-CCS Transport A new associated-CCS transport architecture in A T M / B - I S D N is proposed to provide fast signaling transport for real-time transaction protocols like the proposed R F R T P . With the associated signaling mode, the user and signaling networks would share the physical, virtual circuit ( V C ) and virtual path (VP) transmission facilities over the A T M network. Based on this signalling architecture, we further propose source-routing unicast and multicast services (Fig. 14) and protocols in the signaling A T M adaptation layer ( S A A L ) which is conceptually above the A T M layer (i.e., utilizing services provided by the A T M layer). The source-routing services are supported over logical signaling links established by preassigned permanent V C s and V P s within the A T M network. Once the source-routing information is known (e.g., obtained from the A I N during call processing), the signaling initiator can start transmitting signaling messages without having to establish a signaling connection explicitly; thus enabling fast signaling transport. The format of the associated-CCS signaling cell is illustrated in Fig. 15. A s Table 3 indicates, the associated-CCS signaling cell is identified via the cell header by preassigned V C I / V P I and by the 2-bit payload type (PT) indicator being set to some non-zero code (e.g. "01"). The P T code to identify signaling cells (including operation and maintenance traffic) has not been finalized by the A T M standard; however, the general agreement is to identify user cells by setting the P T bits to "00" [2]. 51  Predetermined Connection Route  (a)  Source-Routing Unicast (Forward)  Predetermined Connection Route  • Copy  m-  (b)  |  n  Copy  •4  Copy |  Source-Routing Multicast (Copy and Forward)  # |  •  n Copy Forward ^  Copy-and-forward Node  Forward Node  Associated Signaling Application Controllers ( e . g . R e s o u r c e R e s e r v e , R e s e r v a t i o n Initiator a n d R e s p o n d e r )  Figure 14 Associated-CCS Transport: Source-Routing Unicast and Multicast Inband Signaling Cell Inband Signaling— A T M Cell Payload  - Inband Signaling A T M Cell Header  Inband Signaling A T M Adaptation Layer Header  Preassigned VCI  VPI  PT = 01  POSTP Header  Preassigned VPI to indicate inband signalling destination endpoint  POSTP Payload  Inband Signaling A T M Adaptation Layer Payload  Associated-CCS Signaling Cell Associated-CCS A T M Cell Header  *~ "*  Preassignei GFC  VPI  VCI  PT RES = 01  CLP HEC  Associated-CCS— A T M Cell Payload  SourceRouting Transport Type  No. of SourceRouting Entries  Next Entry Pointer VPI  Source-Routing Information  VCI  Porl  1  Associated-CCS A T M Adaptation Layer Header  Associated-CCS Source-routing Transport Type: (1) Unicast (Forward) (2) Multicast (Copy-and-Forward) Figure 15 Signaling Cells Format  52  VPI  VCI n  Port n  RFRTP Header  RFRTP Payload  n  I Associated-CCS A T M Adaptation Layer Payload  A T M Switch Functions  A T M Cell Header Fields  VP  VC/VP  Inband  Associated  CrossConnect  CrossConnect  Signaling  Signaling  (User or Inband  (User or  Destination  Transport  Signaling Cells)  Inband  Endpoint  Signaling Cells) VPI (12 bits)  Most Significant Bit of VPI = 0  Preassigned Most Significant Bit VPI. e.g., VPI of VPI = 1 = All l's  Preassigned VPI  V C I (16 bits)  V C I assigned  V C I assigned  V C I assigned  per call  per call  per call  Preassigned VCI  P T = 00 (user)  P T = 00 (user)  PT = 01  PT = 01  or  or P T = 01  P T = 01  (signaling)  P T (2 bits)  (signaling) Table 3 A T M Cell Header Fields Encoding and A T M Switch Functions  The payload of the associated-CCS signaling cell contains the S A A L header and payload. The S A A L header consists of the following fields: (1) protocol type indicating either source-routing unicast or source-routing multicast; (2) number of source-routing information entries; (3) next entry pointer; and (4) source-routing information entries. The source-routing information identifies the series of preassigned destination VCI/VPI/port for the subsequent signaling transport stages.  For k hops associated with the source,  connecting, and destination nodes, the source-routing information is subdivided into k subfields.  Subfield i contains the next forwarding VCI/VPI/port at stage i.  With the  source-routing unicast, a signaling cell is forwarded to the next node according to the corresponding subfield information. With the source-routing multicast, a signaling cell is copied and the copy is passed to the local application protocol entity before it is forwarded to the next node.  There are two approaches for adding the proposed associated-CCS transport service  functions to an A T M network. One is to integrate the service functions into the cell routing mechanism in the A T M switch. A more flexible alternative that can be implemented by retrofitting existing A T M switches is to append the service functions in an independent unit external to the cell routing mechanism, as illustrated in Fig. 16.  The associated  signaling transport server acts as a self-contained unit to process associated signaling cells arriving from one or more outputs of the switch fabric. The processed associated signaling cell returns to one or more inputs of the switch fabric for forwarding over the appropriate outgoing link. While endowed with flexibility, the latter approach does impose extra cell routing delays since an associated signaling cell is routed through each cell routing mechanism twice.  3.2.2 Inband Signaling Transport Besides associated-CCS transport, inband signaling transport is required to transport packet-ordering synchronization markers during connection rerouting to preserve the timing relationship between the user data and the synchronization markers. The inband signaling cell (Fig. 15) consists of the A T M cell header with the A T M cell payload containing the inband S A A L or application layer protocol header and payload. A s illustrated in Table 3, the inband signaling transport server identifies the inband synchronization marker cells via the cell header by the preassigned V P I and passes them to the P O S T P protocol entity or server. Once the P O S T P protocol entity detects out-of-sequence cells that require resequencing, it will request the cell synchronization manager to direct the user information cells to the resequencing buffer. When the synchronization period ends, the P O S T P protocol entity will direct the resequenced cells to the input of the switch fabric to continue their transit to the destination.  54  ATM Switch User Information, Inband & Associated Signaling Cells  Cell Switching Mechanism Switch Fabric  User Information, Inband & Associated Signaling Cells  VP CrossConnect V C CrossConnect  Cell Synchronization! Manager  Routing Control  8§  1 Forward  A$sociated-CCS Transport Server Source-Routing Unicast & Multicast Send  Associated Signaling Cells  Copy  RFRTP Protocol Entity  i 1  si  Inband Signaling Transport Server  Inband Signaling Cells  t °°py POSTP Protocol Entity Control Control  Resequencing Buffer .  User Information Cells  Figure 16 ATM Switch Functions  55  si  3.3 Performance Analysis 3.3.1 Reservation Transaction Delay with RFRTP At the reservation responder, the reception of the allocation status ( A L L O C A T I O N S U C C E S S or A L L O C A T I O N - F A I L U R E ) messages sent by the resource servers is modelled by using the concept of reception cycle (Figs. 11 and 12). The reception cycle starts after the reservation responder has received the R O B U S T - R E S E R V E message. The first cycle duration is bounded by the allocation control processing delay. Subsequent cycle duration is bounded by the local retry transaction delay. During each cycle, none or one or more allocation status messages from the resource servers will arrive at the reservation responder. The cycles end when all the allocation status messages have arrived. Let T u a  oc  be the average allocation control processing delay experienced by each re-  source server, T if ver  y  be the average verification control processing delay experienced by  the reservation responder, Transit be the average transit delay between two neighbouring nodes, T try be the average local retry transaction delay, T } re  cycle delay, T  cyc  reseTve  e  be the average reception  be the average reservation transaction delay.  Let M be the number of connecting nodes or resource servers along the predetermined connection route between the reservation initiator and responder, N i  be the average  cyc e  number of reception cycles, N  r o u t e  be the average number of local alternative routes  originated from a connecting node along the predetermined route, Nn ). be the average n  number of connecting links of an alternative route originated from a connecting node. The transit delay per link Transit composes of switching, transmission, queueing and propagation delays. The transit delay for high priority control packets in high-speed networks is dominated by the propagation delay. Let t be the propagation delay between p  56  two connecting nodes along the predetermined route, Tp be the end-to-end propagation delay between the source and destination nodes along the predetermined route. Therefore,  T  P  = MT•ransit . it =  (1)  Mt  t  p  At the end of the first reception cycle, up to M allocation status messages may have been sent by the resource servers. For those resource servers that do not respond, they would engage in local retry along alternative routes to their next immediate neighbouring nodes along the predetermined route.  We now derive a recursive relation for  P (M), r  the probability distribution that r reception cycles are required to service all M resource servers; i.e.  P\(M)  is the probability that the reservation responder receives all M  allocation status messages in the first reception cycle. Let f(k,j)  be the probability that fcout of j previous non-responding resource servers  fail again to respond in the current reception cycle.  Subsequently, each of the current  non-responding resource servers may require various number of reception cycles to send the allocation status message. Assume that reservation on the predetermined route or each alternative route between two neighbouring connecting switches is a Bernoulli event with failure probability Q and success probability 1-Q,  j  k  Q (i k  then  - Q)  (2)  Therefore, M  (3) k=\  The expected number of reception cycles is CO  (4)  57  Employing Eqn. (2) for induction on Eqn. (3), then  p (M)  =  r  (i-Q ) R  M  -(I-Q*- ) 1  M  (5)  Substituting Eqn. (5) into Eqn. (4) and noting that M  =  (1 - Q'f  M  £ k=0  M  Ncycle  M  ^^ k=l  (-1)  k  From Figs. 11 and 12 and that Tt t  = t,  ransi  (6)  (-Q y r  k  fc+1  (7)  l-Q  k  we can see that  p  Tretry — ^route (^verify  ^^link^p  (8)  ~f" -^a/Zoc)  and Treserve — T ify ver  -f- T n a  -\- 2,Mtp -\- {N y \  oc  C  C  l)T etry  e  (9)  r  Substituting Eqn. (8) into Eqn. (9), Preserve  ~  -f" 2Mtp -f-  Tu  T jjy  a  ver  oc  {^cycle  ^-)\_^route(Tverify  2A^/^ ^ip -f"  -\-  n  T n y\ a  oc  (10) =  [(Ncyde  -  l)N  + l](T if  route  ver  + T ii )  y  a  oc  + 2[M  +  (Ncyde  l)N Nu k]t  —  route  n  p  From Fig. 17, we can see that for link congestion and failure probability Q < 0.1 (which is already quite high for normal network operations), N i  cyc e  M increases; i.e. 1 < N [ cyc  is upper-bounded by 3 as  < 3. For practical operations of the retry transactions, it is  e  safe to assume that 1 < Nn k < 3 and 1 < N ute n  r0  < 4. Substituting these upper-bound  and lower-bound values into Eqn. (10), {Treserve) i  m n  reserve)  max  — T ify ver  9(T fy verl  + Tn a  oc  + 2Mtp  + TaUoc) + ( 2 M + 48)t  58  (11)  p  (12)  A © £3  A 0=0.2 © Q=0.1 ED Q-0.01  2.5  s  0.5  20.0 30.0 No. of Links  Figure 17 R F R T P Performance Analysis: Reception Cycles  3.3.2 Reservation Transaction Robustness with R F R T P Let M be the number of connecting nodes or resource servers along the predetermined connection route between the reservation initiator and responder, (N t ) rou e  of local alternative routes originated from the {Niink)ij the  D  e  i  be the number  i-th node along the predetermined route,  the number of connecting links of the j-th alternative route originated from  i-th node. Let (Preserve)i  3  be the probability of successful reservation along the j-th local alter-  native route originated from the i-th predetermined node, Z be the random variable that %  counts the total number of reservation trials up to and including the first success originated from the i-th node-pair. Then the cumulative distribution function (cdf) of Z is given as t  k=j-i PZi{j)  =  Preserve,o  Preserven (1  Preserve^ ) " i " ~ t ~ P r e s e r v e i j  Y\  Preserveik)-  Therefore, the probability of successful reservation for the whole connection is given as M °- = LI F {{Nrouu)i Zi  + l)-  59  For  a  numerical example,  {(Nnnk)ij} have  =  the  M  {(N  = 5,  )  ROUTE I=1  TO 5  ((1),(1,1),(1,1,2),(1,1,2,3),(1,1,2,3,4)).  same  {{P),(P,P),  let  probability  (p,P,P ),  of  routing  success  p,  thus  }  =  (0,1,2,3,4),  Assume  all  links  | ( Preserve ) i j |  =  (p,P,P ,P ), (p,P,P ,P ,P )) 2  2  3  2  3  4  Assuming reservation failure is due mainly to resource congestion rather than resource failure and p = 0.99 or l-p=0.01 for high congestion condition, the cumulative probabilities of reservation failure experienced by the resource server of a connecting node (employing the predetermined route and increments of alternative routes) are as follows: ( 1 0 ~ , 1 0 ~ , 2 * 1 0 , 6 * 1 0 , 2 * 1 0 ~ ) . Then the probability of reservation failure for 2  4  - 6  - 8  9  the whole connection is given as 1 — a = 10~ . Without the robust source-routing scheme, 8  the probability of connection routing failure becomes 1—0.99 = 5 * 1 0 . Therefore, even 5  -2  for a low probability of single-link routing success (i.e., heavy congestion of transport resource), increasing the number of available alternative routes per determined node-pair decreases significantly the probability of per-connection reservation failure. Fig. 18 shows an example of resource reservation along a predetermined route of 5 connecting nodes.  The top curve is associated with the non-retry scheme, i.e., any  single reservation failure between any node-pair would result in connection reservation failure. The other curves are associated with the localized retry scheme characterizing the proposed R F R T P , the connection reservation failure probability is decreased by increasing the available number of local alternative routes between a node-pair. For a 1 % per-link reservation failure probability, the connection reservation failure probability decreases from 5 * 1 0  - 2  to 1 0  - 8  by dimensioning 4 local alternative routes per any node-pair.  60  Figure 18 R F R T P Performance Analysis: Reservation Robustness  61  Chapter 4 Fast, Robust and Seamless Handoffs over ATM-Based P C N s This chapter illustrates that fast, robust and seamless handoffs over ATM-based P C N s can  be achieved by configuring the proposed generic real-time connection rerouting  service (as described in Chapter 3) to support mobile call handoff over the proposed M V C connection architecture (as described in Chapter 2).  Establishment of a M V C -  based handoff connection is dictated by the time-distributed connection  establishment  scheme which predetermines the connection route during call setup but delays real-time connection resource allocation until actual handoff occurs. Fast and robust connection resource allocation during handoff is supported by the proposed R F R T P .  Seamless  synchronized traffic rerouting from the current connection to the handoff connection is supported by the proposed intra-network P O S T P when the M T is capable of simultaneous access to multiple connections (e.g., when soft handoff is supported over the air interface), or by the proposed inter-network P O S T P when the M T is incapable of simultaneous access to multiple connections.  4.1 Example Scheme This section considers the mobile call handoff example with the M T incapable of simultaneous access to the current and handoff connections.  In configuring the real-  time connection rerouting service for the mobile call handoff environment, the following assumptions on signaling capabilities are made:  (1) the M T is capable of initiating  handoff; (2) the T P , B S and M T are capable of supporting the inter-network P O S T P ; (3) the T P , B S and A T M nodes are capable of supporting the R F R T P . 62  It is assumed that inband signaling and associated-CCS (source-routing unicast and multicast) are available to support the P O S T P and the R F R T P respectively.  Inband  signaling cells (synchronization markers) are transported with the same priority as the user information cells, and they are routed at the A T M layer via V C and V P crossconnection over the same G V C / V P links to transport the user information cells.  The associated  signaling cells are routed at the signaling A T M adaptation layer (above the A T M layer) via source-routing over the pre-assigned signaling V C / V P links.  The  fast, robust and seamless handoff initiated by the M T over duplex M V C is  shown in Fig. 19 and described as follows:  1.  The M T signals the T P via the old B S to switch M T ' s incoming traffic to the mobile links of the handoff connection. The T P is equipped with the source-routing information of the mobile links to the new B S during call initiation.  2.  The T P signals the new B S to assign a duplex radio channel for the handoff connection.  3.  The T P initiates robust and fast resource allocation along the predetermined handoff connection route (including the new BS), as elaborated in the following Section 4.1.1.  4.  After the T P receives the acknowledgment for successful resource allocation, the T P would initiate seamless synchronized traffic rerouting from the current or old connection to the newly established handoff connection.  The synchronization is  carried out by the T P , the old B S , the new B S and the M T , as described in the following Section 4.1.2, to prevent call disruptions due to A T M cell loss and sequencing errors.  63  Handoff-Initiating MT  TP  BS  Robust and ast Resource Reservation over the Predetermined Route of the Handoff Connection via RFRTP J..  T  Synchronized Traffic Rerouting via Inter-Network POSTP  0  Old BS (with ongoing connection)  #  New BS (with handoff connection)  CH  Mobile connecting points of the original connection (ATM switches)  1  Mobile connecting points of the handoff connection (ATM switches)  » ~ . ~ ~ $ g » .  •  Signaling within Air Interface Signaling within Backbone Network  Figure 19 Example of Fast Seamless Handoff Processing with MVC Connection  64  4.1.1 Fast Handoff Connection Establishment The real-time connection rerouting service dictates time-distributed connection establishment for M V C - b a s e d handoff connections. Routes for the potential handoff connections, which are terminated at the BSs of the neighbouring cells and converge at the T P within the ATM-based backbone, are predetermined during call setups without strict real-time requirement. Connection resource is not allocated along any of those routes until a handoff actually occurs.  Fast resource reservation for the M V C - b a s e d handoff  connection during handoff processing is supported by the proposed R F R T P (as described in Section 3.1.1), which dictates the involvement of the T P A T M switch, the B S and the connecting A T M switches between them. Employing the proposed R F R T P , the T P assumes the role of the reservation initiator of the source node, the B S assumes the role of the reservation responder of the destination node. The A T M switches between the T P and the B S assume the roles of the resource servers of the connecting nodes. The sourcerouting unicast and multicast of the associated-CCS transport mechanism as described in Section 3.2.1 would be employed for signaling transport. With references to Figs.  11,  12 and 19, the employment of the R F R T P during M V C handoff is described as follows:  1.  The reservation initiator of the T P signals an initial-commit reservation request message to the new B S over an associated signaling channel along the connecting switches of the mobile links via source-routing multicast.  2.  Upon receiving the initial-commit reservation message, each connecting switch: (1) forwards the signaling message to the next connecting switch along the route designated by the incoming source-routing information; and (2) taps the signaling message for processing by the reservation controller.  3.  Upon receiving the signaling message, the resource server of each connecting switch 65  initiates a statistical or deterministic reservation control algorithm. If the resource server accepts the request, appropriate resources are reserved.  Upon completion  of control processing, each resource server sends either a rejection or acceptance message to the reservation responder of the B S via source-routing unicast over the associated signaling channels.  The source-routing information is encoded in the  reservation request message. 4.  The reservation responder of the B S signals a final-commit request message to the T P over the associated signaling channels along the connecting switches of the M V C mobile links via source-routing multicast.  5.  Upon receiving the final-commit reservation message, each connecting switch must either commit or release the previously reserved resources.  Fig. 11 illustrates the timing diagram of the fast resource reservation scheme without activating the robust adaptation of the associated signaling transport. Fig. 12 illustrates the timing diagrams of the fast resource reservation scheme when the robust adaptation of the associated signaling transport is used to bypass congested or faulty links.  4.1.2 Seamless Synchronized Traffic Rerouting The proposed inter-network P O S T P (as described in Section 3.1.2) is employed to provide the seamless or disruption free handoff capability (i.e., no data loss or sequencing errors) to M V C handoff during traffic rerouting. With reference to Fig. 19, the employment of the inter-network P O S T P is described as follows:  1.  The T P signals the old B S that the M T ' s incoming traffic via the original connection is now terminated by issuing an incoming sequence termination marker via the inband signaling transport mechanism proposed in Section 3.2.2. 66  2. The TP signals the new BS that the M T ' s incoming traffic via the handoff connection is now activated by issuing an incoming sequence beginning marker via the inband signaling transport mechanism. 3.  Upon receiving the incoming sequence termination marker, the old BS signals the M T to switch to the new BS.  4.  Upon receiving the incoming sequence beginning marker, the new BS buffers the M T ' s incoming traffic.  5.  Before switching to the new BS, the M T signals the TP via the old BS that the M T ' s outgoing traffic via the original connection is now terminated by issuing an outgoing sequence termination marker via the inband signaling transport mechanism.  6.  The M T dissociates from the old BS.  7. The M T signals the new BS to activate the assigned radio channels of the duplex handoff connections. The new BS then presents the incoming buffered data to the MT. 8. The M T signals the TP via the new BS that the M T ' s outgoing traffic via the handoff connection is now activated by issuing an outgoing sequence beginning marker via the inband signaling transport mechanism. 9.  The TP synchronizes the M T ' s outgoing traffic via the old and the new mobile links by employing the outgoing sequence termination and beginning markers, and buffering the outgoing traffic on the new link as necessary.  67  4.2 Performance Analysis 4.2.1 Processing Delays The time delay performance parameters characterizing the seamless handoff example include: (1) the resequencing delay, (2) the incoming sequence synchronization buffering delay, and (3) the handoff processing delay. Resequencing is employed to eliminate packet sequence disordering due to the instantaneous mismatch of packet transit time between the original and the handoff connection at the handoff epoch. Incoming sequence buffering is employed to eliminate call disruption due to the inability of the handoffinitiating M T to associate with the original and the handoff connection simultaneously. The handoff processing delay determines the overall cell delay jitter. Let TQ be the signaling transit delay between the BS and the TP, T^ T  M  be the  signaling transit delay between the BS and the M T , T^ be the user information transit T  delay of the synchronization marker between the BS and the TP, T^  M  information transit delay between the BS and the M T , Transit  D e  be the user  the average signaling  or user information transit delay via a mobile connecting link of the M V C , T  reserve  be  the resource reservation or allocation delay, Mo be the number of mobile connecting links of the original connection, and MM be the number of mobile connecting links of the handoff connection. Switching from the original connection to the handoff connection, the resulting difference or mismatch in cell transmit times would generate either a lagging gap (positive difference) or packet-disordering (negative difference).  The seamless handoff scheme  eliminates packet-disordering by introducing resequencing delay  TRSQ-  The cell transit  time consists of the deterministic delays (e.g. transmission, propagation and switching 68  delays) and the stochastic queueing delays. The queueing mechanisms at the output ports of the connecting nodes form a network of queues. Consequently, the difference in cell transit times has a deterministic delay difference component and a stochastic queueing delay difference component. However, the transit delay T t  r a n a i i  for high priority control  packets in high-speed networks is dominated by the per-link propagation delay t . Under p  this situation, TRSQ becomes a deterministic quantity and  TRSQ = (Mo - M )t n  (13)  p  With reference to Fig. 19, the inter-BS incoming sequence synchronization buffer delay on the handoff-initiating side is given by:  TBUF  = T§  M  + TRSQ  (14)  The average handoff processing time is determined as: Thandoff ~ BT + T erve + % BT + ^ BM T  T  + BM + %TRSQ  T  T  reS  (15) = (Mo + 2Mfl)Ttransit  + T rve rese  + % BM + BM + ^RSQ T  T  Derived as the reservation delay dictated by the R F R T P , T  is obtained by substi-  reserve  tuting M by Mg in Eqn. (10) of Section 3.3.1, Trese rve —  T  ver  ify  where T \\ a  oc  + T Uoc a  +  1M t  + (Ncyde  H v  -  l)[N  r o u i e  (T  v e r i  fy  2N t  +  Unk v  + T Uoc)] a  (16)  is the average allocation control processing delay experienced by each  resource server, T f veri  y  is the average verification control processing delay experienced  by the reservation responder, N t rou  e  is the average number of local alternative routes  between two neighbouring nodes of the predetermined reservation route, Nu k n  average number of connecting links of an alternative route, N [ cyc  e  is given by Eqn. (7)  of Section 3.3.1 with M being substituted by Mg and Q being substituted by Bg  H, 69  is the  where Bg is the probability of handoff resource blocking along an alternative route. Replacing T  by t  transit  Mo = M , then  »  reserve  =  T  H  Nroute,Nunk,  in Eqn. (15) and assuming T  p  handoff  M~H,Bff, i.e.,  3M t  + T  H p  Therefore, T  .  reserve  Ti ioff  M  + T%  M  and  is a function of  handoff  f{NrouteNunk,M Bff).  =  lan(  2 r |  H  The upper-bound of the reservation delay along the handoff connection is given by Eqn. (12) in Section 3.3.1 with M being substituted by M # :  < 0.1  (18)  Substituting Eqns. (13) and (18) into Eqns. (14) and (15); and replacing Transit  by t ,  (Treserve)  max  = 9(T  + T  v e r i f y  a l l o c  ) +  (2M + 48)t , H  when  p  B  ff  p  then  = (Mo - M )t  (TBUF)  H  max  = 9(T  (T ndoff) ha  max  Therefore, (TBUF) ax  c  v e r i f y  a  m  n  + TaUoc) + ( 3 M  0  +  + r |  p  2M  ff  (19)  M  + 48)t + 2 r | + p  M  r  V BM  (20)  be minimized by selecting Mo and MJJ as close as possible.  On the other hand, (Thandoff)  c max  a  n  be minimized by decreasing the number of mobile  connecting links; or by selecting the T P to be as close as possible to the mobile endpoints. In the extreme case where the T P of the M V C connection is at the destination endpoint (i.e., maximum mobile connecting links no common connecting links), upper bounds of the delays result.  Factors Affecting Handoff Processing Delay  This section examines how the handoff  processing delay is affected by the M V C connection configuration and the resource reservation performance parameters via R F R T P . Fig. 20 shows Thandoff  as a function of N ute  Nunk = 1. It shows that Thandoff  r0  for varying MJJ when BR - 0.005 and  is directly proportional to N te rou  with a proportionality  factor of 0.02 (per-link propagation delays/route), regardless of Mff.  70  Bj_|= 0.005 =g  200.0  M  ;G—0  "D  •  c  •  H  =  Niink = 1  mobile link  1  M |_| = 5 mobile links  O—O  M  H  = 10 mobile links  A"—A  M  H  = 20 mobile links  M H = 30 mobile links  _ro a> Q  150.0  f  100.0  £  co  as  Q_ O  -£  -A-  &  -A  A-  A  (D D_ O  50.0 a> Q  -B  LP"  o "a c  ©  CO  X  0.0  -4  -0-  0  0  e1  B  - a  Eh  -Q-  -e  2  3  e4  B  •  B  B-  -Q-  -G  e  e  ©-  -e-  5  6  7  N o . of L o c a l Alternative R o u t e s  8  -0  9  10  Nroute  Figure 20 Effect on Handoff Processing Delay due to M V C Connection Configuration Fig. 21 shows Tkandoff as and Nn k n  MJJ  = 10  = 1. It shows that the direct proportionality factor between Th ioff  and  Nroute is non-linearly  function of  a  N  ROUTE  for varying Bg when  an(  proportional to Bff for a given  MJJ  and Nu k. n  For example, when  Bff increases from 0.005 to 0.05 and 0.5, the direct proportionality factor is increased respectively from 0.02 to 0.109 and 0.774. Fig. 22 shows Thandoff as a function of N ute for varying Nu k when Mg = 10 and r0  BE  n  = 0.005. It shows that the direct proportionality factor between Th d ff an  0  and  N t rou  e  is non-linearly proportional to Nn k for a given Mff and Bff. For example, when Nn k n  n  increases from 1 to 2 and 3, the direct proportionality factor is increased respectively from 0.02 to 0.058 and 0.114. 71  _C0  Q Q O  c CO - 1  Q  g  i  L  0  1  i  i  2  3  i  4  i  i  i  5  6  7  i  8  i  9  10  Nroute  N o . of Alternative R o u t e s  Figure 21 Effect on Handoff Processing Delay due to Targeted per-link Handoff Call Blocking Probability The R F R T P bypasses link failures or congestion during reservation via localized retry over alternative routes between two neighbouring nodes along the predetermined reservation route.  The reservation robustness can be enhanced and the handoff delay  can be minimized by minimizing ''Mg, BE and Nu k. n  the number of alternative routes N  r o u t e  would enhance the reservation robustness but at  the expense of increasing the handoff delay T^andoff22 show that Tk doff an  BE  O n the other hand, increasing  increases slightly as N  r o u t e  F°  r a  given ME, Figs. 21 and  increases providing that the targeted  is not greater than 0.015 and the Nn k is not more than 2. Under this situation, n  Nroute can be increased to maximize reservation robustness, and it would not cause a significant increase in  T} d ffian  0  72  o  M|_|= 10  F31—|  = 0.005  200.0  "O c CO  G--©N -Q  l i n k  = 1  N  = 2  N n k  0 - -O N|ink = 3 = 4 A— - A N | <3- —<l Nlink = 5 l n k  150.0 _ro  V - -V N| ~ > N|  CD Q  c o ro cn ro a. o  >-  i n k  j n k  = 6  = 7  100.0  CD D_  O  b  50.0  CD  a "o c «  0.0  0  1  2  3  4  5  N o . of Alternative R o u t e s  6  7  8  10  NrrjutG  Figure 22 Effect on Handoff Processing Delay due to Reservation over Alternative Routes  4.2.2 Effect of Handoff Processing Delay on Unnecessary Handoff Probability Due to signal reception fading in the mobile cellular environment, the handoff process must be initiated at a margin well before the received signal between the M T and the current B S falls below the receiver threshold level.  Increasing the initiation margin  would decrease the probability of forced call terminations but it would also increase the probability of unnecessary handoffs.  To avoid forced call termination due to handoff  failure, the handoff processing delay must not exceed the initiation margin.  For a given established initiation margin as illustrated in Fig. 23, intelligent decision algorithm for handoff initiation tries to minimize the number of unnecessary 73  handoffs  by averaging received signal values and employing hysteresis.  The intelligent decision  process would incur detection delay which should be maximized to minimize the probability of unnecessary handoffs.  However, increasing the detection delay would also  decrease the allowable time for handoff processing. Thus, it is important to minimize the actual handoff processing delay to allow optimal minimization of unnecessary handoffs while maintaining the required probability of forced call terminations.  Precautionary Handoff Threshold  Channel Received Signal Strength  Mandatory Handoff Threshold  Time  Initiation Margin  Initiation Margin = Handoff Processing Delay + Handoff Detection Delay Figure 23 Handoff Initiation Margin  Handoff across an ATM-based P C N would incur handoff processing delays across the fixed A T M network and the mobile subnet. 74  Let T gin mar  be the initiation margin,  (TATM)max  ^  an<  (Tsubnet)  b  max  the actual maximum or worst-case handoff processing  e  delay across the fixed A T M network and the mobile subnet respectively, T^etection be the handoff detection delay.  To avoid forced call terminations,  Trnargin >  (TATM) max  (Tsubnet)  (21)  Tdetection  max  With the mobile call handoff example across the fixed A T M network, (TATM)  is  max  given by Eqn.  (20).  For simplification, assume  ( 3 M + 2M + 48)t » 0  H  2T  (T TM) A  + T  S  p  U  BM  BM  = 9(T r y  MAX  ve  lf  Mg — Mo, and 9(T if ver  +  y  Tu ) + a  0C  then  +  T  ) + (5M  aUoc  with the connecting path propagation delay given by  H  + 48)t  (22)  p  Tp = Mgt  p  In [22], the control processing delay for their proposed bandwidth reservation control  8600)  algorithm (run on a V A X  controller and 3.18 ms (i.e. T f ) veri  delay for a  10 km  is found to be  3.02 ms  (i.e.  T u ) for the connecting a  oc  for the destination controller. In [21], the propagation  y  metropolitan path is assumed to be on the order of  0.05 ms;  km coast-to-coast path around 25 ms. Therefore, for a metropolitan path:  = 58.45  ms for  Mg = 1 (worst case) and (TATM)  max  =  coast-to-coast path: (TATM)  max  1380.8 ms  for  =  56.2 ms  for  5000  for a  {TATM)  max  Mg = 16. For a  Mg = 1 (worst case) and  (T TM) x A  ma  = 255.8 ms for Mg = 16. To give some examples of initiation margins established  under standard mobile  cellular conditions, we will reference the parameters in page 101 of [36].  It is assumed  that the mandatory handoff threshold level is set at -100 d B m (decibels over l m W ) at the cell boundary at which a handoff must be taken, and the precautionary handoff threshold level is set at -95 d B m . The M T is assumed to be roaming at a constant velocity of 50 miles/hour along a straight path between two transmitting antennas of neighbouring cells. 75  Then based on the propagation path loss figure in page 102 of [36], the T  m a r  g  should  i n  be set at 180 seconds for a suburban area with a mobile path loss of 38.4 dB/decade, and at 36 seconds for New York City with a mobile path loss of 48 dB/decade. Thus, we can see that with the proposed handoff scheme over an A T M / B - I S D N P C N , in the worst case, (TATM)  consumes less than 0.77% and 3.8% of the above established  MAX  Tmargin  of 180 seconds and 36 seconds respectively for a coast-to-coast path. Assuming ( T can see that T  s u b n e t  d e t e c t  i  o n  )  m a x  is of the same order of magnitude of  ( T A T M )  m  a  x  ,  can be optimized almost up to the limit of the established  t  h  T  e  n  m a r g  w  e  i . n  According to [37] with a hysteresis margin of 10 d B , when T^uction  is allowed to operate  at 36 seconds, the probability of unnecessary handoff is around 1 0  - 4  around 10~  9  for microcells as illustrated in Table 4.  in macrocells and  When Tdetection is allowed to  operate at 180 seconds, the probability of unnecessary handoff is negligible for both macrocells and microcells.  Therefore, for a given intelligent decision algorithm and a  given established initiation margin, optimizing the handoff delay across networks would minimize the probability of unnecessary handoff while maintaining the probability of forced call terminations. Handoff Processing Delay  Probability of Unnecessary Handoff  (worst case)  Macrocell  Microcell  36 sec  58.45 ms  io-  IO"  GJrban Area)  (Metropolitan Path)  Initiation Margin  4  9  1380.8 ms (Coast-to-Coast Path) 180 sec  58.45 ms  (Suburban Area)  (Metropolitan Path)  10"  4  1380.8 ms (Coast-to-Coast Path) Table 4 Probability of Unnecessary Handoff 76  io-  9  B y employing the proposed reservation and synchronization transaction protocols, the mobile call handoff example results in handoff processing delay that uses up a very small portion of the initiation margin, allowing the handoff detection delay to maximize almost up to the initiation margin. Consequently, the probability of unnecessary handoffs is minimized.  77  Chapter 5 Adaptive Connection Resource Management Architecture to Provision Call Level Performances over an ATM-based PCN In future P C N s supporting network-wide handoffs, new and handoff requests will compete for connection resources in both the mobile and backbone networks.  Forced  call terminations due to handoff call blocking are generally more objectionable than new call blocking. The previously proposed guard channel scheme for radio channel allocation in cellular mobile networks reduces handoff call blocking probability substantially at the expense of slight increases in new call blocking probability by giving resource access priority to handoff calls over new calls in call admission control. While the effectiveness of a fixed number of guard channels has been demonstrated under stationary traffic conditions, with non-stationary call arrival rates in a practical system, the  achieved  handoff call blocking probability may deviate significantly from the desired objective. We propose a novel dynamic guard channel scheme which adapts the number of guard channels in each cell according to the current estimate of the handoff call arrival rate derived from the current number of ongoing calls in neighbouring cells and the mobility pattern, so as to keep the handoff call blocking probability close to the targeted objective while constraining the new call blocking probability to be below a given level.  The  recovery control for either handoff or new call blocking in the cellular mobile network is based on the blocked-calls-dropped discipline. The proposed dynamic guard channel scheme is applicable to channel allocation over cellular mobile networks, and is extended to bandwidth allocation over the backbone network to enable a unified approach to prioritized resource access over the ATM-based P C N . The recovery control for handoff call blocking in the ATM-based backbone network  78  is based on the robust reservation transaction of the proposed R F R T P (presented in Section 3.1.1) via localized retry over alternative reservation routes.  5.1 Dynamic Guard Channel Scheme in Cellular Mobile Networks The fixed guard channel scheme is a prioritized resource access scheme which allows new calls and handoff calls to share capacity while giving resource access priority to handoff calls by assigning a larger resource capacity limit to handoff calls than new calls, with the guard channels being the difference between the capacity limits. Thus call admission control is based on the current bandwidth usage and respective assigned capacity limit of each call class.  Although in this thesis we only consider  the congestion or blocking recovery control in the cellular mobile networks based on the blocked-calls-dropped discipline, the resource access priority for handoff calls may be further increased by employing the blocked-calls-queued discipline.  With shared  channel resources, queueing the blocked handoff calls only would increase the blocking probability of the new calls. For a given stationary call arrival rate, the targeted longterm congestion related performance (i.e., blocking probability) for handoff calls can be satisfied by choosing an appropriate number of fixed guard channels. In general, given the total channel resource that may be allocated to the new and handoff calls, blocking occurs during call admission control when the call requires bandwidth over the radio channel in a cell or the links traversed over the backbone network in excess of what is available. In the ATM-based backbone, admitting the call under such condition would degrade the connection performance of all calls sharing capacity with the setup call.  Without prioritized resource allocation, handoff and new  calls would have the same blocking probability. However, handoff calls can experience a more favorable blocking probability than new calls by prioritizing resource allocation 79  during call admission using the guard channel scheme: with Na guard channels in a pool of NT channels where NT > NG, a handoff call is allocated any free channel while a new call is allocated a channel only when the number of free channels is greater than No-  It has been shown [28] that under stationary traffic, even for a small value of NG,  handoff call blocking probability is substantially reduced at the expense of only a slight increase in the new call blocking probability.  The proposed dynamic guard channel scheme adapts the number of guard channels according to the current estimate of the instantaneous handoff call arrival rate so as to keep the handoff call blocking probability close to the targeted objective while not deteriorating the new call blocking significantly.  The proposed scheme can be implemented in the  BSs in a mobile network or the bandwidth controller of each wired link in the backbone network.  A n implementation model of the proposed scheme is illustrated in Fig. 24.  In addition to the admission and congestion control functions as required by the fixed scheme, the proposed scheme also requires the instantaneous handoff call arrival rate estimation and capacity limits adaptation functions described below.  The arrival processes of new and handoffs calls at a B S or link bandwidth controller are assumed to be Poisson with time-varying rates of X^(t)  and A # ( £ ) respectively. The  call departure process is assumed to be Poisson with a constant rate of ji. The congestionrelated performance parameters at time t are the new call blocking probability B^(t)  and  the handoff call blocking probability Bg(t).  NT,  the number of guard channels at time  For a given total number of channels  t, Nc(t), is employed as a control parameter which  can be set within the range 0 < Nc(t)  < NG  max  with NG  max  < NT.  Assume that changes in the arrival rates occur at a moderate rate, and the time required to reach the steady state after a change is short compared to the time to the 80  Instantaneous Handoff Call Arrival Rate Estimation  Potential Handoff Call Arrivals Potential Handoff Call Releases  Capacity Limit Adaptation  N Q = Adapted Number of Guard Channels  Handoff Call Admission  Handoff Call Arrivals  Acceptance (No. of Free Channels  o o  CD  c  c  co  Blocking  o  New Call  New Call Arrivals  Admission  Acceptance (No. of Free Channels  <  Busy Channel (0 < N w  G  <N  Q  75 +—> o  o  Free Channel  O )  d Z  O  Blocking  #  O o  where,  N  "max  G  = PN  T  and p <  1  I  "max  Figure 24 Dynamic Guard Channel Scheme next change in the arrival rate, then N  a(t) Mi) P (t)^- (l-r(t)) T  B (t) B  =  r  0  (23)  (N -N (i)-1) T  B (t) N  =  a  (24)  l-P {t) 0  i=0  81  where a(t) = [Ajv(t) + A#(i)]/^, r(t) — Xj\f(t)/a(t)fj,, and P (t) is the probability that 0  all channels are unoccupied at time t: ~ N  T  - N  G  ( t )  i,^  N  T  (25) 11 i=0  i=N -N {t)+1 T  G  5.1.1 Instantaneous Handoff Call Arrival Rate Estimation The  instantaneous handoff call arrival rate at a test cell for the next immediate  estimation interval depends on the followings: the number of active M T s with ongoing calls in the neighbouring cells, the mobility patterns of the active M T s in terms of speed and direction during the estimation interval, the sizes of the cells currently resided by the active M T s and the remaining call durations of the ongoing calls. Let P#; • (&) be the probability of handoff call arrival at the test cell j due to an active M T in the neighbouring cell / at the k-th estimation interval, FT . (t) be the cdf of the H  channel holding time in the neighbouring cell i associated with the active M T (the channel holding time depends on the cell residing time and the remaining call duration), r be the estimation interval length, FT .(^)  be the cdf of the unencumbered call duration (i.e., the  C  time an assigned channel would be held if no handoff is required) in the neighbouring cell i, Pxi-*j(k) be the typical transit probability of an active M T from the neighbouring cell i into the test cell j given a call handoff occurs at the k-th estimation interval, A#(fc) be the call handoff arrival rate at the test cell at the k-th estimation interval, Mi be the number of active M T s in the neighbouring cell i, and N be the number of neighbouring cells. Then,  JW*)  = T ,(r){l  E Ps^iWi A^(fc) = —  - F (r))P ^.(k)  F  H  EF  Xi  (r)(l  -  THt  Tci  = —  F (r))P ^(k)Mi Tci  r  T  82  (26)  Xi  (27)  If FT (t) and FT {F) h  C  are assumed to be exponential, and Pxi^j(k) changes slowly  relative to the estimation interval, then Pff^^k)  can be carried forward to the next  (k+])-th estimation interval if the associated call is still active in the neighbouring cell. Instead of exchanging all call state or active M T information among neighbouring BSs regularly after each estimation interval, each B S can just signal changes in the call state to the neighbouring cells for updating A#(&). Then the update of A_#•(&) would become a simple exercise of addition and subtraction for respective increase and decrease of the number of active M T s or ongoing calls in the neighbouring cells.  A s illustrated in  Fig. 25, the number of active M T s or ongoing calls in a cell changes at the following events: call completion, call handoff, new call arrival and handoff call arrival.  Upon  each call completion or call handoff, the B S would signal a decrease in the call state to the neighbouring cells. Upon each new call arrival or handoff call arrival, the B S would signal an increase in the call state to the neighbouring cells. After obtaining A# at an estimation interval , the number of guard channels NG that is required to satisfy a given target objective for either B^ via Eqns. (24) and (23) respectively. targeted B^  or Bg can be determined  Fig. 26 illustrates the required NQ to maintain a  or Bg as A# varies for a given A^r.  5.1.2 Capacity Limit or Guard Bandwidth Adaptation Adaptation control policy can be formulated under a strict or a relaxed criterion. A strict criterion would be to maintain the handoff call blocking probability at the targeted objective at all cost (hard target) by compromising the new call blocking probability indefinitely. A relaxed criterion would be to keep the handoff call blocking probability close to the targeted objective (soft target) while ensuring that new calls are not denied of resource access completely  at any given time. 83  To support the relaxed criterion,  Potential Handoff Call Arrivals in Test Cell: New Call Arrivals and Incoming Call Handoffs in Neighbouring Cells. Potential Handoff Call Releases in Test Cell: Call Completions and Outgoing Call Handoffs in Neighbouring Cells. Figure 25 Estimation of Instantaneous Handoff Call Arrival Rate at a B S  the proposed adaptation control policy allows deviation of the handoff call blocking probability from the targeted objective and restricts the guard bandwidth not to take up the whole resource Let Bpf  target  capacity.  be the targeted objective for new call blocking probability, Bg  hard  84  and  50.0  N X  z  40.0  in  T  = 50  = 0.1/sec,  targeted B  X  N  = 0.2/sec,  targeted B  X  N  = 0.333 sec, targeted B  H  = 0.005  X  N  = 0.1/sec,  N  = 0.05  N  = 0.005  H  = 0.005  H  Q)  c c CO  JZ  O  targeted B  N = 0.2/sec,  30.0  X  T3  ico  targeted B ^ = 0.05  = 0.333 sec, targeted B  N  N  = 0.05  3  o  "o c> z  20.0  TJ  d) i—  10.0  'D (X  d> DC 0.0 0.00  0.10  0.30  0.20  Estimated Handoff Call Arrival Rate Figure 26 Number of Guard Channels Required for Targeted Blocking Probabilities  Bj{  soft  be the hard and soft targeted objectives, respectively, for handoff call blocking  probability, where Bg  = aBu  soJt  with a > 1. A t a given time instant t, Ayy(i) can  hard  be obtained by either ongoing local monitoring or by setting to a time-independent value if stationarity is assumed, fx is assumed to be time-independent, Ajy(t) is estimated by Eqn. (27) based on call state information obtained from neighbouring cells. NN  Let  be  targcl  the  BN(XN{t),X {t),n,N (t)) H  Nu  <  G  respectively  soft  B  value  <  G  B  values  G  i V c ^ ) can  that  to  assume  E q n . (24),  Nc(t)  H  < N  N  H  G  and N  Gmax  < N.  Gmax  T  The proposed policy is described as follows:  1.  Determine N  2-  If N  ,  N  Ntargat  Hhard  <N  Hhard  ,  Ntarget  and N  via Eqns. (24) and (23).  Hso!t  then N (t) = N G  ,  Hhard  85  else N (t) G  =  to  N  Hsojt  satisfy  N  can assume  and B {\ {t),\ {t)^,N {t))  H h a r d  according to Eqn. (23), where 0 < N  that  according  N t a r g e t  be the smallest  B (*N(t)^ff{t),H,N (t)) H  greatest  and  Hhard  to  satisfy  <  B  H s o f t  A strict control criterion to maintain handoff call blocking probability at a targeted B  H  objective can be realized by setting a =  „  g o / t  =1 and Nc  max  criteria can be realized by increasing a and decreasing Nc , max  = NT-  Relaxed control  and the limiting case  would correspond to the strict control criterion to maintain new call blocking probability at a targeted  objective.  5.1.3 Performance Analysis We simulate the operations of the proposed dynamic guard bandwidth scheme in a mobile wireless system with cells characterized by diverse traffic characteristics. Fig. 27 shows the cell coverage over downtown, city areas, suburbs and rural areas configured as follows: (1) the downtown cell is surrounded by six city area cells; (2) each city area cell is surrounded by the downtown cell, two city area cells and three suburbs cells; (3) each suburb cell is surrounded by one city area cell, two suburb cells and one rural area cell. Performance results obtained from a city cell will be illustrated and discussed.  Rural Area  Rural Area Figure 27 Simulation Model 86  As shown in Fig. 25, when a new call is originated in a cell and assigned a channel, the call holds the channel until the call is completed in the cell or the call is handed off to another cell as the mobile moves out of the cell. The probability of requiring a handoff PH depends on the cell coverage area, the M T ' s movement and the call duration. A call handoff must be directed to one of the neighbouring cells. The probability of each neighbouring cell receiving the call handoff depends on the amount of common boundary area and the mobile directional pattern.  Call Blocking Probabilities Under Stationary Traffic  The condition with stationary  traffic is simulated with the following system parameters: the new call and handoff call inter-arrival times are assumed to be exponential and stationary with mean new call arrival rate A# and Ay respectively, the call duration is assumed to be exponential with mean TQ = 150 seconds, the channel holding time is assumed to be exponential with mean T# = 120 seconds and mean service rate fx, the total number of channels Nj> - 50 channels, the adapted number of guard channels at estimation interval k, Nc(k), within the range 0 < NQ < Nc  max  new call blocking probability blocking probability  B{j  hard  with Nc  B^  target  = 0.005,  = NT = 50, the targeted objective of  max  - 0.05,  is set  the hard targeted objective of handoff call  the transit probability of an active M T from one  cell to another cell Px depends on the time of day. There is a greater tendency to transit towards the downtown cell during morning traffic rush hours, and to transit outwards away form the downtown cell during evening traffic rush hours. For example, during morning traffic rush hours,  Px  = 0.1, city-to-downtown  0.7, city-to-suburb  to-city =  =  is set as follows: downtown-to-city =  0.5, suburb-to-rural = 0.167,  =  0.1, suburb-to-suburb  rural-to-rural =  87  0.167,  0.5, and  =  city-to-city  0.167,  suburb-  rural-to-suburb =  0.167.  Long-term handoff call and new call blocking probabilities B^  and Bg  are obtained for  a 120 hours simulation time duration. A reference scheme with a fixed guard size of 2 channels is also employed.  Com-  paring the fixed guard scheme and the proposed dynamic guard scheme, Figs. 28 and 29 show Bjq and Bff  respectively as functions of Pff for different values of Xjy, whereas  Figs. 30 and 31 show B^  and Bff respectively as functions of Xjy for different values of  Pff, where Pff is the probability that an ongoing call will require at least one more handoff before completion. When resource access priority (via the number of guard channels) given to handoff calls is adapted according to the instantaneous handoff call arrival rate, Bff  decreases significantly with only negligible or small increases in  B^.  Handoff Prob. P,  Figure 28 Effect of Handoff Probability on New Call Blocking Probability  Let RN be the ratio of B^ the ratio of Bff  under dynamic guard to that under fixed guard, RJJ be  under fixed guard to that under dynamic guard. We define adaptive gain 88  Handoff Prob. P.  Figure 29 Effect of Handoff Probability on Handoff Call Blocking Probability  0.10  0.15  0.20 0.25 New Call Arrival Rate (call/sec)  0.30  Figure 30 Effect of New Call Arrival Rate on New Call Blocking Probability  GA,  which measures the relative trade-off between the decrease in Bg 89  and the increase  10 Target Objective  10'  m  o CL  cn c !2 o  o  CO  10  « O a: O T3 C « X  Fixed Guard Dynamic Guard  10'  10'  0.10  0.15  0.20 0.25 New Call Arrival Rate  0.30 >- (call/sec) N  Figure 31 Effect of New Call Arrival Rate on Handoff Call Blocking Probability in Bflf, as GA =  RH/R-N-  It is imperative that GA  fixed guard to dynamic guard, i.e., an increase in BN larger decrease in By.  >  1 to justify the switch from  is compensated with a relative  Based on the simulation results, Fig. 32 shows the probability  distribution function of GA and it can be seen that 1 < GA < 3 with a mean gain of about 1.3. A s dropped calls due to handoff call blocking are likely to be more costly to the service provider than blocked calls, in terms of customer dissatisfaction, the average 30% adaptive gain attributed to the proposed dynamic guard scheme may translate to a more substantial cost saving for the service provider. Fig.  33  shows B^  (Xff + XJ\[)/NTP 1.2).  and By  as functions  of the channel loading factor p  =  which is varied from a medium level (p = 0.5) to overloading (p =  It can be seen that Bps resulting from the fixed guard scheme and the proposed  dynamic guard schemes are very similar under both medium and heavy loadings. the other hand, By  On  resulting from the dynamic guard scheme is generally lower than 90  0.25  =g 0.20 to b  ra 0.15 O  0.10  0.05  0.00  1.0  1.2  1.4  m i l i •• I i • i 1.6  1.8 2.0 2.2 Adaptive Gain  2.4  2.6  2.8  3.0  Figure 32 Adaptive Gain  10 10  10"  I  10'  3  (0  xi o  ? 10"  o  m 10'  G * • O  10'  10."  0.50  0.60  0.70  0 * • O  B B B B  N  N  H  H  (Fixed Guard) (Dynamic Guard) (Fixed Guard) (Dynamic Guard)  0.80  0.90  1.00  1.10  Channel Loading Factor p  Figure 33 Effect of Channel Loading Factor on Blocking Probabilities that resulting from the fixed guard scheme. The difference in By  decreases as loading  increases, and becomes negligible under heavy loading or overloading. When a call handoff results in connection rerouting over the backbone network, the 91  overall blocking probability increases as the number of rerouting links is increased. Let L be the number of links, including the wireless link and the wired links in the logical virtual path. Assume all links have the same bandwidth capacity and are subjected to similar A# and AH due to fixed call routing. Let Bjj  con  call blocking; Bj^  con  be the probability of per-connection new call blocking. Then,  B  Hcon  B con  and Bjy  con  = l - ( l - B  f  H  (28)  = 1 - (1 - B )  (29)  L  Ncon  Fig. 34 shows Bff  be the probability of per-connection handoff  N  as functions of L when XN = 0.333 calls/second and p  = 0.85. To maintain the per-connection handoff call blocking probability at the targeted objective of 0.005, one wireless link and one wired link can be accommodated under the fixed guard scheme, whereas one wireless link and 4 wired links can be accommodated under the dynamic guard scheme.  This shows that the dynamic guard scheme allows  the handoff call blocking probability objective to be met over a network with a greater diameter than the fixed guard scheme, without reducing the loading of the links as would be required for the fixed guard scheme to meet the above objective.  Call Blocking Probabilities Under Non-stationary Traffic  Actual traffic and the as-  sociated call arrival rate are seldom stationary or have the same level as the nominal, as new call arrival rate changes with the hours of the day and handoff call arrival rates depend on the number and movements of callers in adjacent cells. Based on the simulation model as specified previously, non-stationary traffic condition is introduced by increasing the average new call arrival rate during morning and afternoon traffic rush hours from the nominal rate. Fig. 35 shows the instantaneous (averaged over 20 minute time segments) and running average values (averaged from time 0) of the new and handoff call blocking probabilities 92  0  10 20 Number of Connecting Links L  30  Figure 34 Per Connection Blocking Probabilities over a 24 hours duration, with the average new call arrival rate increasing by 3 times to 0.6 call/second during rush hours from the nominal rate of 0.2 call/second. Average handoff call arrival rate would also increase as average new call arrival rate increases. Under the fixed guard scheme, the handoff call blocking probability increases to 0.015 from the targeted objective of 0.005 while the new call blocking probability increases to 0.25 from the targeted objective of 0.05. In contrast, under the proposed dynamic guard scheme, the handoff call blocking probability stays close to the targeted objective at 0.0033, which is substantially lower than the corresponding value for the fixed guard scheme, while the new call blocking probability deviates from the targeted objective by increasing to 0.265, 6% higher than the corresponding value for the fixed guard scheme. By varying the deviation from the nominal rate and keeping other parameters constant, Fig. 36 shows the call blocking probabilities as functions of the ratio of deviation of new call arrival rates from the nominal rate of 0.2 call/second. It shows that the proposed 93  ca JQ  o  al  o o m ca O *o -a ca  <x>  0  3.0  J3.0  6.0  1 2 . 0  1 8 . 0  21 .0  Time of Day (h«  « * a>  oo ca oo a> ca  1 5 .0  3ur)  "ca  0.20  r  ,  0.60  0.60  r  0.20 ^  0.20  a>  ca .a o  n n i n g A v e r a g e f o r f i> e d  Guard  nning A v e r a g e for D y n a m i c  Guard  o _o m "co O  Instantaneous Averages  12.0  9.0  15.0  1 8.0  21.0  Time of Day (hour)  Figure 35 A n example of Non-Stationary Call Traffic Condition  dynamic guard scheme is able to maintain the handoff call blocking probability at the targeted objective, but the fixed guard scheme fails to accomplish that when the ratio of deviation is more than 1.5; while the new call blocking probabilities are comparable 94  10 G * • 3  OB * B • BH OB  N  H  N  (Fixed Guard) (Dynamic Guard) (Fixed Guard) (Dynamic Guard)  <A 10  XI  o  12 o Ci  m 10'  10 '  4.0  2.0 3.0 Ratio of Deviation from Nominal New Call Arrival Rate  1.0  Figure 36 Effect of Non-Stationary Condition on Call Blocking Probabilities under the two schemes.  Forced Call Termination and Overall Call Failure Probabilities  Various  system  traffic performance characteristics can be derived from the new call and handoff call blocking probabilities.  When the mobile handoff probability is small due to large cell  area or slow mobile movement, B^  con  would be a good indication of system traffic per-  formance. However, when the mobile handoff probability is large due to microcells or high speed mobile movement,  Bff  con  and the resulting forced call termination probability  have increasing influence on system traffic performance. probability PFCT  The forced call termination  is the probability that an ongoing call is forced into termination, e.g.,  as a result of handoff failure. The overall call failure probability POCF,  or the probability  that a call cannot be completed because of either blocking or unsuccessful handoff, would be a good unified performance measure of the effects of both 95  B^  con  and  Bn . con  Recall that PH is the probability that, either a new call which is not blocked or a call which has already been handed off successfully, will require at least one more handoff before  completion, oo  PH = P{Tc > Tff} = J [1 where  FT (t) and fT (t) c  3 1 6  H  the  (30)  F (t)]f (t)dt Tc  TH  cdf and pdfof the unencumbered call duration Tc and  the channel holding time Tjj respectively. Recalling that By  and Bff are the per-link  new and handoff call blocking probabilities, respectively, and B^con  and Bffcon  are the  per-connection new and handoff call blocking probabilities, respectively. Then, PpcT  due to handoff call blocking or resource unavailability is given as  follows:  PFCT  =  Y,[ ^-B i=l B  Bff Pff \-P (\-B  J - Ph l  H c  con  1  H  E e m  )  (3D  and POCF is given as follows:  POCF = B  N  C  + P CT(1  M  - B  F  )  (32)  NCON  Substituting Eqn. (31) into Eqn. (32), D  . Hcon  D  POCF = B  +  NCON  Since B  = 1 - (1 - B )  L  H  A  M  H  H^  B  and B  x  _  P  h  = 1 - ( 1 - B  N  [  NCM  1  _  B  H  M  j  c  )  ^  = 1 - (1 - B ) with L as the number of links, l-(l-B ) \P (l-B y L  L  NCON  H  POCF  ~ B )  P  )  L  +  N  H  N  \-p (\-B y H  L  04)  H  We can see that when Pg or the probability of cell transition of M T is large due to small cell size area, high M T roaming speed or large call duration, PFCT more dominant effect on POCF-  would have a  A t the extreme with Pff = 1, it means that successive 96  handoffs continue indefinitely, resulting in of cell transition is small, extreme with  P  H  Fig. 37 shows  POCF  =  1- Conversely, when the probability  would have a more dominant effect on  Bjy  con  = 0,  POCF  =  PFCT  and  POCF  B  N c o n  At the  POCF-  .  as functions of  L  when  =  0.333 calls/second and  p = 0.85. It illustrates that while the fixed and the dynamic guard schemes offers similar POCF,  the dynamic guard scheme offers a lower  PFCT-  10°  o CL 3  o  • <3> O $  •2  10  B Forced Call Termination Prob. (Fixed Guard) O Forced Call Termination Prob. (Dynamic Guard) ©Overall Call Failure Prob. (Fixed Guard) *- Overall Call Failure Prob. (Dynamic Guard)  -3  •e-a-f  c o E •4 £ 10  o o 0  10 20 Number of Connecting Links L  30  Figure 37 Forced Call Termination and Overall Call Failure Probabilities  97  5.2 Bandwidth Allocation over ATM-Based Backbone Network  5.2.1 Dynamic Guard Channel Extension The fixed or dynamic guard channel methods can be extended to the nodes in the backbone network for link bandwidth allocation to enable prioritized call admission control. Different bandwidth assignment schemes may be employed depending on the traffic characterization and multiplexing scheme. Cellular mobile systems employ per call multiplexing and deterministic bandwidth or channel assignment, whereas ATM-based backbone networks may employ statistical multiplexing and per-call peak or statistical bandwidth assignment.  Since statistical bandwidth assignment can be mapped into per-  call equivalent bandwidth assignment [38, 39], the concept of guard channels can be directly applied to set aside reserved bandwidth for handoff calls. In this thesis, model for the dynamic guard bandwidth scheme is based on per-call equivalent bandwidth or deterministic channel assignment.  To extend the dynamic guard channel scheme to the backbone network, the estimation of the instantaneous handoff call arrival rate is required to adapt the resource access priority to changing handoff call arrival rates.  The estimation process is based on  the accounting of all potential handoff connections (associated with active M T s with ongoing calls) passing through a network link and the determination of the rerouting probabilities from the ongoing connections to the potential handoff connection. In the mobile network, handoffs into a radio cell can only originate from the neighbouring cells and the accounting of all potential handoffs at a radio cell can be performed by monitoring the state information of the neighbouring cells.  In the backbone network,  however, handoffs to a wired link is dependent on the network topology and call routing. 98  With fixed routing, a route is selected for each source-destination pair of access nodes within the ATM-based backbone network. Under this situation, potential handoff connections can be extended from the radio cells to a given destination access node through the backbone network.  Consequently, the accounting of all potential handoff  connections passing through a wired link can be performed. However, fixed routing is typically not used in large network because of its lack of flexibility to adapt to network congestion or failures. With adaptive routing, a route is selected on a per call basis within the ATM-based backbone network. Consequently, potential handoff connections cannot be extended from the radio cells into the backbone network to reach the destination node, as it is not possible to predict what routing decision will be made in the future. Thus, adaptation to handoff rate fluctuations is not possible because of the inability to estimate the instantaneous handoff call arrival rate at each network link. We propose to overcome this inability by employing the mobile virtual circuit ( M V C ) as described in Chapter 2.  Extension via Mobile Virtual Circuit (MVC)  The M V C architecture introduces the  connection control services of potential handoff connection setup and release to support fast seamless handoff and optimal resource utilization, with the setup service involving route determination and logical links reservation (but no bandwidth reservation) and the release service involving logical links release. Each M V C connects the B S in the current cell serving a M T and the BSs in all the immediate neighboring cells to which the M T may potentially handoff its call, to the T P within the ATM-based backbone.  After a  handoff, the T P may shift to another node as the M V C is reconfigured. Because of the continuous route updates for potential handoff connections during the lifetime of a mobile call, the M V C naturally accommodates the proposed dynamic guard bandwidth scheme. 99  To extend the proposed dynamic guard bandwidth scheme into the backbone network via the M V C , we propose to enhance these M V C connection control services with dynamic guard bandwidth management functions. connection establishment, handoff connection  disconnection  This involves, at the time of M V C  or reconfiguration, interpreting the potential  setup or release at each mobile-specific  link respectively  as an  increase or decrease in the number of potential handoff calls for the respective link, and propagating the estimated probability of handoff call arrival as derived from the M T mobility pattern over each link set during setup. Consequently, the dynamic guard bandwidth controller would update the estimated instantaneous handoff call arrival rate at the wire-link as discussed  below.  Dynamic Guard Bandwidth Management Functions with MVC  By  employing  M V C to support connection rerouting during call handoff, the number of active M T s that can initiate handoff calls at a wired link is limited by the number of potential handoff connections of the M V C s passing through the link as illustrated in Fig. 38 (e.g., M V C s 1 and 2 passes through link xy which is associated with two active M T s ) . A n active M T situated at the cell terminating the current ongoing connection of a M V C may generate a handoff call to the neighbouring cells terminating the potential handoff connections of the M V C . Consequently, the probability of a handoff call initiated by the active M T at the wired links along a potential handoff connection is determined by the corresponding handoff probability between the respective cells. As discussed in Section 5.1.1, the probability of a handoff call arrival at a cell generated by an active M T in a neighboring cell during an estimation interval depends on its mobility pattern, the cell size, the remaining call duration and the length of the estimation interval. 100  M Mobile Terminal  Mobile Switching Centre  A O ®  B a s e Station  Cell  Tethered-point ( A T M Switch)  o Connecting Point ( A T M Switch)  Figure 38 Estimation of Instantaneous Handoff Call Arrival Rate at a Wired Link with M V C  To enable continuous update of the estimated instantaneous handoff call arrival rate at a wired link, the signalling of changes in the number of ongoing calls that may handoff to the wired-link occurs during potential handoff connection setup and release associated, respectively, with the establishment and release/reconfiguration of a M V C . It is relatively simple to integrate the dynamic guard bandwidth management functions into the potential 101  handoff connection control services associated with the M V C . In general, new or handoff call arrivals at a test cell will increase the probability of potential handoff call arrival at the neighbouring cells and at the potential handoff connecting link-sets of the M V C , which are terminated at the BSs of the neighbouring cells and converge  at the T P within the wired ATM-based backbone.  Routes are  predetermined for the potential handoff connecting link-sets but actual bandwidth is not reserved, and a fast bandwidth reservation scheme will be employed to reserve actual bandwidth if the anticipated handoff does occur. Conversely, call completions or call departures due to handoffs in a test cell will decrease the probability of potential handoff call arrival at the neighbouring cells and at the potential handoff connecting  link-sets  of the M V C . During M V C setup to support new call arrivals or M V C reconfiguration to support handoff call arrivals, the T P would signal the dynamic guard bandwidth controllers along the potential handoff connecting link-sets to update the increase in the instantaneous handoff call arrival rate at the corresponding links and the B S , and to increase the guard bandwidth by employing respectively the proposed estimation and adaptation schemes as described in this chapter.  During M V C release to support call completions or call  departures due to handoffs, the T P would signal the dynamic guard bandwidth controllers along the previous potential handoff connecting link-sets to update the decrease in the instantaneous handoff call arrival rate at the corresponding link and to decrease the guard bandwidth accordingly.  102  5.2.2 Performance Analysis Given that the dynamic guard channel scheme is extended for bandwidth allocation over the proposed M V C connection architecture (Chapter 2) to keep the per-link handoff call blocking probability close to the targeted objective under non-stationary traffic condition,, this section determines the reservation failure and the forced call termination probabilities when the recovery control for handoff call blocking in the ATM-based backbone network is based on the robust reservation transaction of the proposed R F R T P (Section 3.1.1) via localized retry over alternative reservation routes. When the recovery control is based on the blocked-calls-dropped discipline, the reservation failure probability is the same as the blocking probability.  Reservation Failure Probabilities  Let MH be the number of "mobile" links, between  the T P and the B S , of the selected handoff connection of the M V C , N t , rou  e  be the  number of local alternative routes between the node-pair of the i-th "mobile" link along the predetermined route,  A^,-,^..  be the number of connecting links of the j-th alternative  route between the node-pair of the i-th "mobile" link, (Nun^  — I), Bff be the targeted  per-link handoff call blocking probability. Let Proute.j be the probability of successful reservation along the j-th local alternative route between the node-pair of the  i-th "mobile" link, then  Pro»te  = (I ~ B ) ' *  (35)  N ink  h  H  Let Pnodepain be the probability of successful resource reservation between the node-pair of the  i-th "mobile" link, then k = N tei rou  Pnodepain = (1 ~ Bff) + Prouten Bff +  +  Proute  .  lNroute  Bff  JJ  —1  (1 -  P oute ) r  tk  k=l  (36) 103  Consequently, the per-connection reservation failure probability during handoff Fff  con  is  given as follows:  Fflcon — I  For simplification, let N  = N  r o u U i  P  Pnodepairi  , N  r o u t e  = N  l i n k i j  = (1 - B )  r o u t e  l i n k  (37)  , P de air, no  P  = Pnodepair, then  (38)  N l i n k  H  •A/route 1 Pnodepair = (1 ~ Bff) + P uteBff  ^  TO  (1 - Proute) ~  (39)  k=0  F  = 1 - (Pnodep r)  (40)  MH  Hcm  al  Substituting Eqn. (38) into Eqn. (39), •Nroute —1 Pnodepair = (I - B  H  ) + (l-B  H  )  N  l  i  n  k  B  £  H  7  [l ~ (1 -  link  (41)  Substituting Eqn. (41) into Eqn. (40), -Nroute — 1 FH  = l - { ( l - B  cm  B  )  + ( l - B  H  f ^ B  Forced Call Termination Probability  [l-(l-iW"'-*  E  k "  (42)  The probability of forced call termination due  to reservation failure probability is given as follows:  p  Fff  where, Fff  con  PJJ  con  F  C  T  -  l-P  H  (l-F  TO  LLNK  Fig. and  H  H  c  o  n  )  is defined in Eqn. (42), and P # is the probability that a call will re-  quire at least one more handoff before completion. N ute,N ,M B ,  H  i.e., P  39 shows PFCT  F  C  T  =  is a function of  f{NrouteNl ,M B ). LNK  as a function of N t  Nunt = 1. It shows that PFCT  Therefore, PFCT  T0U  e  H  H  for varying MH when By  - 0.005  can be decreased by a power factor of about 2.3 104  BH=0.005  Nlink = 1  No. of Local Alternative Routes  Nrgute  Figure 39 Effect on Forced Call Termination Probability due to M V C Connection Configuration (0.005) for each increase of is directly proportional to Nroute  >  4 for  1 <  Fig. 40 shows  MH-  M  <  H  PFCT  N  by 1, regardless of  r o u t e  TO  achieve  PFCT  For a given  MH-  < 10~ , 9  N  > 3 for  N te rou  r o u t e  ,  M  = 1, and  MH  = 10 and  E  30.  as a function of  for varying  N te rou  E>H  when  Nunk = 1- Fig- 40 shows that the inverse power proportionality factor between  and  Nroute  is directly proportional to  to 0.05 and 0.5,  PFCT  each increase of  N^ute  and  for  N^ute  >  7  Fig. 41 shows BH  B  For example, when  BH-  BH  PFCT  increases from 0.005  is decreased respectively by a factor of 0.005, 0.05 and 0.5 for by 1. To achieve H  PFCT  =  P CT  <  F  10~ , 9  N  > 4  r o u t e  for  B  H  = 0.005,  0.05.  as a function of  N  r o u  t  e  for varying  Nn k n  when  MH  = 0.005. It shows that the inverse power proportionality factor between  Nroute  PFCT  is directly proportional to  Nn kn  For example, when 105  Nu j. n  = 10 and PFCT  and  increases from 1 to  No. of Alternative Routes  Nrgute  Figure 40 Effect on Forced Call Termination Probability due to Targeted per-link Handoff Call Blocking Probability 2 and 3, PFCT increase of N  is decreased respectively by a factor of 0.005, 0.01 and 0.015 for each  r o u t e  by 1. To achieve PFCT  Nroute > 5 for 3 < N  PFCT  BH  < 5, and N  T O u U  - 9  r o u t e  > 6 for 6 < N  can be minimized by decreasing ME, BH and  also minimize PFCT Thandoff-  l m k  < 10 , N  I  n  > 4 for 1 <  l m k  Nu k. n  Nu  nk  < 2,  and  < 7.  Increasing N  r o u  t  e  would  but at the expense of increasing the handoff processing delay  Section 4.2.1, it has been shown that for a given MH with the targeted  being not greater than 0.015 and the  increased to maximize PFCT  Nn k n  being not more than 2, N  without causing a significant increase in  106  r o u t e  Ti i ff. ian( 0  can be  M|_|= 10  B  = 0.005  H  ^ N  l  i  n  k  - Q Niink O ON, A—AN| O ONiinK V VN|i > E>N| i n k  l n k  n k  ink  3  4  5  No. of Alternative Routes  6  =1 =3 =4 =5 =6 =7  7  Nj-Qytg  Figure 41 Effect on Forced Call Termination Probability due to Reservation over Alternative Routes  107  10  Chapter 6  Conclusions  Seamless mobile call handoffs over an A T M / B - I S D N require real-time connection rerouting capabilities which are currently not available. We have solved this problem by developing a time-distributed handoff connection establishment scheme which distributes the routing and resource allocation tasks among the call processing and handoff processing phases. This scheme employs a mobile virtual circuit ( M V C ) connection architecture in the A T M network to link current and potential handoff connections (with predetermined routes but without resource allocation) to the root of a connection tree, which reconfigures after each handoff to enable continuous support of successive handoffs. During a handoff, a fast and robust resource reservation service establishes the handoff connection via distributed control, and a seamless synchronized traffic rerouting service prevents cell loss and sequencing errors. transaction protocols:  (1)  These services employ the following real-time signaling the robust fast reservation transaction protocol  (RFRTP);  and (2) the packet-ordering synchronization transaction protocol (POSTP). The R F R T P employs parallel commit operations to minimize contention delays, and localized retries to maximize robustness. Performance analyses are presented to evaluate the lower and upper bound delays of the R F R T P for a given allocation control algorithm and the robustness improvement due to localized retry.  These signaling transaction protocols  are supported by fast unicast and multicast associate-CCS transport services and an inband signaling transport service, employing new signaling protocols in the signaling A T M adaptation layer. Integration of the above services and protocols enables seamless handoff processing.  Handoff performance is analyzed in terms of the synchronization  buffering and resequencing delays, the forced-termination vulnerability interval, and the  108  handoff processing delay. The proposals extend the current A T M / B - I S D N standards to support wireless personal communications and are backward compatible with current ATM  implementations.  The proposed dynamic guard channel scheme adapts the resource access priority via guard bandwidth adjustments in response to changes in instantaneous handoff call arrival rates, estimated by observing the state information of neighbouring cells. Under stationary traffic condition, the proposed scheme offers better resource utilization than that of the fixed guard channel scheme by providing lower and comparable long-term handoff call blocking probabilities under respective medium and heavy bandwidth loadings. Under non-stationary traffic condition, the proposed dynamic guard scheme enables the handoff call blocking probability to stay close to its targeted objective, without  significantly  increasing the new call blocking probability relative to the fixed guard scheme.  The  distributed adaptive algorithm is simple and only requires a B S to exchange a small amount of information (number of ongoing call, or equivalently, call arrival and departure events) with its neighbors whenever a call arrives at or leave the corresponding cell. The required number of guard channels can be calculated off-line and loaded into lookup tables to facilitate the dynamic allocation. With adaptive call routing within the A T M based backbone network, estimation of instantaneous handoff call arrival rate for each network link is not feasible and the dynamic guard scheme cannot be applied.  We  propose to overcome this difficulty by enhancing the connection control services of the proposed M V C to support the dynamic guard management functions.  In so doing, the  proposed dynamic guard scheme is applicable to channel allocation over cellular mobile networks, and is extended to bandwidth allocation over backbone network to enable a unified approach to prioritized call admission control over the ATM-based P C N .  109  6.1 Possible Future Works It is expected that further future generations of P C N s will evolve to support multimedia and multiparty (e.g., conferencing) applications, thus there should be growing interests in researching on the issues and problems of extending P C S capability to the A T M / B - I S D N to support user terminal mobility during calls under these communications environments. This thesis has examined these issues and problems under the bipartisan single-media communication environment, and it would be of interest to examine if the proposed M V C connection architecture can be employed or extended easily to facilitate fast handoffs under those communication environments. Under the bipartisan multimedia communication environment, the proposed M V C could be employed to facilitate fast handoffs if the media of a call are multiplexed onto a single virtual circuit. Similarly, it would be of interest to examine the applicability of the proposed real-time disruptionfree connection service to support seamless network-wide handoffs (non-diversity based) over connection architectures that may be employed to support those communication environments. This thesis maintains two call blocking priorities for the handoff and new call requests which compete for shared connection resources over the ATM-based P C N under nonstationary call arrival condition.  Since forced call terminations due to handoff call  blocking are generally more objectionable than new call blocking, new call requests from low mobility terminal should be assigned a higher blocking probability to offset its inherent advantage of low forced call termination probability due to low probability of cell transition. It would be of interest to extend the proposed dynamic guard channel scheme to maintain three call blocking priorities for the following call classes in descending priority order: handoff calls, new high mobility calls (e.g., from vehicle) and new low  110  mobility calls (e.g., from pedestrian).  Ill  Bibliography  [1] M . Kawarasaki and B . Jabbari, " B - I S D N Architecture and Protocol,"  IEEE J. Sel.  Areas Commun., pp. 1405-1415, December 1991. [2] H . Breuer, " A T M - L a y e r O A M : Prinicples and Open Issues," vol.  IEEE Commun. Mag.,  29, pp. 75-78, September 1991.  [3] J. Anderson and M . Nguyen, " A T M - L a y e r O A M Implementationlssues,"  IEEE  Commun. Mag., vol. 29, pp. 79-81, September 1991. [4] D . L . Schilling et al., "Broadband C D M A for Personal Communications Systems,"  IEEE Commun. Mag., vol. 29, pp. 86-93, November 1991. [5] R . L . Pickholtz, D . L . Schilling and L . B . Milstein, "Spread Spectrum for Mobile Communications,"  IEEE Trans. Veh. Technol., vol. 40, pp. 313-322, M a y 1991.  [6] J . N . H . Grond, H . P J . Hecker and C . Shepherd, "Broadband on the Move," IEEE  Int.  Conf. onlnteg. Broadband Services and Networks, pp. 206-210, October 1990. [7] B . Materna, B . Vaughan and C . Britney, "Evolution from L A N and M A N Access Networks Towards the Integrated Network," pp.  1455-1461, November  in Proc. IEEE Globecom, Dallas, TX,  1989.  [8] A . D . Malyan, L J . N g , V . C . M . Leung, and R . W . Donaldson, "Network Architecture and Signalling for Wireless Personal Communications," vol.  11, pp. 830-841, August  IEEE J. Sel. Areas Commun.,  1993.  [9] V . C . M . Leung, N . Qian, A . D . Malyan, and R . W . Donaldson, "Call Control and Traffic Transport for Connection-Oriented High Speed Wireless Personal Communications over Metropolitan Area Networks," 1388, October  IEEE J. Sel. Areas Commun., vol. 12, pp. 1376-  1994. 112  [10JA.S. Acampora, and M . Naghshineh, " A n Architecture and Methodology for MobileExecuted Handoff in Cellular A T M Network," pp.  IEEE J. Sel. Areas Commun., vol. 12,  1365-1375, Oct. 1994.  [11]S. Suzuki and K . Funakawa, "Signalling Protocol Architecture for Digital Mobile System,"  in Proc. IEEE Veh. Technol, pp. 729-734, 1989.  [12]B. Jabbari, "Intelligent Network  Concepts  in Mobile Communications,"  IEEE  Commun., pp. 64-69, Feb. 1992. [13]J. Homa and S. Harris, "Intelligent Network Requirements for Personal Communications Services,"  IEEE Communications, pp. 70-76, Feb. 1992.  [14]O.T.W. Yu and V . C . M . Leung, "Connection Architecture and Protocols to Support Efficient Handoffs  over an A T M / B - I S D N  Personal Communications  Network,"  ACMIBaltzer Journal on Mobile Networks and Applications, vol. 1, pp. 123-139, Oct.  1996.  [15]C. Toh, "Crossover Switch Discovery in Wireless A T M L A N ' s , "  ACM Mobile  Networks and Appl. J., vol. 1, no. 2, pp. 141-165, 1996. [16]O.T.W. Yu and V . C . M . Leung, "Connection Establishment over B - I S D N to Support Wireless P C N Handoff Processing," vol.  in Proc. IEEE Pacific Rim Conf., Victoria, BC,  1, pp. 318-321, M a y 1995.  [17]O.T.W. Yu and V . C . M .  Leung, " B - I S D N Architecture and Protocols to Support  Wireless Personal Communications Internetworking," ONT, vol. 2, pp. 768-772, Sept. [18JO.T.W. Yu and V . C . M .  in Proc. IEEE PIMRC, Toronto,  1995.  Leung, "Extending B - I S D N to Support User Terminal  Mobility over an A T M - B a s e d Personal Communications Network,"  Globecom, Singapore, vol. 3, pp. 2289-2293, Dec. 1995. 113  in Proc. IEEE  [19]O.T.W. Y u and V . C . M . Leung, "Signalling Network Architecture and Transaction Protocols to Support Realtime Connection Rerouting in A T M / B - I S D N s , " Submitted  to IEEE/ACM  Trans, on Networking.  [20JO.T.W. Yu and V . C . M . Leung, "Signalling Network Architecture and Transaction Protocols to Support Realtime Connection Rerouting in A T M / B - I S D N s , " in Proc.  IEEEInfocom, San Francisco, CA, vol. 3, pp. 1012-1019, March 1996. [21]I. Cidon, I. Gopal, and A . Segall, Networks,"  IEEE/ACM  "Connection  Establishment  in  High-Speed  Trans, on Networking, vol. 1, pp. 469-481, August 1993.  [22]D. Ferrari and D . Verma, " A Scheme for Real-Time Channel Establishment in WideArea Networks,"  IEEE J. Select. Areas Commun., vol. 8, pp. 368-379, April 1990.  [23]R.K. Berman and J . H . Brewster, "Perspectives on the A I N Architecture,"  IEEE  Commun., vol. 31, pp. 27-33, Feb. 1992. [24]P.J. Kuhn, C D . Pack and R . A . Skoog, "Common Channel Signalling Networks: Past, Present, Future," IEEE  J. Sel. Areas Commun., vol. 12, pp. 383-394, April 1994.  [25]M. Naghshineh, and Anthony S. Acampora, " Q O S Provisioning in Micro-Cellular Networks pp.  Supporting Multimedia Traffic,"  8d.4.1-8d.4.10, April  in Proc. IEEE Infocom, Boston, MA,  1995.  [26]A. Gavious and Z . Rosberg, " A Restricted Complete Sharing Policy for a Stochastic Knapsack Problem in B - I S D N , "  IEEE Trans. Commun., vol. 42, pp. 2375-2379, July  1994.  [27]M. Naghshineh, and M . Schwartz, "Distributed Call Admission Control in M o bile/Wireless Networks,"  IEEE J. Sel. Areas Commun., vol. 14, pp. 711-717, M a y  1996. 114  [28]D. Hong, and S.S. Rappaport, "Traffic Model and Performance Analysis for Cellular Mobile Radio Telephone Procedures,"  Systems with Prioritized and Nonprioritized Handoff  IEEE Trans. Veh. Technol., pp. 77-92, A u g . 1986.  [29]J. Daigle and N . Jain, " A Queueing System with Two Arrival Streams and Reserved Servers with Application to Cellular Telephone," 2167, April  1992.  [30]R. Guerin, "Queueing-Blocking System Channels,"  in Proc. IEEE Infocom, pp. 2161—  with Two Arrival  Streams and Guard  IEEE Trans, on Commun., vol. 36, pp. 153-163, Feb. 1988.  [31]C.H. Yoon, and C . K . U n , "Performance of Personal Portable Radio Telephone Systems with and without Guard Channels,"  IEEE J. Sel. Areas Commun., vol. 11,  pp. 911-917, A u g . 1993. [32]W.C.Y. Lee, "Smaller Cells for greater performance," pp.  19-23, Nov.  IEEE Commun. Mag., vol. 29,  1991.  [33]O.T.W. Yu and V . C . M . Leung, "Adaptive Resource Allocation for Prioritized Call Admission over an ATM-based Wireless P C N , "  Submitted to IEEE J. Sel. Areas  Commun. [34]O.T.W. Yu and V . C . M . Leung, "Self-Tuning Priorized Call Handling Mechanism with Dynamic Guard Channels for Mobile Cellular Systems," in Proc. Veh. Technol.  Conf., Atlanta, GA, pp. 1520-1524, April 1996. [35]R. Cohen and A . Segall, "Connection Management and Rerouting in A T M Networks,"  in Proc. IEEE Infocom, Toronto, ONT, pp. 2 b . l . l - 2 b . l . 8 , 1994. [36]W. Lee, Hill,  Mobile Cellular Telecommunications Systems, ch. 4. New York: McGraw-  1989. 115  [37]A. Murase,  I.  Symington,  and E . Green, "Handoff  Criterion for Macro  Microcellular Systems," in Proc. IEEE Veh. TechnoL, pp. 524-530, M a y  and  1991.  [38]J.Y. H u i , M . Gursoy, N . Moayeri, and R. Yates, " A Layered Broadband Switching Architecture with Physical or Virtual Path Configurations," IEEE Commun., vol. 9, pp. 1416-1426, December  J. Sel.  Areas  1991.  [39]D. Hong and T . Suda, "Congestion Control and Prevention in A T M Networks," IEEE Network Magazine, vol. 5, pp. 10-16, July  116  1991.  Appendix List of A c r o n y m s and Abbreviations  AIN  Advanced Intelligent Network  ATM  Asynchronous Transfer Mode  B-ISDN  Broadband Integrated Services Digital Networks  BS  Base Station  CCS  Common Channel Signaling  GVC  Group Virtual Circuit  GVCI  Group Virtual Circuit Identifier  IN  Intelligent Network  MAP  Mobile Application Part  MSC  Mobile Switching Center  MT  Mobile Terminal  MVC  Mobile Virtual Circuit  MT  Mobile Terminal  MTP  Mobile Transfer Part  N-ISDN  Narrowband Integrated Services Digital Networks  PCN  Personal Communication Network  PCS  Personal Communication Services  POSTP  Packet Ordering Synchronization Protocol  PSTN  Public Switched Telephone Network  PT  Payload Type  PVC  Permanent Virtual Circuit  QoS  Quality of Service  RFRTP  Robust Fast Reservation Transaction Protocol  SAAL  Signaling A T M Adaptation Layer  SCCP  Signaling Connection Control Part  SS7  Signaling System No. 7  TCAP  Transaction Capabilities Application Part  TP  Tethered Point  VC  Virtual Circuit  VCI  Virtual Circuit Identifier  117  VP VPI  Virtual Path Virtual Path Identifie  118  

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}]}"
                            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:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0065156/manifest

Comment

Related Items