Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An intelligent location management algorithm for cellular systems Koh, Jason Jae Hyun 2000

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

Item Metadata

Download

Media
831-ubc_2000-0445.pdf [ 3.68MB ]
Metadata
JSON: 831-1.0065263.json
JSON-LD: 831-1.0065263-ld.json
RDF/XML (Pretty): 831-1.0065263-rdf.xml
RDF/JSON: 831-1.0065263-rdf.json
Turtle: 831-1.0065263-turtle.txt
N-Triples: 831-1.0065263-rdf-ntriples.txt
Original Record: 831-1.0065263-source.json
Full Text
831-1.0065263-fulltext.txt
Citation
831-1.0065263.ris

Full Text

AN INTELLIGENT LOCATION MANAGEMENT ALGORITHM FOR C E L L U L A R SYSTEMS by JASON J A E H Y U N K O H B A . S c . (Electrical and Computer Engineering), University of British Columbia, 1 9 9 8 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E OF M A S T E R OF A P P L I E D S C I E N C E in T H E F A C U L T Y O F G R A D U A T E STUDIES D E P A R T M E N T OF E L E C T R I C A L A N D C O M P U T E R E N G I N E E R I N G We accept this thesis as conforming to the required standard T H E U N I V E R S I T Y OF BRITISH C O L U M B I A August 2000 © Jason Jae Hyun Koh, 2000 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of The University of British Columbia Vancouver, Canada Date iw, °\ DE-6 (2/88) Abstract With the explosive growth in the number of cellular subscribers in the past decade, cellular systems are faced with the challenge of increasing the efficiency of the usage of the limited frequency spectrum. The service providers are forced to use increasingly smaller cells to increase system capacity; however, smaller cell sizes result in higher signalling loads generated by location management procedures. The intelligent location management algorithm which utilizes the user profile of individ-ual users is proposed. The user profile contains a history of the user's cell-to-cell transitions and the total time spent in a cell. Using this information, the intelligent algorithm generates location areas dynamically around the user in such a way that the mean location updating traffic is reduced, and pages within the location area in multiple steps such that the mean paging traffic is reduced by paging the most probable cells first. A user mobility model is proposed for the purpose of quantifying the performance results of location management algorithms. The proposed model, activity-based mobility model with directional random walk, simulates mobility behaviour patterns of users of different classes. Such a model which captures an individual user's daily movement pattern is important in properly evaluating location management algorithms that utilize user mobility history. Performance in terms of radio signalling is evaluated and compared with other location management algorithms such as the G S M and dynamic location management algorithms. The intelligent algorithm produces less radio signalling at the cost of increased memory storage and processing capabilities at the mobile station and network. i i Table of Contents Abstract ii List of Tables v List of Figures vi Acknowledgments ix Chapter 1 Introduction 1 1.1 Location management concepts 1 1.2 Problem definition and motivations 2 1.3 Thesis objectives and outline 3 Chapter 2 Previous Work on Location Management 5 2.1 Background 5 2.2 Description of location management procedures in GSM 7 2.2.1 GSM network architecture 7 2.2.2 Location updating procedures 9 2.2.3 Paging procedures 12 2.3 Location management using dynamic individualized location areas 14 2.3.1 Dynamic location updating algorithm 15 2.3.2 Paging algorithm 17 Chapter 3 Activity-based Mobility Model with Directional Random Walk 18 3.1 Description of the mobility model 18 3.1.1 Activity transition and duration 20 3.1.2 Selection of activity location 22 3.1.3 Intermediate path 23 i i i 3.2 Simulation results 26 Chapter 4 Intelligent Location Management Algorithm 32 4.1 User profiles 33 4.2 Intelligent paging algorithm 37 4.3 Simulation results 43 Chapter 5 Performance Analysis Results 47 5.1 Description of the simulated environment 47 5.2 Performance comparisons of location management algorithms 48 5.2.1 Location updating traffic : 49 5.2.2 Paging traffic 51 5.2.3 Paging delay 54 5.2.4 Total location management cost 56 5.3 Sensitivity analysis 62 5.3.1 Effect of different mean velocities 63 5.3.2 Effect of different time period lengths 65 5.3.3 Effect of different dynamic location area sizes 67 5.3.4 Effect of different circular region sizes 69 5.3.5 Effect of different call arrival rates 71 Chapter 6 Conclusions 73 6.1 Future work 75 Glossary 77 Bibliography 80 iv List of Tables Table 3.1: Sample data from the activity transition matrix 21 Table 3.2: Sample data from the activity duration matrix 21 Table 5.1: Success rate of each paging step for the dynamic algorithm and the intelligent algorithm 55 List of Figures Figure 2.1: The network architecture of GSM 8 Figure 2.2: Signalling flow of an inter-VLR location update using IMSI 10 Figure 2.3: Signalling flow of an incoming call delivery from the PSTN 13 Figure 2.4: An example of location area cell inclusion with corresponding link weights greater than or equal to N 16 Figure 3.1: Intermediate path based on the Directional random walk model 24 Figure 3.2: Number of visits to each cell, for user type 0 (Full-time employed) 28 Figure 3.3: Total duration (in minutes) in each cell, for user type 0 (Full-time employed)28 Figure 3.4: Number of visits to each cell, for user type 1 (Part-time employed, not a student) 29 Figure 3.5: Total duration (in minutes) in each cell, for user type 1 (Part-time employed, not a student) 29 Figure 3.6: Number of visits to each cell, for user type 2 (Student, possibly employed part-time) 30 Figure 3.7: Total duration (in minutes) in each cell, for user type 2 (Student, possibly employed part-time) 30 Figure 3.8: Number of visits to each cell, for user type 3 (Unemployed, not a student) ..31 Figure 3.9: Total duration (in minutes) in each cell, for user type 3 (Unemployed, not a student) 31 Figure 4.1: Dividing the location area into two paging zones, Zj and Zj, based on re' ....39 Figure 4.2: A cell configuration of the paging zone Zj, consisting of m cells 41 Figure 4.3: Distribution of the sizes of the paging zones for various ae 44 Figure 4.4: Probability of finding user in paging zone Zj for various ae 45 Figure 5.1: Mean number of location updates per user versus elapsed time 49 Figure 5.2: Learning curve of the dynamic location update algorithm 50 Figure 5.3: Mean number of cells paged versus elapsed time, for Xca[[ = 3 calls/day 51 Figure 5.4: Mean number of cells paged versus elapsed time, for Xcall = 6 calls/day 52 Figure 5.5: Mean number of cells paged versus elapsed time, for A , c a ; / = 9 calls/day 52 Figure 5.6: Mean number of cells paged versus elapsed time, for )ical[ = 12 calls/day ....53 Figure 5.7: Mean normalized delay versus call arrival rate 55 Figure 5.8: Total location management cost versus elapsed time, for c = 5 and Xcatl = 3 calls/ day 57 Figure 5.9: Total location management cost versus elapsed time, for c = 5 and Xcan = 6 calls/ day 58 Figure 5.10: Total location management cost versus elapsed time, for c = 5 and Xcal[ = 9 calls/ day 58 Figure 5.11: Total location management cost versus elapsed time, for c = 5 and ~kcall = 12 calls/day 59 Figure 5.12: Total location management cost versus elapsed time, for c = 10 and Xca[l = 3 calls/day 60 Figure 5.13: Total location management cost versus elapsed time, for c = 10 and Xcau = 6 calls/day 61 Figure 5.14: Total location management cost versus elapsed time, for c = 10 and ~kcan = 9 calls/day 61 Figure 5.15: Total location management cost versus elapsed time, for c = 10 and Xcall = 12 calls/day 62 Figure 5.16: Mean number of location updates per user versus mean velocity, over a period of 50 days 63 Figure 5.17: Mean number of cells paged versus mean velocity, for Xcall = 6 calls/day, over a period of 50 days 64 Figure 5.18: Number of cells paged per user over a period of 50 days using the intelligent algorithm with various time period lengths and Xcall = 6 calls/day (Note that the lowest value of the y-axis is 830 66 Figure 5.19: Mean normalized delay using the intelligent algorithm with various time period lengths and X , c a / / = 6 calls/day 67 v i i Figure 5.20: Mean number of location updates per user versus maximum location area size Lmax, over a period of 50 days 68 Figure 5.21: Mean number of cells paged per user versus maximum location area size Lmax^ for Xcal[ = 6 calls/day, over a period of 50 days 69 Figure 5.22: Mean number of cells paged per user versus ae, for Xcall = 6 calls/day and Lmax = 20, over a period of 50 days 70 Figure 5.23: Mean number of cells paged per call for various call arrival rates 71 v i i i Acknowledgments I would like thank the Department of Electrical and Computer Engineering at the University of British Columbia, for giving me the opportunity to continue learning and to complete this program. I would like to express my sincere gratitude to my supervisor, Dr. Cyril Leung, for his invaluable advice, constructive criticism, and constant support throughout the course of my research. It has been a rewarding experience working with him. I would also like to thank my friends and colleagues at the University of British Columbia. I am particularly thankful to Kar Wing Ho, Kwok Hung Lee, Xiao Yan Feng, David Martin, and Hansen Wang for their invaluable assistance and friendship. I would like to thank my parents, and my brother Jeff, for their constant love, encourage-ment, and ceaseless prayers. I am also thankful to Pastor David Oh and the rest of my brothers and sisters at Philadelphia church for their support of prayers. Above all, I thank Lord Jesus for his unconditional love, sacrifice, and grace. By his grace alone I am what I am. I dedicate this work to him. This research was supported in part by NSERC grant OGP0001731. ix Chapter 1 Introduction In recent years, the personal wireless communications industry has been experiencing tremendous growth. When cellular communication service was introduced in the early 1980's, it was costly with limited availability [1]. Since then, factors including decreasing prices, improved radio coverage, and compact, lightweight user terminals have helped wireless service become an affordable alternative to the wired telephone service. To accommodate the continuously increas-ing number of subscribers and their demands for a wider range of services, while constrained by the limited frequency spectrum, higher efficiency in the usage of the frequency spectrum is required. A large number of research studies are being conducted in this area, including the development of location management, multiple access and channel allocation schemes [2]. 1.1 Location management concepts The number of wireless channels available was very limited in the early mobile radio systems, for only a single high power base station was used [3]. To accommodate a large number of subscrib-ers without requiring more of the scarce spectrum, wireless systems adopted a 'cellular' architec-ture. This architecture replaces the single high power base station with many low power base stations. Base stations are assigned with shares of wireless channels available to the entire system, with each of them covering only a small portion of the service area called a 'cell'. The wireless channels assigned to a base station may be re-used at other base stations, if the base stations are spaced sufficiently apart that the interference levels are tolerable. This approach of re-using wireless frequencies significantly increases the system capacity, allowing wireless systems to accommodate a larger number of subscribers. Handheld terminals also benefit from employing the cellular architecture. Since a handheld terminal only needs to communicate to the base station 1 Chapter 1 Introduction 2 of its residing cell , the transmit power level is reduced, resulting in longer battery life. Location management in a cellular system is required for effective delivery of incoming calls. To deliver an incoming call for a mobile user, the mobile switching center needs to know which cell the user is in, so that a connection can be made through the cell's base station. The two basic operations for this purpose are location updating and paging. Loca t ion updating is a procedure initiated by the mobile terminal, informing the network about its current location area. A location area is a group of contiguous cells, over which a mobile station may travel without performing location updates. Paging is a procedure initiated by the network, informing the mobile terminal of an incoming call. Paging messages are broadcast to all or just a group of selected cells within the mobile terminal's current location area. The selection of cells is determined by the paging scheme employed by the network. Both location updating and paging operations consume radio resources, since both involve radio signalling between the mobile stations and the base stations. 1.2 Problem definition and motivations A s cellular services have become more affordable with higher availability in recent years, the number of cellular subscribers worldwide has increased enormously. This has forced service providers to use increasingly smaller cells to meet the growing demand for additional system capacity. Although smaller sized cells provide higher system capacity, it results in higher signal-ling loads generated by location management procedures. This is because it increases either the number of cells in a location area or the number of location areas in the service area. Wi th the location management algorithm of G S M , the number of paging messages is increased in the former case, and the number of location updates is increased in the latter case. Chapter 1 Introduction 3 The above situation is a result of using geographically fixed location areas that are applied to all mobile users, and sending paging messages to all cells in a location area upon call arrival. This method of broadcasting paging messages is also known as flood paging. The motivation behind this thesis is the realization that a mobile user's individual mobility pattern, with respect to time, may be captured using the user's history of movements, from which some intelligence may be derived and applied to the location management procedures. For instance, at the time of a location update, location areas can be dynamically formed with cells that are visited more often than others, centered at the user's current cell. Similarly, when delivering a call , cells in which the user had spent more time than others can be paged first. 1.3 Thesis objectives and outline The main objectives of this thesis are: (i) To propose a realistic mobility model for the purpose of comparing different location management algorithms. (ii) To propose an improved location management algorithm with added intelligence. (iii) To analyze and compare the performances of the proposed location management algo-rithm with other existing algorithms. (iv) To study the sensitivity of the performance results with variations in mobility and loca-tion management parameter values. Chapter 2 reviews the location management procedures of the current second generation G S M cellular system and describes a recently proposed dynamic location management algorithm. The proposed mobility model is explained in Chapter 3. This model combines two existing mobility Chapter 1 Introduction 4 models proposed by other authors. In Chapter 4, the proposed location management algorithm is described. This algorithm improves on the dynamic location management algorithm by introduc-ing a new paging scheme with added intelligence. Chapter 5 discusses the performance of the proposed algorithm in terms of location update traffic, paging traffic, and call setup delay. The results are compared with the dynamic location management algorithm and the G S M algorithm with three different parameter values. Chapter 2 Previous Work on Location Management Location management has received a fair amount of attention in recent years. In this chapter, some background on location management is presented, along with an overview of the location management in G S M (Global System for Mobile communications), and the dynamic individual-ized location management algorithm proposed in [13]. The location management algorithm in G S M is used as the standard of comparison for many location management proposals. In Chapter 5, the G S M location management algorithm and the dynamic individualized location management algorithm are compared with the intelligent algorithm proposed in this thesis. 2.1 Background In first generation cellular systems, when a call arrives for a user at the mobile switching center (MSC), a paging message is sent out simultaneously on every cell's base station forward control channel (FCC) in the system with the user's mobile identification number (MIN). If the user's mobile station successfully receives the message, it responds by transmitting an acknowledge-ment message on the base station reverse control channel (RCC) of the current cell. The M S C then sets up the call through the base station of the user's current cell. The broadcasting of a paging message to all cells in the system is called system wide paging, and a significant amount of radio resource is wasted due to this. However, at the time of first generation cellular systems (in the 1980's), traffic was highly unbalanced [5]. Less than one third of calls were incoming calls to the cellular user. It is mentioned in [5] that frequency of a paging procedure was rather low, and system wide paging had little impact on the overall radio bandwidth usage. The paging procedure became more important as the number of subscribers increased and the incoming and outgoing call rates became more balanced. In order to reduce the amount of 5 Chapter 2 Previous Work on Location Management 6 wasted radio resource due to system wide paging, the concept of location management, with location updating and paging as the two basic operations, was introduced in the second generation cellular systems. In second generation cellular systems such as the G S M system, the system coverage area is divided into geographical areas called location areas. Each location area is a group of contiguous cells defined by the network. The network keeps track of the location of every user to within a location area, and the identity of the location area within which a user currently resides is stored in the network database. In order to maintain this information for a user, a location update procedure is triggered whenever the user crosses a location area boundary. When a call arrives for a user, the identity of its current location area is retrieved from the network database, and a paging message is broadcast simultaneously to all cells in the location area. This is also known as flood paging. Additional signalling is required in G S M since a message is transmitted over the radio channel for a location update. However, the wasted radio resource due to paging is considerably lower when compared to system wide paging, since only a subset of cells in the system needs to be paged. Further details of the location management procedures in G S M are presented in Section 2.2. With the explosive growth in the number of cellular subscribers since the introduction of G S M in 1991, and the use of smaller sized cells to meet the growing demand for additional system capacity, location management has recently become a popular research topic aimed at further reducing the wasted signalling due to location updating and paging [4]. Many location management algorithms have been proposed in the past years, attempting to reduce signalling bandwidth for additional costs of increased implementation complexity and call setup delay. Various location updating algorithms have been studied in [6]-[14]. Some authors have proposed the idea of defining multiple location area layers [7]-[9], while others incorporated the concept of Chapter 2 Previous Work on Location Management 7 dynamically defining location areas [10]-[12]. These algorithms aim to reduce the mean number of location updates, and at the same time, avoid heavy congestion of location updates which occur only in the border cells of location areas. Numerous paging algorithms have been proposed in [13]-[27]. All of the algorithms use multiple-step paging, where cells of higher probabilities are paged in the earlier paging steps. The concept of user profiles was adopted in many proposals. Some user profiles include certain user characteristics which may help predict a user's mobility behaviour [21]-[25]. Others collected movement histories of individual users to estimate the cell probabilities [13], [15]-[20]. The common goal of the proposed paging algorithms was to reduce the mean number of cells paged per call, while maintaining the call setup delay at a tolerable level. 2.2 Description of location management procedures in G S M Global System for Mobile communications (GSM) is a second generation cellular system standard that was first introduced into the European market in 1991 [3]. In this section, the location management concepts and procedures of G S M are explained in greater detail. A brief overview of the G S M network architecture is given in Section 2.2.1, which will help understand the subsequent sections on the location updating and paging procedures used in GSM. 2.2.1 G S M network architecture As shown in Figure 2.1, the network architecture of G S M consists of three components, the mobile station, the network switching subsystem, and the base station subsystem. Chapter 2 Previous Work on Location Management 8 Public Switched Figure 2.1: The network architecture of G S M The central component of the network switching subsystem (NSS) is the mobile switching center (MSC). The M S C provides typical switching functions and handles location management, call delivery, and security. The home location register (HLR) is a centralized database that permanently stores subscriber information, and information related to location management and call delivery for all users of the network. In general, there is only one H L R in a G S M network, although it may be implemented as a distributed database using several physical HLRs . The visitor location register (VLR) temporarily stores similar information as in the HLR, but only of users that are currently located in the associated geographical area. The V L R is associated with a particular geographical area covered by one MSC. The M S C and the databases communicate with each other using the mobile application part (MAP) protocol, which uses the transport protocols in the Signalling System No. 7 (SS7) [4]. An M S C is in charge of one or more base station controllers (BSCs), and each BSC in turn controls several base station transceivers (BSTs). A subnetwork of the BSCs and their associated Chapter 2 Previous Work on Location Management 9 BSTs is called the base station subsystem (BSS). The BSS handles most of the radio resource aspects of G S M . Each BST is the actual antenna site that defines a single cell of the cellular layout, and the radio resource management of several BSTs and BST handover control are handled by a BSC. 2.2.2 Location updating procedures A location update is triggered when the mobile station (MS) enters a cell with a location area identifier (LAI) different from the current LAI stored at the MS. Additionally, a location update is triggered upon powering on of the MS (attach procedure), and the network may also force location updates to be triggered periodically. If the new location area belongs to the same V L R as the previous location area, the subscriber information already exists in the V L R and it is updated with the new LAI. This is called the intra-VLR location update. Otherwise, an inter-VLR location update must be performed and a number of extra steps are required. There are two types of inter-V L R location updates: inter-VLR location update using temporary mobile subscriber identity (TMSI), and inter-VLR location update using international mobile subscriber identity (IMSI) [5]. The IMSI is used to identify a subscriber, while the TMSI is assigned by a V L R to the subscriber, and is only valid while the subscriber is registered to that VLR. Although the TMSI alone cannot uniquely identify a subscriber, it is used to save paging bandwidth, since it is shorter than the IMSI. When the MS registers at a new V L R during a inter-VLR location update, no subscriber information exists and it must be requested from the HLR (for inter-VLR location update using IMSI) or from the previous V L R (for inter-VLR location update using TMSI). Then, the HLR is updated to point to the new V L R , and the subscriber information from the previous V L R is deleted. The signalling flow of an inter-VLR location update using IMSI is illustrated in Figure 2.2. Chapter 2 Previous Work on Location Management 10 MS BST BSC MSC V L R HLR Prev. V L R CHANNEL REQUEST IMMEDIATE ASSIGNMENT LOCATION UPDATE REQUEST CHANNEL REQUIRED CHANNEL ACTIVATION CHANNEL ACTIVATION ackl IMMEDIATE ASSIGN ESTABLISH INDICATION COMPLETE H LAYER 3 INFO LOCATION UPDATE ACCBPT CHANNEL RELEASE MAP_UPDATE_ LOCATION_AREA| |1rfAP_UPDATE_ LOCATION. AREA ack MAP_UPDAl'E. H LOCATION |^IAP_INSERT_ SUBSCRIBER. DATA MAP_INSERT_ SUBSCRIBER. DATA ack l lAP_UPDATE_ LOCATION ack MAP_CANCEL_ LOCATION MAP_CANCEL_ LOCATION ack Figure 2.2: Signalling flow of an inter-VLR location update using IMSI The procedure of an inter-VLR location update mainly consists of the following steps: 1. Allocate signalling channel for the MS - The MS transmits a C H A N N E L REQUEST mes-sage to the BST. The BST measures the transmission delay, and transmits a C H A N N E L REQUIRED message to the BSC. The BSC then assigns a signalling channel, and sends a C H A N N E L ACTIVATION message back to the BST with the description of the assigned channel. Upon successful allocation on that channel, the BST responds with a C H A N N E L ACTIVATION acknowledgement. An IMMEDIATE ASSIGN command is sent to the Chapter 2 Previous Work on Location Management 1 ] BTS, which forwards it to the MS as an IMMEDIATE ASSIGNMENT message with the description of the allocated channel. 2. Location update request - The MS sends a LOCATION UPDATE REQUEST message to the BST through the allocated channel. The layer 3 location update request information containing the identity of the MS (IMSI or TMSI), the mobile equipment classmark, LAI, and the type of location update (intra-VLR, inter-VLR with IMSI/TMSI, or attach proce-dure), is sent along with the request message. The BST forwards the layer 3 information along with an ESTABLISH INDICATION message to the BSC, approving the setup of the channel. The BSC then sets up a SS7 connection with the MSC, allowing the MSC to communicate transparently with the MS through the BSS. The BSC sends a C O M P L E T E L A Y E R 3 INFO message to the MSC, containing the location update request information from the MS. 3. Update new location information - The MSC uses the M A P protocol to communicate with the appropriate databases. The MSC sends a MAP_UPDATE_LOCATION_AREA mes-sage to the new VLR, which includes the layer 3 information of the MS. The HLR is then notified of the new V L R through the MAP_UPDATE_LOCATION message. The HLR sends a MAP_CANCEL_LOCATION message to the previous VLR, and requests removal of any subscriber information from the previous VLR, which then returns an acknowledgement back to the HLR. The HLR sends the subscriber information (e.g. sub-scribed services, any restrictions..etc) to the new V L R through the request message MAP_INSERT_SUBSCRIBER_DATA. An acknowledgement is sent by the V L R after successfully registering the subscriber information. 4. Release signalling channel - After a successful acknowledgement of the subscriber infor-Chapter 2 Previous Work on Location Management 12 mation from the new VLR, the HLR sends an acknowledgement of the MAP_UPDATE_LOCATION message, and the V L R in turn sends an acknowledgement of the MAP_UPDATE_LOCATION_AREA message which was previously sent by the MSC. The MSC then sends a LOCATION UPDATE ACCEPT message to the mobile sta-tion to the MS, completing the location update. The allocated channel is released after receiving a C H A N N E L RELEASE message from the MSC. 2.2.3 Paging procedures When there is an incoming call to the mobile network from the public switched telephone network (PSTN), a connection is established to the target MS through its currently serving MSC. To find the correct MSC, the location information needs to be looked up from the HLR. However, the PSTN is not able to directly query network databases such as the HLR, and therefore the HLR is queried through a gateway MSC (GMSC) which is designated for the purpose of identifying the MS's currently serving MSC [28]. After identifying the MSC, the current LAI is retrieved from the database and page messages are broadcast to the cells of the current LAI. The signalling flow of an incoming call delivery is illustrated in Figure 2.3. Chapter 2 Previous Work on Location Management 13 MS BST BSC MSC V L R HLR GMSC PAGING R E Q U E S T C H A N N E L R E Q U E S T ' PAGING C O M M A N D PAGING 1vlAPJ>ROVIDE_ R O A M I N G _ N U M B E R M A P _ S E N D _ INFO_FOR_ INCOMING_CALL| "MAP_PAGE M A P _ P R O V I D E : R O A M I N G . N U M B E R ack I A M ' M A P _ S E N D _ ROUTING_ INFORMATION M A P _ S E N D _ ' R O U T I N G . INFORMATION ack I A M Figure 2.3: Signalling flow of an incoming call delivery from the PSTN The procedure of an incoming call delivery from the PSTN are as follows: 1. Identify call routing information - The PSTN sends an initial address message (IAM), with the mobile station ISDN number (MSISDN) of the target MS, to the GMSC notifying an incoming call. The GMSC then sends the MSISDN through the request message MAP_SEND_ROUTING_LNFORMATION to the HLR. The HLR maps the MSISDN to the subscriber's IMSI, and identifies the registered V L R and thus the corresponding MSC. The HLR sends a MAP_PROVIDE_ROAMING_NUMBER message to the VLR, along with the IMSI of the subscriber. The V L R assigns a mobile station roaming number (MSRN), and maps the IMSI to the MSRN. An acknowledgement to the MAP_PROVIDE_ROAMING_NUMBER with the MSRN is sent back to the HLR, which Chapter 2 Previous Work on Location Management 14 in turn forwards the MSRN to the GMSC as an acknowledgement to the MAP_SEND_ROUTING_ENFORMATION request message. 2. Send paging messages - The GMSC continues with the call routing process by sending an IAM to the subscriber's corresponding MSC with the new MSRN. The MSC then sends a MAP_SEND_INFO_FOR_INCOMING_CALL message to the V L R with the MSRN. This message is not acknowledged until the subscriber successfully responds to the pag-ing. The V L R requests to page the MS by sending a MAP_PAGE message back to the MSC, along with the IMSI and the LAI of the MS stored in the VLR. The MSC then sends a PAGING message to the appropriate BSCs. This message contains the IMSI, TMSI, a list of cells (i.e. BSTs) to be paged, and other additional information such as the type of channel required (e.g. speech or data). The BSC sends a PAGING C O M M A N D message to each BST of the specified list, which in turn transmits a PAGING REQUEST message over the radio interface. When the MS successfully receives the PAGING REQUEST message, a procedure of requesting a signalling channel is initiated which then establishes a connection from the MS to the PSTN. 2.3 Location management using dynamic individualized location areas The algorithm proposed in [13] attempts to utilize the mobility history, contained in the user profile of an individual user, to dynamically create location areas and to determine the paging zone that contains the most probable cells. The user profile consists of a counter N(a, b), which contains the number of transitions the user has made from cell a to cell b, and T(b), which contains the average time per visit that the user has spent in cell b. As the user moves from cell a to cell b, the counter N(a, b) is incremented, and a timer tb starts counting until the user moves out Chapter 2 Previous Work on Location Management 15 of cell b, or the mobile station is switched off. The average value of T(b) is then updated using the final value of tb. The following subsections describe the dynamic location updating algorithm and the paging algorithm proposed in [13]. 2.3.1 Dynamic location updating algorithm The information stored in N(a, b) of the user profile is utilized to dynamically create individual-ized location areas. A location update is triggered when one of the following events occurs: (i) User crosses the location area boundary - When the user enters a cell that is not part of the present location area, a location update is triggered. (ii) Mobile station is first powered on in a cell - Initial location area is formed around the us-er's present cell upon powering on of the mobile station. (iii) Periodically as required by the network - Period length is a parameter defined by the network. When a location update is triggered, the mobile station generates a list of cells to be included in the new location area, and transmits the corresponding cell identifiers along with the identifier of the current cell to the network. The sequence of steps for forming a dynamic location area is outlined as below: 1. As the mobile station enters a new cell and decides that a location update is needed, it searches the user profile to determine whether the user has previously visited that cell. If the user has never visited that cell, a classical location update (similar to current G S M implementations) is performed. Relatively large fixed location areas are pre-defined by the network for such purpose. Chapter 2 Previous Work on Location Management 16 2. If the new cell has been previously visited by the user, the list of its neighbouring cells, along with the number of times the user has moved to each of the neighbouring cells from the new cell (i.e. link weight, N(a, b), for all possible b, where a is the new cell), are retrieved from the user profile. The number of neighbouring cells is normally six, assum-ing that cells are hexagonal in shape. 3. The mean weight, N, of the links to the neighbouring cells is calculated as £ N{a, b) N = b e Nf« — , where NCa is the list of neighbouring cells of cell a, and \NCa\ is the neighbouring cell count. This value is used to distinguish paths that are frequently tra-versed, and paths that are seldom traversed. The mean is a relatively simple way of deter-mining a break between frequently and seldom traversed paths. The cells corresponding to the link weights that are greater than or equal to the mean weight N are included in the new location area, as shown in the example in Figure 2.4 below. The cells are always added in decreasing order of weight. Order of inclusion Included in the new location area Mean weight AT =22.8 0 = New cell - 14 3 25 Figure 2.4: An example of location area cell inclusion with corresponding link weights greater than or equal to N 4. After adding the qualified cells from the first ring of neighbouring cells, steps 2 and 3 are Chapter 2 Previous Work on Location Management 17 repeated for the newly included cells in the order of decreasing link weight. In the case where one of the selected links lead to a cell which is already included in the new location area, the link is ignored as a particular cell needs to be listed in the new location area only once. The steps are repeated until no other qualifying cells remain, or the size of the new location area reaches a maximum, Lmax. The maximum allowed size of a dynamic location area Lmax is defined externally by the network. The effect of defining larger and smaller dynamic location areas is discussed in Chapter 5. 5. Finally, the user's current cell location and the list of cells which make up the new location area are transmitted to the network. 2.3.2 Paging algorithm In this paging algorithm, the paging procedure is divided into two paging steps. In the first step, the most probable cells are paged, and if the user is not found, all cells in the dynamic location area are paged in the second step. The paging algorithm utilizes the values of T(b) of the user profile, in order to determine the paging zone Pz that contains the most probable cells. The problem of assigning cells of the dynamic location area into the paging zone Pz is similar to the problem of selecting the links to include in the location updating algorithm. Upon an incoming call, the average value of T(b) among the cells in the dynamic location area is calculated, then the cells with a T(b) value greater than the average is included in Pz. The cells included in Pz are paged first, and if this attempt is unsuccessful, all cells of the dynamic location area are paged. The maximum normalized delay of this paging algorithm is 2. Chapter 3 Activity-based Mobility Model with Directional Random Walk A mobility model is a model of the movements of mobile users. In the study of location manage-ment algorithms, a mobility model is of crucial importance in quantifying the performances of proposed algorithms. In the past, early users of cellular communications, typically, were highly mobile [28]. For this reason, as well as for their simplicity, mobility models which have been frequently used in the past were completely random, or dependent on many assumed typical values for key variables. However, these models no longer accurately model the movement behaviours of the majority of the users today. As the number of cellular subscribers has grown tremendously, cellular communications are now being used mostly by people whose movement behaviours are more routine [28]. Therefore, to accurately quantify and evaluate the performances of location management algorithms today, a mobility model that can generate realistic movements of an individual user is needed. A number of different mobility models have been proposed in [11], [16], [29]-[33] for evaluating the performances of various location management algorithms. The mobility model described in this chapter is a combination of two existing mobility models - the Activity-based mobility model [30] and the Directional random walk model [29]. 3.1 Description of the mobility model The activity-based approach models a particular user's movements as a sequence of activities, with each activity having an associated time of day, duration, and location. An activity is classi-fied into one of the following nine categories: 18 Chapter 3 Activity-based Mobility Model with Directional Random Walk 19 1. Work 2. Work-related 3. School 4. Serve passenger 5. Shopping 6. Social/Recreation 7. Personal business 8. Return home 9. Other A user is also classified into one of the following four categories: 1. Full-time employed outside the home 2. Part-time employed outside the home, but not a student 3. Student, secondary or post-secondary, possibly employed part-time outside the home 4. Not employed outside the home, and not a student As is further explained in section 3.1.1, the number of user types, the number of activities, and the types of activity may vary depending on the availability of raw data. Each user has defined parameters of home, work, and school locations. These locations can be set randomly within the service area for each user. Users who are employed have another parameter which specifies whether their work locations are fixed from day to day. Users who are employed full-time have additional parameters to specify time to leave for work, and time leaving work. Similarly for students, an additional parameter of time leaving for school is defined. The nominal time parameter values, for each user, is randomly chosen within a user-specified range (e.g. 7 ~ 9am for time to leave for work and 4 ~ 6pm for time leaving work). The actual times of the corresponding events are randomly selected within a user-specified tolerance range (e.g. 30 minutes) centered at the fixed values. The 24-hour day is divided into a user-defined number of time periods of equal lengths. Based on the current time period, user type, and the user's previous activity, the next activity is Chapter 3 Activity-based Mobility Model with Directional Random Walk 20 selected. The transition probability from one activity to another is given in the activity transition matrix, which contains the probabilities of the activity transitions for all time periods, activities, and user types. Once the next activity is selected, the duration of the selected activity is chosen from a distribution based on the current time period, the user type, and the type of activity. The activity duration matrix contains the probability distributions of the activity durations for all time periods, activities, and user types. The activity transition and duration matrices are described in further detail in section 3.1.1. After selecting the next activity and its duration, the corresponding location is selected based on certain heuristics and the type of activity. Since the current location of the user is already known, once the destination location is selected, the intermediate path is determined based on the directional random walk model. In the directional random walk model, the user travels towards the destination in discrete and statistically independent steps, where in each step, the user travels according to several parameters whose values are selected from various mobility distributions. Details of the parameters and distributions are described in section 3.1.3. Upon reaching the destination, the user remains stationary for the selected duration of the activity, and the sequence is repeated. 3.1.1 Activity transition and duration As mentioned earlier, the movement of a mobile user is modelled as a sequence of activities, each with an associated time of day, duration, and location. The next activity of a user is determined based upon the probabilities of the possible activities depending on the previous activity, the user type, and the current time period. The distributions of the transition probabilities are stored in a Chapter 3 Activity-based Mobility Model with Directional Random Walk 21 four-dimensional array called the activity transition matrix, shown in Table 3.1 below. The associ-User type Time period Previous activity Next activity Cumulative probability 5 6 1 0.103448 5 6 2 0.103448 5 6 3 0.137931 5 6 4 0.275862 5 6 5 0.344828 5 6 6 0.551724 5 6 7 0.620690 5 6 8 1.000000 5 6 9 1.000000 5 7 1 0.119048 5 7 2 0.119048 5 7 3 0.142857 Table 3.1: Sample data from the activity transition matrix ated duration of the selected activity is determined similarly using the duration probabilities which depend on the selected activity, user type, and the current time period. Another four-dimensional array called the activity duration matrix, as shown in Table 3.2, contains the probability distributions of all possible activity durations, set at multiples of 5 minutes. Users of User type Time period Activity Duration Cumulative probability 3 6 6 200 0.910448 3 6 6 320 0.925373 3 6 6 345 0.940298 3 6 6 415 0.955224 3 6 6 420 0.970149 3 6 6 450 0.985075 3 6 6 510 1.000000 3 6 7 0 0.081481 3 6 7 5 0.148148 3 6 7 10 0.237037 3 6 7 15 0.340741 3 6 7 20 0.429630 Table 3.2: Sample data from the activity duration matrix Chapter 3 Activity-based Mobility Model with Directional Random Walk 22 type 0 (Full-time employed) and type 2 (Student, possibly employed part-time) have individual time parameters which are used in generating the times at which users go to work, return home from work, or go to school. In order to accommodate these times, the durations of certain activi-ties, in particular the activities prior to going to work, returning home from work, or going to school, need to be adjusted. In such cases the duration values selected from the activity duration matrix are overridden by the adjusted values. An adjustment is made such that the next activity would occur uniformly random at within ±15 minutes of its specified time. The sizes of the activity transition and duration matrices depend on the number of time periods, the number of user types, and the number of activities defined in the mobility model. The probability entries in the above two matrices should be based on data collected through field experiments or user surveys. One way of collecting such data is to conduct a survey of daily trips of users of various types. Each user is provided with a travel diary in which all trips taken during the survey period are recorded. Each recorded trip should include the start time, the arrival time, and the activity type. The person completing the survey also specifies his or her employment status, and whether the work location is fixed. The activity transition and duration matrices used in [30] were constructed using the data derived from a trip survey conducted by the Regional Municipality of Waterloo in 1987. The 24-hour day was divided into twelve 2-hour time periods, and the four user types and nine activity types described earlier in this chapter were used. All simulations in this thesis were performed using the same activity transition and duration matrices as in [30]. 3.1.2 Selection of activity location For all users except those with non-fixed work locations, their work, home, and school locations Chapter 3 Activity-based Mobility Model with Directional Random Walk 23 are pre-defined and fixed. Therefore, among the nine activity categories, the associated locations for the work, school, and return home activities are always known. For shopping, the user selects one shopping area from among a number of shopping areas, each at its own fixed location. Since it is more likely for users to shop at shopping areas that are closer to them, a shopping area is randomly selected from the top three nearest shopping areas from the current location of the user. For the rest of the activities, and in the case of work with non-fixed work location, the associated locations are selected uniformly and at random within the service area. 3.1.3 Intermediate path The intermediate path from a mobile user's current location to the next destination is determined according to the directional random walk model of [31]. After the location associated with the mobile user's next activity is found, the user travels towards the destination in a series of discrete and statistically independent steps, as illustrated in Figure 3.1. A travelling step is a straight path along which the velocity is constant. There are three parameters associated with each travelling step: random angular deviation p\ random velocity v, and random hold time tn. The principal direction is defined by the path from the user's current position to the destination. The direction of each travelling step is set by a random angular deviation p, with p.d.f. about the principal direction. The user then travels in that direction at a constant speed v, chosen at random from a p.d.f. fy( v)> f ° r a duration of th, a random hold time with p.d.f. fn(t). Chapter 3 Activity-based Mobility Model with Directional Random Walk 24 destination Vi x thi starting point Figure 3.1: Intermediate path based on the Directional random walk model The values of the three parameters are randomly selected based on their distributions for each travelling step until the user arrives within a distance dtoi from the destination. The user is then considered to have reached the destination. The distribution of the random angular deviation |3 from the principal direction is modeled by an Alpha p.d.f. with a mean deviation of zero. Thus, /*(P) = 1 a '[1 +(XpP ]tan (apTi) 0 ; -n < P < n ; otherwise (3.1) Tt It The probability of a user travelling in the forward direction, i.e. - - < P < - , is assumed to be 0.95, and therefore the corresponding ap value of 4.2 is used. The random velocity distribution, based on previous radar speedometer measurements in rural areas, is modeled using a Gaussian p.d.f. with a standard deviation cd = 0.17 x v, where v is the mean velocity. Thus, Chapter 3 Activity-based Mobility Model with Directional Random Walk 25 ( v - v ) 2 1 2oj fv(v) = —±=e ' (3.2) where Gd = 0.17 x v. The arrival of events that initiate a new travelling step is modeled to be Poisson. Therefore the hold time of each travelling step, i.e. inter-arrival time of Poisson arrivals, is exponentially distributed with parameter Xh - =, where th is the mean hold time. Thus, h MO = h (3.3) l 0 ; otherwise The radial distance is the distance between the user's current position and the starting position of the trip. The mean radial distance of a mobile user along the intermediate path is a function of average hold time, average velocity, and elapsed time. This function, as derived and approximated in [29], is given by r(v, t, th) = vtcosp + c(v, th), (3.4) where the offset c(v, th) is approximated as c(v, th) = 3.24(v)2 - 37.8(v) + 0.0096(rA)2 + 1.3586(^) + 150.3608. (3.5) The mean value, cos [3, is Chapter 3 Activity-based Mobility Model with Directional Random Walk 26 COS (3 = J c o s p - / f l ( P ) d p . (3.6) Substituting (3.1) into (3.6) yields cos : 1 CCpCOSp 9 2 2 —1 TI z [ 1 + otp P ]tan (C(p7t) JP (3.7) Jp 2 • tan (ap7t) -TC |_1 + ocp (3 which can be numerically integrated to obtain cosP = 0.834. 3.2 Simulation results A simulation of the mobility model was performed so as to visualize the movement patterns of the four different user types. Movements of four mobile users, one of each user type, are collected from a simulated environment consisting of a service area containing 100 cells (10x10) and twelve uniformly spaced shopping locations. All four simulated users travel at a mean velocity v of 25 kilometers per hour with a mean hold time th of 60 seconds, and are assigned with the same home, work, and school locations. The movements generated for each user were collected over 50 simulated days. For each cell, the total number of visits and the total duration of each user during the 50-day period were extracted from the collected data. Figure 3.2 through Figure 3.9 display the results for each user in terms of the total number of visits and the total duration in each cell. Each figure marks the twelve shopping locations as well as the user's work, home, and school locations. Note that the sets of ranges shown in the legend can differ in the figures. Chapter 3 Activity-based Mobility Model with Directional Random Walk 27 As illustrated in Figure 3.2 through Figure 3.5, it is clear that the visits of user type 0 (Full-time employed) and user type 1 (Part-time employed, not a student) are concentrated on the home and work cells and the intermediate path between the two. More variability is shown in Figures 3.4 and 3.5 when compared to Figures 3.2 and 3.3, since users of type 1 spend less time at work than users of type 0. Similarly for user type 2 (Student, possibly employed part-time), as shown in Figures 3.6 and 3.7, the visits are concentrated in home and school cells and the cells along the home/school path. As expected, Figures 3.8 and 3.9 show that the movement pattern for user type 3 (Unemployed, not a student) is not as predictable when compared to the movement patterns of other user types. Nevertheless, the home cell and the cells that are nearer to the home cell appear to have relatively higher number of visits. In terms of shopping cells, the figures show that for all user types, the shopping cells that are nearer to home, work, and school cells have relatively higher duration of stay. The number of visits to these shopping cells may not necessarily be high, since a user may have a high mean duration per visit with low number visits. Chapter 3 Activity-based Mobility Model with Directional Random Walk 28 m 120 + • 90- 119 • 70-89 • 50-69 • 30-49 • 10-29 • 1-9 • 0 Figure 3.2: Number of visits to each cell, for user type 0 (Full-time employed) • 23000 + • 700 - 900 • 500 - 699 • 300 - 499 • 200 - 299 • 100 - 199 • 1 -99 • 0 Figure 3.3: Total duration (in minutes) in each cell, for user type 0 (Full-time employed) Chapter 3 Activity-based Mobility Model with Directional Random Walk 29 • 100 + • 80-99 • 60-79 • 40-59 • 20-39 • 10- 19 • 1-9 • 0 Figure 3.4: Number of visits to each cell, for user type 1 (Part-time employed, not a student) • 3000 + • 700- 1100 • 500 - 699 • 300 - 499 200 - 299 • 100 - 199 • 1-99 • 0 Figure 3.5: Total duration (in minutes) in each cell, for user type 1 (Part-time employed, not a student) ler 3 Activity-based Mobility Model with Directional Random Walk • 120 + • 100- 119 • 70-99 • 50-69 • 30-49 • 10-29 • 1-9 • 0 Figure 3.6: Number of visits to each cell, for user type 2 (Student, possibly employed part-time) • 18000 + • 700 - 999 • 400 - 699 • 200 - 399 • 100 ~199 • 50-99 • 1-49 • 0 Figure 3.7: Total duration (in minutes) in each cell, for user type 2 (Student, possibly employed part-time) Chapter 3 Activity-based Mobility Model with Directional Random Walk 31 • 100 + • 8 0 - 9 9 • 6 0 - 7 9 4 0 - 5 9 • 2 0 - 3 9 • 1 0 - 1 9 • 1 - 9 • 0 Figure 3.8: Number of visits to each cell, for user type 3 (Unemployed, not a student) • 56000 + • 800 - 1200 • 600 - 799 • 400 - 599 • 200 - 399 • 100 - 199 • 1 - 9 9 • 0 Figure 3.9: Total duration (in minutes) in each cell, for user type 3 (Unemployed, not a student) Chapter 4 Intelligent Location Management Algorithm Location management is required for effective delivery of incoming calls for a mobile user. The two basic operations of location management are location updating and paging. Both operations involve radio signalling between the mobile station and the base station, thus consuming radio resources. The goal of developing a new location management algorithm is to reduce, under given conditions, such consumption of radio resources. The location management algorithm described in this chapter makes use of the fact that an average mobile user has a limited number of frequently visited locations. In most cases, a user goes on a trip for a particular purpose, usually a type of activity such as work, school, or shopping, occurring at a particular location. The possible locations where such activities can take place are limited. As an example, the shopping locations are located at specific sites, and for most users, the work, school, and home locations are fixed. The home location normally acts as the hub from which most trips are taken, and it is also a place where a user spends a significant amount of time. Frequently visited locations of a user are not pre-defined in location management, since this information is not always collected from cellular customers. Therefore, it is the responsibility of the location management algorithm to intelligently determine such information. The intelligent algorithm described in this chapter determines these frequently visited locations of a user at a given time of day based on the individual user profile. The user profile is a record of the mobility history of the user. Section 4.1 explains the details of the data that are collected and maintained by the user profile. Knowledge of the frequently visited locations of a user at a given time of day can significantly cut down on the radio resources due to paging, when compared to the conventional 32 Chapter 4 Intelligent Location Management Algorithm 33 flood paging. Furthermore, by dynamically creating location areas that consist of frequently visited cells and their intermediate cells, the radio resource consumption due to location updates can be reduced significantly when compared to the conventional method of defining fixed location areas. Detailed numerical results and analyses of the radio signalling reductions are presented and discussed in the next chapter. Several factors determine the performance of a location management algorithm. These can be described as bandwidth due to location updating and paging, call setup delay, computational complexity in the mobile terminal, and memory requirements at the mobile terminal and the network databases. In this algorithm, priority is given to minimizing the bandwidth due to location updating and paging, while maintaining the remaining factors at an acceptable level. The additional computational complexity and memory requirements are discussed in this chapter. The location management algorithm described in this chapter is an improvement on an existing algorithm described in [13]. The location updating algorithm of [13] remains unchanged, while a new paging algorithm with added intelligence is introduced. The description of the location updating algorithm of [13] is reviewed in Section 2.3.1. Definition of the user profile is modified accordingly to accommodate the new paging algorithm. 4.1 User profiles The mobility history of an individual user is stored in its user profile. The mobility history consists of the number of transitions the user has made from one cell to another, and the total duration the user stayed in each cell at a given time period of the day. The definitions of these data components are as follows: Chapter 4 Intelligent Location Management Algorithm 34 • N(a, b) - a counter which stores the number of times the user has moved from cell a to each neighbouring cell b, also called the link weight. • T(b, k) - a timer which stores the user's total duration of stay (in minutes) in cell b, during time period k. • te- elapsed time since the last location update or page. • Cprev - the user's previously known cell location. As the user moves from cell a to cell b, the counter N(a, b)is incremented by 1, and the timer T(b, k), starts incrementing every minute starting from its previous value. The timer stops when the user moves out of cell b, or the user's mobile station is switched off. During a location updating procedure or after a paging procedure, the network stores the user's residing cell number into Cprev, and starts the timer te from 0. The 24-hour day is broken down into a user-defined number of time periods. Defining a large number of short time periods (e.g. 5-10 minutes), allows the user profile to build up mobility data that is more sensitive to the time of day, when compared to defining a small number of long time periods. The disadvantage is that the profile takes longer to build for capturing the user's mobility patterns. However, this problem disappears once the profile is built up after a certain learning duration. Chapter 6 discusses the time that is required for building a user profile, which sufficiently captures the user's mobility patterns, for various time period lengths. In order to reduce the time needed to adapt to drastic changes in user mobility patterns, the user profile incorporates a moving window scheme which employs the concept of aging of the user profile data. When a data entry reaches a certain age (e.g. 50 days), it is considered outdated and is purged from the user profile. This reduces the influence of previously active cells which Chapter 4 Intelligent Location Management Algorithm 35 may no longer be important. As an extreme example, consider a user moving to a new home or switching his or her work location. The total number of visits and the total duration of stay for the previous home or work cell are reduced in time, and after a certain span of time (e.g. 50 days), the old values will disappear. This span of time is referred to as the window size. The counters and timers are stored in non-volatile memory of the mobile station (e.g. SIM card of a G S M handheld) as well as in the network database. Specifically, the counter N(a, b) is stored at the mobile station for location updating purposes (see Section 2.3.1), and the timer T(b, k) is stored on the network side for paging purposes (see Section 4.2). Although T(b, k) is stored on the network, the mobile station is responsible for collecting and transmitting the relevant information. It is always desired to have up-to-date T(b, k) values on the network to perform more efficient paging. However, since transmitting the T(b, k) values to the network involve radio signalling, this data should be sent as infrequently as possible. Therefore, the mobile station maintains a small version of T(b, k) which stores a day worth of information, which is then transmitted to the network daily during a low-traffic period (e.g. 2am ~ 6am). The network collects and accumulates this data on to the actual T(b, k) values. Storage and transmission requirements of the user profile are explained below: Storage requirement of the user profile on the mobile station A day worth of Tib, k) : Number of entries = Number of cells in the system x Number of Time periods Maximum value per entry = 24 hrs x 60 min/hr = max. 1440 minutes(i.e. 11 bits per entry) e.g. For a 100 hexagonal cell system with 30 minute time periods Required memory for T(b, k)= 100 cells x 48 time periods x 11 bits Chapter 4 Intelligent Location Management Algorithm 36 = 52800 bits, or 6.45 Kbytes For N(a, b) : Number of entries = Number of cells in the system x Number of neighbours per cell x Window size Maximum value per entry = max. allowed visits to a cell per day e.g. For a 100 hexagonal cell system with 30 minute time periods, a window size of 50 days, and a max. allowed visits to a cell of 30 visits/day Maximum value of N(a, b) per entry = max. 30 visits/day (i.e. 5 bits per entry) Required memory for N(a, b) = 100 cells x 6 neighbours per hexagonal cell x 50 days x 5 bits per entry = 150000 bits, or 18.32 Kbytes Total memory required on the mobile station = 6.45 Kbytes +18.32 Kbytes = approx. 25 Kbytes Storage requirement of a single user profile on the network For T(b, k) : Number of entries = Number of cells in the system x Number of Time periods x Window size Maximum value per entry = 24 hr/day x 60 min/hr = max. 1440 minutes (i.e. 11 bits) e.g. For a 100 cell system with 30 minute time periods and a window size of 50 days Storage required on the network for a single user profile = 100 cells x 48 time periods x 50 days x 11 bits per entry = 264000 bits, or 322.3 kbytes Chapter 4 Intelligent Location Management Algorithm 37 Transmission requirement Size of transmission data per day = One day worth of T(b, k) Transfer time = Size of transmitted data / Transmission speed e.g. One day worth ofT(b, k) is 52800 bits, and the transmission speed is at 10 Kbits/sec Transfer time = 52800 bits / 10240 bits per sec = 5.16 seconds 4.2 Intelligent paging algorithm When the current G S M system attempts to locate a user, the system simply broadcasts paging messages to all cells in the user's location area. In such a case, the cost of paging is directly related to the size of the location area. However, paging costs may be reduced by using the knowledge of a user's frequently visited cells. The paging algorithm proposed in this section significantly reduces the average paging cost by utilizing the mobility information stored in the user profile. The basic idea of any improved paging algorithm is to divide the paging procedure into multiple paging steps, so that the system pages only a subset of the cells within the location area during each step. The intelligent part of the proposed algorithm focuses on the process of grouping the cells into different paging steps, so that a user may possibly be reached after paging a minimal number of cells. The probability of finding a user in a given cell during a given time period, is determined using the information stored in the timer T(b, k) of the user profile. Let Ld be the list of cells making up the user's dynamic location area. We define the cell probability Pc(b\k,Ld) as the corresponding probability of cell b of location area Ld, in which the user may reside in during time period k, i.e. Chapter 4 Intelligent Location Management Algorithm 38 Pc(b\k,Ld) = Tib, k) (4.1) The probability of a given cell may differ during different time periods. The cell probability at a given time period is independent of the probability of the same cell at a different time period. This is obvious in the case of the work cell of a full-time employed user, as an extreme example. During work hours, a full-time employed user may be found at work with a high probability, whereas at other times, the probability of finding the same user at work may be very low. It is important to obtain cell probabilities as a function of time, because otherwise in the case of the above example, the system will page the user at the work cell with a high probability even during off-hours, as the work cell is a place at which the user spends a great deal of time overall during a 24-hour day. The first step of grouping the cells is to divide the location area into two paging zones, Zj and Z 2 . The deciding factors for this particular grouping of cells are te, the elapsed time since the last location update or page, and Cprev, the user's previously known cell location. With the value of te, we can calculate re, the mean distance the user travels during a time period te, assuming that the user moves according to the directional random walk model towards the same principal direction without stoppage. In Section 3.1.3, the expression for the mean radial distance r(v, t, th) of the directional random walk model was presented. The distance re is equivalent to the mean radial distance r(v, t, th), with parameter t replaced by the elapsed time te. Hence, re = r(v, te, th) (4.2) Chapter 4 Intelligent Location Management Algorithm 39 Upon a paging event, the network fetches the values of te and Cprev, then forms a circular region centered at Cprev, with radius rj, where rj = ae • rg, and ae is the scaling factor of the circular region defined by the network. The cells of the current location area which fall within the circular region are included in paging zone Z\, while the remaining cells are included in Z 2 , as shown in Figure 4.1 below. Dynamic location area Figure 4.1: Dividing the location area into two paging zones, Zj and Z2, based on re'. Suppose a user moves towards a destination outside the circular region with ae = 1, starting from the center of Cprev When the user moves without stoppage, on average, the user can be found at some point along the circular boundary after a duration of te. However, in most cases the user remains stationary in one or more particular cells for some portion of te, or in some cases, all of te. For this reason, chances of locating the user in paging zone Zj is very high. Simulation results Chapter 4 Intelligent Location Management Algorithm 40 have shown that during paging events, the user is found in paging zone Zj in over 97 percent of the occurrences for ae = 1. Section 4.3 presents the simulation results in greater detail. Note that the value for te can get large enough for the circular region to cover every cell in the location area. In such a case, all cells belong to paging zone Z 1 ; and paging zone Z 2 is empty. After dividing the location area into two paging zones, the network further partitions paging zone Zj into three smaller paging zones, Zn, Z 1 2, and Z 1 3, such that the network pages the cells of Zj in three successive steps. The paging order of the location area is Z n , Z 1 2 , Z 1 3 , and finally Z2, a maximum of four paging steps. If the network finds the user during one of the earlier paging steps, evidently there is no reason to proceed to the succeeding steps, and therefore the paging procedure terminates. Given the above conditions, together with the probabilities of the cells in Z 1 ; we wish to find the optimal configuration of cells such that the mean paging traffic is minimized. A cell configuration defines the sizes of the three paging zones and the list of cells that are assigned to each of them. We define Tz^, the normalized mean paging traffic for paging Zj, as follows: TZx = Pn • \Zn\ +Pn • (|Z„| + |Z12|) + P 1 3 • (|Zn| + |Z12| + |Z13|) (4.3) where IZ1(I is the size of paging zone Z1(- in terms of number of cells. The probability of locating the user in paging zone Z1(- is defined as, p u = X Pc(b\k>Z0- (4.4) beZu Figure 4.2 shows a cell configuration of the paging zone Zj, consisting of m cells. In order to find Chapter 4 Intelligent Location Management Algorithm 41 the optimal cell configuration, the cells must be arranged in decreasing order of probabilities, and the two break points which divide the cells into three groups chosen such that the corresponding normalized mean paging traffic Tz is minimized. 1 — 2 1 1 — 1 [Z 1 2 || Z13 | Cj... Ci ci+l • • • cj cj+l • • • cm i '— Break points —' Figure 4.2: A cell configuration of the paging zone Z j , consisting of m cells The cells are sorted in decreasing order of probability, i.e. c n > c n + x for all n<m , since cells with higher probabilities should be paged earlier than cells with lower probabilities. We can prove this using (4.3). Suppose that the cells are sorted in decreasing order, and we swap cell cx from Z n with cell cy from Z 1 2 , where cx > cy. The values of the terms \ZU\ and (IZ n l + IZ12I) remain unchanged, but P\\ decreases by (cx - cy) while / > j 2 increases also by (cx - cy). The normalized mean paging traffic for this modified configuration, Tz', is expressed as Tzl = ( P n - Ac) • | Z n | + ( P 1 2 + Ac) • ( | Z n | + |Z 1 2 | ) + P 1 3 - ( | Z n | + | Z 1 2 | + |Z 1 3 | ) (4.5) where Ac = c x - c y . The change in normalized mean paging traffic, Tz^ - Tz^, is thus Tz, ~ ?z; = ^ 11 • P i i | + Pn • (|Z n| + |Z 1 2 | ) + Pu • (|Z n| + | Z 1 2 | + |Z 1 3 | ) Chapter 4 Intelligent Location Management Algorithm 42 -{Pn - Ac) • \Zn\ - (Pn + Ac) • ( | Z H | + |Z 1 2 | ) - P 1 3 • ( | Z „ | + | Z 1 2 | + |Z 1 3 | ) = ( -Ac - |Z 1 2 | ) (4.6) Since the values of both Ac and IZ12I are greater than zero, the change in normalized mean paging traffic shown in (4.6) is negative. This means that 7*z' is greater than r Z ] , and the modified configuration yields a higher mean traffic when compared to the original configuration. Therefore, it is required that the cells be ordered in decreasing probabilities as the first step to minimizing the mean paging traffic. The next step to minimizing the mean paging traffic is to determine the positions of the two break points such that the value of T Z ] is minimized. One way of doing this is to calculate the mean paging traffic Tz^ for all possible configurations of Zj , given that the cells are sorted in decreasing probabilities, then to select the configuration which produces the least amount of traffic. Using this brute-force method, the number of configurations that needs to be checked for Zj is ^ Z ' 2 _ 1 j = — — ^ * ^ z ' l 2 ) . For example, the number of configurations for a Zj size of 20 cells, is ^ 2 9j = 1 9 * 1 8 = 171. For a faster search for the optimal configuration, we introduce a simple algorithm which reduces the number of configurations that needs to be checked. Referring to Figure 4.2, the steps below describe this simple algorithm: 1. Place the break points at positions / = 1, and j = i+ 1. Find Tz^. 2. Increment j by 1. Find T Z ) . Repeat this step until the new Tz is higher than the previous Chapter 4 Intelligent Location Management Algorithm 43 3. Increment /by 1, let j - i + 1. Find r Z ) . If the new r Z ] is higher than the r Z ] of / = i - 1, and j = i+ 1, stop. Otherwise, repeat from step 2 until i reaches m - 2. In all steps, the minimum value of Tz^ and its corresponding configuration needs to be maintained. When the algorithm stops, the configuration with the lowest T Z | is selected. The number of configurations that are checked varies depending on the cell probabilities of Zj . Using the cell probabilities that are results of the mobility model of Chapter 3, simulations have shown that the number usually ranges from 30 to 40 for a Zj size of 20 cells, compared to 171 for the brute-force method. The assumption in this algorithm is that when the cells are sorted in decreas-ing order of probability, the cost curve as a function of break point position is always concave up. In other words, if a break point is set at a position where the cost is optimal, the cost always increases as the breakpoint moves away from its optimal position. This assumption was verified using computer simulation. 4.3 Simulation results Simulation of the intelligent algorithm was performed to analyze the effects of varying the value of ae. The results in terms of paging traffic and delay are presented and discussed in the next chapter. This section presents and discusses the mean sizes of the paging areas, and the probabil-ity of finding the user in paging zone Zj for different values of ae. The simulation was based on a 100 hexagonal cell (10x10) system, with 100 simulated users (25 of each user type), for a duration of 50 simulated days. All simulated users travelled at a mean velocity v of 25 kilometers Chapter 4 Intelligent Location Management Algorithm 44 per hour with a mean hold time th of 60 seconds. The maximum location area size Lmax was set at 20. Call arrivals were modeled to be Poisson, where the mean call arrival rate X c a / / was set at 6 calls/day. From the simulation, the mean location area size was 18.09 cells out of a maximum of 20 cells. Figure 4.3 indicates that the mean size of paging zone Z] is much greater (14 ~ 16 cells) than the size of paging zone Z 2 (2 ~ 4 cells). This means that the value of the elapsed time te, on average, is relatively large such that the circular region with radius rj covers a large portion of the location area. o JQ E 3 z 20 18 16 14 12 10 8 6 4 2 0 3.866 11.125 2.030 1.066 3.583 11.369 2.066 1.069 3.353 11.565 2.097 1.072 3.158 11.732 2.124 1.074 2.974 11.887 2.150 1.076 2.840 12.004 2.166 1.077 2.706 12.117 2.186 1.079 2.604 12.206 2.197 1.080 0.8 0.9 2.490 ' 12.302 2.214 1.081 1.1 1.2 1.3 1.4 1.5 1.6 CLe m IZ2I • IZ13I • IZ12I • I Z n l Figure 4.3: Distribution of the sizes of the paging zones for various ae Chapter 4 Intelligent Location Management Algorithm 45 For all values of ae, the paging zone Z n only contains about one cell. This means that in the first paging step, the intelligent algorithm nearly always pages only one cell, which likely is one of work, school, or home cells depending on the user type and the time of the day. Figure 4.4 shows the probability of finding the user in paging zone Zj for different values of ae. For ae = 1, on average the user stays within Zj with a probability of 0.97. However, note that Figure 4.4 shows that for ae = 1, Zj adds up to about 14.7 cells, which is only 81% of the total 18.09 cells in the location area. This means that during an incoming call, with a mean probability of 0.97, the network pages only within 14.7 cells, whereas if we did not incorporate the idea of elapsed time te, the network pages within all 18.09 cells of the location area. Figure 4.4: Probability of finding user in paging zone Zj for various ae We can expect the mean size of Zj to fluctuate when we vary the values of either one or both of the call arrival rate and the maximum location area size Lmax. A higher call arrival rate Chapter 4 Intelligent Location Management Algorithm 46 yields shorter inter-arrival times between pages, and a smaller Lmax shortens the inter-arrival times between location updates. Shorter inter-arrival times between pages and location updates results in lower values of te, and thus yield smaller sizes of Zj. Smaller sizes of Z] likely result in less paging traffic. Especially when we reduce the value of Lmax, not only does the size of Zj decrease, the mean size of the dynamic location area decreases as well. This further reduces the paging traffic, although a balance with the addition of location update traffic is needed to achieve the lowest overall bandwidth consumption. Chapter 5 Performance Analysis Results One of the important goals of location management is efficient utilization of the limited radio spectrum. This goal can be achieved by minimizing the location updating and paging traffic that is transmitted across the radio interface. In this chapter, the proposed intelligent location manage-ment algorithm is compared with existing algorithms in terms of radio resource consumption. Location management of G S M with three types of fixed location area layouts, the dynamic individualized location management algorithm proposed in [13], and the algorithm proposed in this thesis are analyzed under a common simulated environment. In addition, the sensitivity of the performances with variations in several mobility and location management parameter values is analyzed. For convenience, from this point and on we will refer to the three types of G S M location management algorithms as GSM1, GSM2, and GSM3, the dynamic individualized location management algorithm of [13] as the dynamic algorithm, and finally the proposed algorithm in this thesis as the intelligent algorithm. All simulation results presented in this chapter have 99% confidence intervals within ± 5% of their respective mean values. 5.1 Description of the simulated environment The simulated environment consists of one hundred (10x10) hexagonal cells with a radius of 1.5 kilometers each, twelve uniformly spaced shopping locations, and 100 simulated users. The 1998 American community survey (ACS) of several large cities in the U.S. was obtained from the U.S. Census Bureau [34], in order to determine a realistic distribution of the user classes for the simulated environment. According to the survey, 49.4 percent of the population were employed, while 25.2 percent were students. Only about 2 percent of the population had jobs which did not have fixed work locations. There are no indications of the ratio of full-time and part-time 47 Chapter 5 Performance Analysis Results 48 employed workers in the survey, therefore in this simulation, we divide the employed population into full-time and part-time evenly. The 100 cells in the system are grouped into three different fixed location area layouts, designated for the three types of GSM algorithms, GSM1, GSM2, and GSM3. The fixed layout for GSM1 has 100 location areas of only one cell each, and the fixed layout for GSM2 has 10 location areas of 10 cells each. The fixed layout for GSM3 has 5 location areas of 20 cells each, and this layout is also used by the dynamic algorithm and the intelligent algorithm when users perform location updates from cells which were never visited. All simulated users travel at a mean velocity v of 25 kilometers per hour with a mean hold time th of 60 seconds. To model the arrival of calls, a Poisson process with rate X c a / / calls per unit time is used. Each simulation run generates a log of the number of location updates, the number of paging messages, and the average paging delay for all users of each simulated day, for a total of 50 days. 5.2 Performance comparisons of location management algorithms The mean values of the location updating traffic, paging traffic, and normalized paging delay of a user for the three G S M algorithms and the dynamic algorithm are compared with the proposed intelligent algorithm. In addition, a function for the total location management cost is defined in terms of location updating and paging traffic, and is used to compare the overall performance of the different algorithms. Paging delay is not considered in the total cost function because the goal of the proposed algorithm is mainly to minimize radio bandwidth, and paging delay does not relate to radio signalling. Nevertheless, paging delay is a significant component of location management costs, as it affects the setup time for the incoming calls to mobile users. The proposed algorithm does not attempt to optimize paging delay, however, its value should be kept Chapter 5 Performance Analysis Results 49 at a relatively low level. Section 5.2.3 covers the analysis of the different algorithms in terms of mean and maximum paging delay. 5.2.1 Location updating traffic The mean location updating traffic per user for the different algorithms are compared in Figure 5.1:. G S M 1, with only one cell per location area, has the highest number of location updates since every cell crossing triggers a location update in GSM1. GSM2, with 10 cells per location area, performs much better than GSM1, and GSM3 performs the best among the fixed G S M algorithms. As expected, the location updating traffic for the G S M algorithms are directly related to the fixed location area sizes. 1400 -GSM 1 -GSM 2 -GSM 3 - Dynamic/Intelligent Figure 5.1: Mean number of location updates per user versus elapsed time The dynamic algorithm and the intelligent algorithm uses a common dynamic location Chapter 5 Performance Analysis Results 50 updating algorithm. To demonstrate the efficiency of the dynamic location updating algorithm, the parameter for the maximum location area size Lmax was set to 20 cells, matching the size of GSM3's fixed location area. The resulting mean dynamic location area size was found to be approximately 18 cells. Yet, the dynamic location updating algorithm outperforms GSM3 after an initial learning period of 3 to 4 days. The initial learning period exists because of the time that is required for the user profile to build up and provide better information for cell selection. Figure 5.2: below shows a close up view of the learning curve of the dynamic location updating algorithm. Elapsed Days Figure 5.2: Learning curve of the dynamic location update algorithm The dynamic location updating algorithm performs worse than GSM3 for the first 4 days, and then shows a steady improvement. Chapter 5 Performance Analysis Results 51 5.2.2 Paging traffic Figures 5.3: to 5.6: show the mean number of cells paged per user for the different algorithms. Results for incoming call arrival rates X , c a / / of 3, 6, 9, and 12 calls per day are shown in Figures 5.3:, 5.4:, 5.5:, and 5.6:, respectively. The results for GSM1 show that it has the lowest paging costs, since there is only one cell per location area and therefore the current cell is always known. GSM3 has the highest mean number of cells paged among the different algorithms due to its large location areas. For the GSM algorithms, since the number of cells paged per call is equal to the size of the location area, the number of cells paged per user is simply the size of the fixed location area x XcaU (calls/day) x Elapsed days. The results of GSM algorithms increase linearly with the call arrival rate. Elapsed Days Figure 5.3: Mean number of cells paged versus elapsed time, for X c a „ = 3 calls/day Chapter 5 Performance Analysis Results 52 lie Elapsed Days Figure 5.4: Mean number of cells paged versus elapsed time, for Xcall = 6 calls/day Elapsed Days Figure 5.5: Mean number of cells paged versus elapsed time, for Xcall = 9 calls/day Chapter 5 Performance Analysis Results 53 14000 Elapsed Days Figure 5.6: Mean number of cells paged versus elapsed time, for Xcall = 12 calls/day The average dynamic location area size for both the dynamic algorithm and the intelligent algorithm was approximately 18 cells, which is comparable to GSM3 with 20 cells per location area. However, the paging algorithms of the dynamic and the intelligent location management algorithms performed much better than GSM3 and even GSM2. Furthermore, even though it uses the same dynamic location areas as the dynamic algorithm, the intelligent algorithm outperforms the dynamic algorithm by about 45 percent at a call arrival rate A . c a / ; of 6 calls per day. For the intelligent paging algorithm, the scaling factor of the circular region ae was set to 1.4, and the 24-hour day was divided into 30 minute time periods for a total of 48 time periods. As with the G S M algorithms, the mean number of cells paged for the dynamic algorithm is directly proportional to Xcall. However, the paging traffic for the intelligent algorithm is not directly proportional to Chapter 5 Performance Analysis Results 54 Xcall. This is discussed further in Section 5.3.4, where the effect on the intelligent algorithm performance for different call arrival rates are analyzed. 5.2.3 Paging delay One of the trade-offs involved in minimizing radio bandwidth in the intelligent algorithm is an increased paging delay. Although paging delay does not relate to radio signalling, it affects the call setup time for a mobile user, and therefore it is desirable to keep the delay at a reasonably low level. The G S M algorithm uses flood paging which simply broadcasts paging messages to all cells in the location area. The normalized paging delay in the case of GSM is always 1, since the user is always found on the first try. For the dynamic algorithm, the dynamic location area is broken down into two paging zones, allowing a maximum normalized paging delay of 2. According to the wireless mobile communication standard adopted in North America, an incoming call for a mobile user needs to be set up, in the worst case, within 10 to 12 seconds [18]. In general, the elapsed time for the cellular system to detect whether or not a mobile station responded to a paging message is 3 seconds [18]. For this practical reason, the intelligent algorithm divides the dynamic location area into a maximum of four zones (only three zones when paging zone Z 2 is empty), allowing a maximum normalized paging delay of 4. The maximum normalized delay allowed in the intelligent algorithm may seem relatively high especially when compared to G S M and the dynamic algorithm. Yet, the resulting mean normalized delay is fairly low. Figure 5.7: shows the mean normalized delay of the different algorithms for various call arrival rates. Chapter 5 Performance Analysis Results 55 — * - -GSM —e— - Dynamic — X - - Intelligent 0 3 6 9 Call arrival rate Figure 5.7: Mean normalized delay versus call arrival rate For call arrival rates of 3, 6, 9, and 12, the mean normalized delay of G S M is always equal to 1 as expected, while the delay of the dynamic algorithm is slightly higher, consistently at around 1.15. The intelligent algorithm has the highest delay, ranging from 1.26 to 1.29 for the different call arrival rates. According to Figure 5.7:, the mean delay decreases slightly with the call arrival rate. This is a result of defining smaller sizes for paging zone Zj at higher call arrival rates. With an increase in Xca[l, the mean inter-arrival time of incoming calls, and hence the mean value of te, is reduced. Since the radius of the circular region which defines Zj is directly related to te, the number of cells in Zj decreases as te is reduced. Paginj jstep 1 2 3 4 Dynamic 86% 14% Intelligent 84% 6% 9% 1 % Table 5.1: Success rate of each paging step for the dynamic algorithm and the intelligent algorithm Chapter 5 Performance Analysis Results 56 Table 5.1: shows the probability distribution of the normalized paging delay for the dynamic algorithm and the intelligent algorithm. The dynamic algorithm, with a maximum normalized delay of 2, is successful at the first paging step for 86 percent of the incoming calls. Despite the high maximum normalized delay of 4, the intelligent algorithm achieves a success rate of 84% at the first paging step, while the delay equals 4 for only 1% of the incoming calls. 5.2.4 Total location management cost The dynamic location updating algorithm performs best in terms of location updating traffic, although in terms of paging traffic, GSM1 outperforms all the other algorithms. To assess the combined effect of location updating and paging traffic, we define the total location management cost as a linear combination of the two. As the goal of the proposed algorithm is to reduce radio bandwidth, paging delay is not considered in the total cost function as long as the delay is at an acceptable level. The total location management cost is given by: CTotal = c ' TLU+TPG t5-1) where TLTJ is the location updating traffic, TPG is the paging traffic, and c is a constant represent-ing the relative cost of a location update to a paging message. A location updating procedure is considered more expensive in terms of signalling than just paging a cell, and therefore we analyze the total cost of each location management algorithm using 5 and 10 as the two representative values of c [13]. Figures 5.8: to 5.15: illustrate the comparisons of the total costs for the different algorithms. Figures 5.8:, 5.9:, 5.10:, and 5.11: show results for incoming call arrival rates X , c a / ; of 3, 6, 9, and 12 calls per day, respectively, with constant c = 5. The relative costs of the different Chapter 5 Performance Analysis Results 57 algorithms differ considerably with different call arrival rates Xcall. At 3 incoming calls per day, and for c = 5, GSM1 has the highest cost, as shown in Figure 5.8. This is a result of its high location updating traffic, which its optimum paging traffic could not compensate at a Xcall of only 3 calls per day. However, at a Xcall of 6 calls per day, GSM1 outperforms GSM3, and at Xcall of 9 and 12 calls per day, it performs the best among the G S M algorithms. GSM3, which has the highest paging cost, is better than GSM1 and slightly worse than GSM2 at Xcall of 3 calls per day. Its performance, however, drops rapidly as the number of incoming calls increases. Chapter 5 Performance Analysis Results 58 8000 Elapsed Days Figure 5.9: Total location management cost versus elapsed time, for c = 5 and Xcall = 6 calls/ day 12000 -, 1 Elapsed Days Figure 5.10:Total location management cost versus elapsed time, for c = 5 and Xcall = 9 calls/ day Chapter 5 Performance Analysis Results 59 14000 Elapsed Days Figure 5.11:Total location management cost versus elapsed time, for c = 5 and Xcall = 12 calls/ day The dynamic and the intelligent algorithms outperforms the G S M algorithms by a signifi-cant margin for c = 5 and for 7icall = 3, 6, 9, and 12 calls per day. The intelligent algorithm provides a 18% improvement in total cost over the dynamic algorithm at A , c a / ; of 3 calls per day, and the improvement increases to 36% at X. C f l / / of 12 calls per day. This is expected since the improvement in total cost is a result of less paging traffic, and at higher call arrival rates, the improvement in paging traffic becomes more apparent. Figure 5.12: to 5.15: show results for incoming call arrival rates \ a u of 3, 6, 9, and 12 calls per day, with c = 10. At c = 10, GSM1 has the worst total cost for Xcall = 3, 6, and 9; for hcall = 12, it outperforms GSM3. This is because GSM1 generates the largest number of location Chapter 5 Performance Analysis Results 60 updates, and location updates are now relatively more expensive. GSM3 performs better, since its high paging traffic is offset by its low location update traffic. The figures show that the dynamic and the intelligent algorithms still outperform the other algorithms by a significant margin at c = 10. The improvement of the intelligent algorithm over the dynamic algorithm is now smaller since more weight is placed on location updates, and the two algorithms use a common location updating algorithm. At Xcau = 3 and 12, the intelligent algorithm yields 10% and 30% improvement over the dynamic algorithm. Chapter 5 Performance Analysis Results 61 14000 Elapsed Days F igu re 5.13:Tota l loca t ion management cost versus e lapsed t ime, f o r c = 10 and \ c a l l = 6 ca l l s/ day 14000 Elapsed Days F igu re 5.14:Total loca t ion management cost versus e lapsed t ime, f o r c = 10 and Xcall = 9 ca l l s/ day Chapter 5 Performance Analysis Results 62 16000 Elapsed Days Figure 5.15:Total location management cost versus elapsed time, for c = 10 and Xcall = 12 calls/day 5.3 Sensitivity analysis There are several mobility and location management parameters which may affect the perfor-mance of the proposed intelligent algorithm. In Section 5.2, all simulation results were obtained using fixed values for parameters such as mean velocity v, time period length, maximum dynamic location area size Lmax, and scaling factor of the circular region ae. The effects of variations in mean velocity v, time period length, dynamic location area size Lmax, scaling factor of the circular region ae, and call arrival rate X- c a^ on the performance of the intelligent algorithm are analyzed in this section. Chapters Performance Analysis Results 63 5.3.1 Effect of different mean velocities According to the Region of Waterloo trip survey, over 85% of all trips were made using a motorized vehicle [28]. Considering traffic lights, stop signs, and traffic congestion, the mean velocity should be set relatively low. The mean velocity in Section 5.2 was set to 25 kilometers per hour. Since this is a rough estimate, the resulting performances using different mean velocities are compared. Figure 5.16: illustrates the sensitivity of location update traffic to changes in mean velocity. 1 _ 1400 -i 0) (0 3 1200 -0) Q . CO 1000 -« T J Q . 3 800 -C O *3 600 -CC O O —I 400 -o CO n 200 -E 3 Z 0 -10 15 20 25 30 Mean velocity 35 GSM 1 GSM 2 GSM 3 Intelligent/Dynamic Figure 5.16:Mean number of location updates per user versus mean velocity, over a period of 50 days The location update traffic increases with mean velocity for all the algorithms, although each algorithm shows a different rate of increase. At a higher mean velocity, the mean length of each travelling step in the directional random walk model (Figure 3.1:) is correspondingly higher. This means that there are higher fluctuations within the intermediate path from the starting point to the Chapter 5 Performance Analysis Results 64 destination at a higher mean velocity, thus increasing the possibility of the user exiting the cell and possibly the location area of each travelling step. GSM1 shows the highest rate of increase, while GSM3 has the lowest. The rate of increase is inversely proportional to the fixed location area size of the G S M algorithms, because the higher fluctuations within the intermediate path have less impact with larger location areas. The location update algorithm for the intelligent and dynamic algorithms generates a mean location area size of approximately 18 cells (see Section 4.3), which is comparable to the 20 cells of GSM3. However, the rate of increase is slightly higher than that of GSM3, due to the irregular shapes of the dynamic location areas. Figure 5.17: shows the mean paging traffic over a period of 50 days, for different mean velocities. For the GSM algorithms, paging traffic is not affected by the change in velocity, since the number of cells paged is always fixed at the time of an incoming call. However, there are slight improvements with increasing velocity for the dynamic and intelligent paging algorithms. 7000 10 15 20 25 Mean velocity • - G S M 1 — • - -GSM 2 —• - GSM 3 — © - - Dynamic - Intelligent Figure 5.17:Mean number of cells paged versus mean velocity, for Xcall = 6 calls/day, over a period of 50 days Chapter 5 Performance Analysis Results 65 According to the dynamic and intelligent paging algorithms, destination cells in which the user spends relatively large amounts of time have higher priorities in paging over the cells of their intermediate paths. Since an intermediate path may consist of several cells, it is more expensive to locate a user while the user is travelling on the intermediate path. At higher mean velocities, a user reaches travel destinations faster and spends less time on intermediate paths. This increases the probability of finding the user in destination cells. 5.3.2 Effect of different time period lengths For the intelligent algorithm, the time period length determines how sensitive the user profile data is to the time of day. The results of Section 5.2 were based on 30 minute time periods (48 time periods in one day), which was selected based on an assumption that it covers the uncertainties of the occurrence times of certain events of a user, such as going to work and returning home from work. Defining smaller time periods yields more accurate user profile data conditioned on time, although factors such as memory requirements and profile build time should be considered. Figure 5.18: below shows the mean number of cells paged per user over a period of 50 days, for time period lengths of 5, 15, 30,45, 60, 90, 120, 240, and 1440 minutes, and a call arrival rate of 6 calls per day. Chapter 5 Performance Analysis Results 66 910 900 5 ID Z 890 v Q . a> oou a ro <n 870 a> o •5 860 a f 850 3 Z 840 -I 830 867.818 868.796 862.649 854.94 Time period lengths (in minutes) Figure 5.18:Number of cells paged per user over a period of 50 days using the intelligent algorithm with various time period lengths and Xcall = 6 calls/day (Note that the lowest value of the y-axis is 830) From Figure 5.18:, the intelligent algorithm produces more paging traffic with increased time period length. The time period length of 1440 minutes (i.e. 24 hours) which is equivalent to disregarding the idea of having time periods, produced the highest paging traffic. There is a 5% improvement in paging traffic for defining 5 minute time periods over not defining any time periods at all. Figure 5.19: shows the mean normalized delay produced by incorporating the different time period lengths. The performance in terms of normalized delay was better with shorter time periods as well, due to a similar reason of having more accurate user profiles available at times of paging. Chapter 5 Performance Analysis Results 67 Time period lengths (in minutes) Figure 5.19:Mean normalized delay using the intelligent algorithm with various time period lengths and Xcall = 6 calls/day 5.3.3 Effect of different dynamic location area sizes As mentioned in Chapter 4, the dynamic location update algorithm generates location areas dynamically based on the user profile. The maximum allowed size Lmax of dynamic location areas is one of the independent variables defined by the network, and this variable was set to 20 cells per location area for the simulation in Section 5.2. Intuitively, larger location areas produce less location updating traffic at the cost of higher paging traffic. Figure 5.20: shows the location updating traffic for Lmax values of 10, 15, 20, 25, and 30. Chapter 5 Performance Analysis Results 68 0 ) in 1200 -3 V-fl> O . 1000 -(A 0> « TJ 800 -a. 3 C O 600 -« u o 400 -o d) qui 200 -3 Z 0 -- » - G S M 1 - • - G S M 2 -A—GSM 3 -x— Intelligent/Dynamic Figure 5.20:Mean number of location updates per user versus maximum location area size Lmax, over a period of 50 days As the maximum dynamic location area size L m a x increases, the number of location updates decreases, but at a decreasing rate. With L m a x =10, the dynamic location update algorithm produces slightly more location updates than GSM3 (20 cells per location area), although it is better than GSM2 (10 cells per location area). The dynamic location update algorithm outperforms the GSM algorithms for Lmax > 15. Chapter 5 Performance Analysis Results 69 -•—GSM 1 « - G S M 2 -A— GSM 3 Dynamic Intelligent 5 10 15 20 25 30 Figure 5.21:Mean number of cells paged per user versus maximum location area size Lmax, for Xcall = 6 calls/day, over a period of 50 days Figure 5.21: illustrates the mean paging traffic of the different algorithms for various sizes of Lmax. As expected, the number of cells paged by the intelligent algorithm and the dynamic algorithm increases with Lmax. According to the figure, the intelligent paging algorithm is more efficient than the dynamic algorithm for different sizes of Lmax. Doubling Lmax from 10 cells to 20 cells causes a 39% increase in paging messages for the intelligent algorithm, while the dynamic algorithm causes a 46% increase. 5.3.4 Effect of different circular region sizes The constant parameter ae plays a role of scaling the circular region which defines the paging zone Zj. A smaller ae results in fewer cells in Z 1 ; but more in Z2, thereby reducing the paging cost if the user is in Zj . However, having a smaller paging zone Zj yields a higher probability of the user being in paging zone Z 2 . In this case, all cells in the dynamic location area need to be paged. Chapter 5 Performance Analysis Results 70 Increasing ae decreases the probability of paging all cells in the location area, but increases paging cost due to the increased size of Zj. Therefore it is important to evaluate the performance using different values of ae, in order to find a balance between the two costs. 930 o. a> 830 -O 810-JO E = 790-770 -750 -I , , , , , , , , , 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 a e Figure 5.22:Mean number of cells paged per user versus ae, for Xcall = 6 calls/day and Lmax = 20, over a period of 50 days Figure 5.22: shows the paging traffic of the intelligent paging algorithm for values of ae ranging from 0.8 to 1.6 with Lmax = 20 and X, C f l / / = 6. The intelligent algorithm performs better with increasing ae until it reaches a value of 1.4, at which point the paging traffic starts increasing slightly with ae. The paging traffic changes only slightly for 1.2 < ae < 1.6. This relates to the probability of finding the user in paging zone Zj , as discussed in Section 4.3.1. Referring back to Figure 4.5, the probability increases at a certain rate until ae = 1.2, after which the rate of increase becomes lower as the probability approaches 1. Figure 5.22: shows that the paging traffic Chapter 5 Performance Analysis Results 71 decreases as the probability of finding the user in Zj increases. But starting from ae = 1.2, the probability increase of Zj is small. Further increasing ae enlarges the size of Zj , which causes the paging traffic to increase after ae = 1.4. 5.3.5 Effect of different call arrival rates The call arrival rate Xcall determines the mean inter-arrival time between pages. It was previously mentioned that for the intelligent paging algorithm, shorter inter-arrival times between pages and location updates result in lower values of te, and thus yield smaller sizes of Zj . Since the size of Zj directly affects the number of cells paged, the paging traffic for different call arrival rates is examined. 20 i (Q O 18 -• 3 Calls/day • 6 Calls/day • 9 Calls/day • 12 Calls/day o 10 i o c c 4 -(0 0) £ 2 -0 G S M 1 G S M 2 G S M 3 Dynamic Intelligent Figure 5.23:Mean number of cells paged per call for various call arrival rates Figure 5.23: shows the mean number of cells paged per call for the different algorithms, Chapter 5 Performance Analysis Results 72 and for call arrival rates Xcall of 3, 6, 9, and 12 calls per day. The mean number of cells paged per call for the G S M algorithms are obviously not affected by Xcall, as the number of paged cells per call is equal to the size of the fixed location area. The dynamic algorithm is also not affected, since the technique of dividing the dynamic location area into two paging zones does not account for elapsed time since the last location update or page. The intelligent algorithm, however, performs slightly better at higher call arrival rates. According to Figure 5.23:, the mean number of cells paged reduces at a decreasing rate as the call arrival rate increases. This is because at a higher Xcall, the mean size of Zx is smaller due to a smaller mean te. Note that when the mean size of Zj is reduced by using a lower ae value, the performance in paging traffic is not necessarily better. This is due to the decrease in the probability of finding the user in Zx, and thus yielding a higher probability of proceeding on to the final paging step of paging the cells in Z 2 . The mean number of cells paged per call for the dynamic algorithm is about 5.2, while for the intelligent algorithm, it is 3 cells per call for Xcall = 3, but drops 10% to 2.7 cells per call when Xcall is increased to 12. Chapter 6 Conclusions The activity-based mobility model with directional random walk, a model designed to simulate user mobility behaviour patterns, was proposed. Multiple user classes with distinctive behaviour patterns are defined in this model. The movements of a mobile user are modelled as a sequence of activities, with a particular location and duration associated with each activity [30]. The interme-diate path between two activity locations is modelled using a directional random walk. In a directional random walk, the user travels towards a destination in a series of discrete and statisti-cally independent steps, with several mobility parameters associated with each travelling step [31]. The activity-based mobility model with directional random walk is used as a tool to evaluate the performances of different location management algorithms. An intelligent location management algorithm which utilizes user profiles for dynamic creation of location areas and paging zones was proposed. A user profile contains the number of times an individual user has moved from a cell to another, and the total time spent in each cell during each time period of the day. Any information that is stored on the user profile for a certain duration of time is considered outdated and is purged. The intelligent algorithm employs an existing location update algorithm, which dynamically forms a location area around the user's current cell based on the cell transition information of the user profile [13]. The paging algorithm utilizes the total time spent in each cell during the current time period for determining the cell probabilities of the cells within the location area. The paging procedure is composed of a maximum of four paging steps, and the most probable cells are paged in the earlier steps. The performance of the intelligent algorithm was compared with the dynamic location management algorithm, and the GSM location management algorithm with fixed location area 73 Chapter 6 Conclusions 74 sizes of 1, 10, and 20. The performances were evaluated in terms of location updating and paging traffic, and normalized paging delay. The results show that the GSM algorithm is optimal in terms of paging delay; however, in terms of radio signalling costs (a linear combination of the location updating traffic and the paging traffic) it performs significantly worse than the dynamic and the intelligent algorithms for all three fixed location area sizes. Although it shares a common location updating algorithm with the dynamic algorithm, the intelligent algorithm is more efficient in terms of paging. As an example, there is a 45% improvement in paging traffic over the dynamic algorithm with X C f l / ; = 6, ae = 1.4, and using 30-minute time periods. The effects of variations in mean velocity v, time period length, maximum dynamic location area size Lmax, circular region parameter ae, and call arrival rate Xcall on the perfor-mance of the intelligent algorithm were studied. The location updating traffic increases slightly with mean velocity, while the paging traffic decreases slightly. The intelligent paging algorithm performs better in terms of paging traffic and delay with shorter time period lengths; however, the trade-off is higher memory requirements and profile build time. Increasing the maximum dynamic location area size Lmax decreases the location updating traffic but at a decreasing rate. The paging traffic increases, albeit not linearly with Lmax. For example, doubling Lmax from 10 cells to 20 cells only results in a 39% increase in paging traffic. The circular region parameter ae set at 1.4 yields the best performance in terms of paging traffic among the ae values from 0.8 to 1.6. The performance increases with ae until ae = 1.4, at which point the performance starts to degrade slightly with ae. The call arrival rate Xcall also affects the performance in paging for the intelligent algorithm. The performance increases with Xcall, but at a decreasing rate approaching Chapter 6 Conclusions 75 a steady state limit. Increasing Xcall from 3 to 12 calls per day yields a 10% reduction in paging traffic. 6.1 Future work The following are some suggestions of topics which can be further investigated: 1. A realistic call arrival/duration model - Realistic arrivals and durations of calls for dif-ferent user classes may yield somewhat different performances of location management algorithms. 2. Added-intelligence based on call arrival patterns - The maximum dynamic location area size L m a x can be set dynamically based on the user's call arrival pattern. Since the cost of a location update maybe considerably higher than the cost of a page, the value of Lmax c a n D e s e t higher during times of low call arrivals. This will decrease the location updating traffic and increase the paging traffic. However, the increase in paging traffic will be less significant due to the low call arrival rate, and therefore the total cost may be lower. Conversely, at times of high call arrivals, the value of L m a x can be set lower to reduce the paging traffic for a cost of higher location updating traffic. This may again yield a lower total cost since the paging traffic is more significant at higher call rates. The call arrival pattern can be captured by storing the call arrival history as a function of time on the user profile, similar to the total time spent in each cell T(b, k). 3. Performance comparison of the proposed intelligent location management algorithm using different mobility models - The intelligent algorithm is expected not to perform as well when using mobility models which do not capture a user's daily movement pat-er6 Conclusions 76 terns, since the resulting user profile will not be useful to the algorithm. 4. Performance comparison of other recently proposed location management algorithms, e.g. [18], using the activity-based mobility model with directional random walk. Glossary BSC Base Station Controller BSS Base Station Subsystem BST Base Station Transceiver F C C Forward Control Channel GMSC Gateway Mobile Switching Center G S M Global System for Mobile Communications HLR Home Location Register IAM Initial Address Message IMSI International Mobile Subscriber Identity LAI Location Area Identifier M A P Mobile Application Part MIN Mobile Identification Number MS Mobile Station MSC Mobile Switching Center MSISDN Mobile Station ISDN Number MSRN Mobile Station Roaming Number NSS Network Switching Subsystem PSTN Public Switched Telephone Network RCC Reverse Control Channel SS7 Signalling System No. 7 TMSI Temporary Mobile Subscriber Identity V L R Visitor Location Register 78 ap Constant parameter for the random angular deviation distribution ae Scaling factor of the circular region (3 Random angular deviation c Relative cost constant of a location update to a page Cprev The user's previously known cell location C Total Total location management cost Xcall Call arrival rate Xh Constant parameter for the random hold time distribution cd Standard deviation for the random velocity distribution Ld List of cells making up the dynamic location area L m a x Maximum allowed size of a dynamic location area N Mean weight of the links to the neighbouring cells N(a, b) Number of transitions the user has made from cell a to cell b NCa List of neighbouring cells of cell a Pxi Probability of locating the user in paging zone Zu Pz Paging zone containing the most probable cells re Mean radial distance for elapsed time te tb Total time the user has spent in cell b for current visit te Elapsed time since the last location update or page th Random hold time T(b) Average time the user has spent in cell b T(b, k) Total time the user has spent in cell b during time period k Location updating traffic Paging traffic Normalized mean paging traffic for paging Zj Random velocity Paging zone within the circular region Paging zone outside of the circular region Sub-paging zones within paging zone Zj Bibliography [1] W. C. Y. Lee, Mobile Cellular Communication Systems, 2nd Edition. New York: McGraw-Hill, Inc., 1995. [2] I. F. Akyildiz and J. S. M . Ho, "On Location Management for Personal Communications Networks," IEEE Commun. Mag., pp. 138-145, Sept. 1996 [3] T. S. Rappaport, Wireless Communications Principles and Practice, Prentice-Hall, Inc., 1996. [4] N. E. Kruijt, D. Sparreboom, F. C. Schoute, and R. Prasad, "Location management strategies for cellular mobile networks," Electronics & Communication Engineering Journal, pp. 64-72, April 1998. [5] S. Tabbane, "Location Management Methods for Third-Generation Mobile Systems," IEEE Commun. Mag., pp. 72-84, Aug. 1997. [6] A. Bar-Noy, I. Kessler, and M . Sidi, "Mobile Users: To Update or not to Update?," IEEE INFOCOM, vol. 2, pp. 570-576, 1994. [7] G. Colombo and H. Hegeman, "Network architecture and functionalities in UMTS, " Proc. IEEE/ICCC PIMRC '94, Wireless Networks, pp. 844-851, Sept. 1994. [8] S. Okasaka, S. Onoe, S. Yasuda, and A. Maebara, "A new location updating method for digital cellular systems," Proc. IEEE VTC '91, Saint Louis, MI, pp. 345-350, May 1991. [9] H. C. Lee and J. Sun, "Multilayered Model Strategy for Optimal Mobile Location Tracking," Proc. IEEE PIMRC '97, vol. 3, Helsinki, Finland, pp. 1009-1013, Sept. 1997. [10] S. J. Kim and C. Y. Lee, "Modeling and Analysis of the Dynamic Location Registration and Paging in Microcellular Systems," IEEE Trans. Veh. Technoi, vol. 45, no. 1, pp. 82-90, Feb. 1996. [11] R. Chang and K. Chen, "Dynamic Mobility Tracking for Wireless Personal Communication Networks," Proc. IEEE ICUPC '97, vol. 2, San Diego, CA, pp. 448-452, Oct. 1997. 80 Bibliography 81 [12] L . -R. Hu and S. S. Rappaport, "Adaptive location management scheme for global personal communications," IEEProc. Commun., vol. 144, no. 1, pp. 54-60, Feb. 1997. [13] J. Scourias and T. Kunz, "A Dynamic Individualized Location Management Algorithm," Proc. IEEE PIMRC '97, vol. 3, Helsinki, Finland, pp. 1004-1008, Sept. 1997. [14] B. Suh, J. Choi, and J. Kim, "Mobile Location Management Strategy with Implicit Location Registration," IEEE 49th VTC '99, vol. 3, Houston, TX, pp. 2129-2133, May 1999. [15] S. Tabbane, "An Alternative Strategy for Location Tracking," IEEE lournal on Selected Areas in Communications, vol. 13, no. 5, pp. 880-892, June 1995. [16] D. Gu and S. S. Rappaport, "A Dynamic Location Tracking Strategy for Mobile Communication Systems," IEEE 48th VTC '98, vol. 1, Ottawa, ON, pp. 259-263, May 1998. [17] J. S. M . Ho and J. Xu, "History-Based Location Tracking for Personal Communications Networks," IEEE 48th VTC '98, vol. 1, Ottawa, ON, pp. 244-248, May 1998 [18] H. C. Lee and J. Sun, "Mobile Location Tracking by Optimal Paging Zone Partitioning," IEEEICUPC '97, vol. 1, San Diego, CA, pp. 168-172, Oct. 1997. [19] S. Madhavapeddy, K. Basu, and A. Roberts, "Adaptive Paging Algorithms for Cellular Systems," IEEE 45th IEEE VTC '95, vol. 2, Chicago, IL, pp. 976-980, July 1995. [20] F. Sarkar, S. Subramaniyam, and S. Madhavapeddy, "Iterative Algorithm for Uniform Page Traffic Reduction in a Cellular System," IEEE 47th VTC, vol. 2, Phoenix, AZ, pp. 500-504, May 1997. [21] G. L . Lyberopoulos, J. G. Markoulidakis, D. V. Polymeros, D. F. Tsirkas, and E. D. Sykas, "Intelligent Paging Strategies for Third Generation Mobile Telecommunication Systems," IEEE Trans. Veh. Technol., vol. 44, no. 3, pp.543-554, Aug. 1995. [22] B. Lee and C. Hwang, "A Predictive Paging Scheme Based on the Movement Direction of a Mobile Host," IEEE 50th VTC '99-Fall, vol. 4, Amsterdam, Netherlands, pp. 2158-2162, Sept. 1999. [23] S. Mishra and O. K. Tonguz, "Most Recent Interaction Area and Speed-based Intelligent Paging in PCS," IEEE 47th VTC '97, vol. 2, Phoenix, AZ, pp. 505-509, May 1997. Bibliography 82 [24] G. Wan and E . Lin, "A Dynamic Paging Scheme for Wireless Communication Systems," ACM/IEEE MobiCom '97, Budapest, Hungary, pp. 195-203, Sept. 1997. [25] D. G. Jeong and W. S. Jeon, "Effective Location Management Strategy Based on User Mobility Classes," IEEE GLOBECOM '98, vol. 3, Sydney, Australia, pp. 1426-1430, Nov. 1998. [26] T. K. Kim and C. Leung, "Generalized Paging Schemes for Cellular Communication Systems," IEEE PACRIM '99, Victoria, BC, pp. 217-220, Aug. 1999. [27] D. Munoz-Rodriguez, "Cluster Paging for Traveling Subscribers," IEEE 40th VTC '90, pp. 748-753, May 1990. [28] J. Scourias, "Dynamic Location Management and Activity-based Mobility Modelling for Cellular Networks," Master of Mathematics thesis, Dept. of Computer Science, University of Waterloo, 1997. [29] K. C. H. Lam, "Location Updating and Paging Strategies in a Cellular System," M.A.Sc. thesis, Dept. of Electrical Engineering, University of British Columbia, April 1993. [30] J. Scourias and T. Kunz, "An Activity-based Mobility Model and Location Management Simulation Framework," ACM MSWiM '99, Seattle, WA, pp. 61-68, Aug. 1999. [31] D. Lam, J. Jannink, D. C. Cox, and J. Widom, "Modeling Location Management in Personal Communication Services," IEEE 5th ICUPC '96, vol. 2, Cambridge, M A , pp. 596-601, Sept.29-Oct.2, 1996. [32] A. Lombardo, S. Palazzo, and M . Tedesco, "A Comparative Study of Tracking Strategies using Directional Mobility Models," IEEE 8th PIMRC '97, vol. 3, Helsinki, Finland, pp. 994-998, Sept. 1997. [33] E. Jugl and H. Boche, "A New Mobility Model for Performance Evaluation of Future Mobile Communication Systems," IEEE ICCC '99, vol. 3, Vancouver, BC, pp. 1751-1755, June 1999. [34] "American Community Survey," U. S. Census Bureau, http://www.census.gov/acs/www/, 1999. 

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:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0065263/manifest

Comment

Related Items