Citation
Optimizations for Vertical Handoff in Next Generation Wireless Networks

Material Information

Title:
Optimizations for Vertical Handoff in Next Generation Wireless Networks
Creator:
ZHU, FANG ( Author, Primary )
Copyright Date:
2008

Subjects

Subjects / Keywords:
Authentication ( jstor )
Bandwidth ( jstor )
Cosmic microwave background radiation ( jstor )
Cost analysis ( jstor )
Cost functions ( jstor )
Distance functions ( jstor )
News content ( jstor )
Propagation delay ( jstor )
Signals ( jstor )
User preferences ( jstor )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Fang Zhu. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
12/31/2007
Resource Identifier:
496175033 ( OCLC )

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

OPTIMIZATIONS FOR VERTICAL HANDOFF IN NEXT GENERATION WIRELESS NETWORKS By FANG ZHU A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2005

PAGE 2

Copyright 2005 by Fang Zhu

PAGE 3

This document is dedicated to my parents.

PAGE 4

iv ACKNOWLEDGEMENTS First of all, I would like to express my sincere gratitude to my advisor, Prof. Janise McNair, for her guidance, encouragement, a nd contributions in the development of my research. Without her vision, deep insight, advice, and willingness to provide funding, this work would not have been possible. I would also like to than k Prof. Yuguang Fang, Prof. Jianbo Gao, and Shigang Chen for serving on my dissertation comm ittee and providing valuable advice on my research. I extend my special thank to Prof. Dapeng Wu for valuable discussions. I would like to extend my sincere appreci ation to my fellow graduate students (Yuan Guo, Kartikeya Tripathi, Pankaj Agga rwal, Hetal Patel, Dawood Al-Abri, Aarti Bharathan, Sitarama Penumetsa), in the labora tories for their assistance, and graduate students in the Wireless Netw orks Laboratory (WINET) for their valuable inputs and support. Finally, this dissertation is dedicated to my parents for their love, sacrifice and support.

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGEMENTS...............................................................................................iv LIST OF TABLES............................................................................................................vii LIST OF FIGURES.........................................................................................................viii ABSTRACT....................................................................................................................... .x CHAPTER 1 INTRODUCTION..................................................................................................1 2 BACKGROUND....................................................................................................4 2.1 Architectures and Enabling Technologies..................................................4 2.1.1 Micromobility Protocols.................................................................6 2.1.2 Mobile IP Version 6........................................................................7 2.2 Vertical Handoffs........................................................................................8 2.2.1 Vertical Handoff Decision............................................................10 2.2.2 Vertical Handoff Radio Link Transfer Design.............................13 2.2.2.1 Context transfer.................................................................16 2.2.2.2 Related work on context transfer......................................16 2.3 Related Work............................................................................................19 3 POLICY-BASED VERTICAL HA NDOFF NETWORK ARCHITECTURE.....23 3.1 UMTS/WLAN Integration........................................................................24 3.2 Vertical Handoff Policy Architecture.......................................................26 3.2.1 The Internet Engineering Ta sk Force (IETF) Guidelines.............26 3.2.2 Proposed Vertical Handoff Interworking Scenariors....................28 3.2.2.1 NCHO/MAHO..................................................................29 3.2.2.2 MCHO...............................................................................32 3.3 Performance Analysis...............................................................................34 3.3.1 Policy-Based Handoff Overhead..................................................34 3.3.2 Policy-Based Handoff Decision Latency......................................34 3.3.3 Probability of Policy-Base d Vertical Handoff Failure..................35 3.4 Numerical Results.....................................................................................36

PAGE 6

vi 4 MULTI-SERVICE VERTICAL HANDOFF DECISION ALGORITHM (MUSE-VDA).......................................................................................................40 4.1 MUSE-VDA Algorithm and Cost Function.............................................44 4.2 Performance Analysis...............................................................................46 4.2.1 Blocking Probability.....................................................................48 4.2.2 Mobility Model.............................................................................51 4.3 Numerical Results.....................................................................................51 4.3.1 RSS Performance..........................................................................53 4.3.2 RSS with Mobility Performance...................................................55 4.3.3 MUSE-VDA performance............................................................56 4.3.3.1 Scenario 1: Collective Handoff.........................................57 4.3.3.2 Scenario 2: Prioritized Multi-Network Handoff...............57 4.3.4 Results for More Demanding CBR Services................................60 5 LAYER 3 MOVEMENT ESTIMATION (L3ME) USING LAYER 2 HINTS...66 5.1 L3ME Architecture...................................................................................68 5.2 Layer 2 Hints.............................................................................................70 5.2.1 Categories of Layer 2 Hints..........................................................70 5.2.2 L3ME Likelihood Function..........................................................72 5.3 Application in Pre-Re gistration Handoff..................................................75 5.3.1 Handoff Latency Analysis............................................................76 5.3.2 Signaling Cost Analysis................................................................77 5.3.3 Signaling Efficiency Analysis.......................................................78 5.4 Performance Analysis...............................................................................79 6 SUMMARY AND FUTURE WORK..................................................................83 6.1 Summary...................................................................................................83 6.2 Future Work..............................................................................................84 LIST OF REFERENCES..................................................................................................85 BIOGRAPHICAL SKETCH............................................................................................91

PAGE 7

vii LIST OF TABLES Table page 1 Parameters...............................................................................................................37 2 Region names and descriptions...............................................................................50 3 Parameters used in the numerical results................................................................52 4 Sets of signaling cost parameters............................................................................79

PAGE 8

viii LIST OF FIGURES Figure page 1 Mobile IP and micro-mobility...................................................................................7 2 Traditional handoff policies....................................................................................12 3 Soft handoff policies...............................................................................................13 4 Signal flow for handoff into a boundary cell..........................................................15 5 Context transfer protocol operation.........................................................................17 6 Message sequences for proactive and r eactive context transfer initiation..............19 7 UMTS and WLAN integration (ETSI)....................................................................25 8 Two possible policy-based network architectures..................................................27 9 NCHO or MAHO handoff decision procedure.......................................................29 10 Signaling during vertical handoff............................................................................30 11 HO trigger message format.....................................................................................31 12 MCHO handoff decision procedure........................................................................32 13 Signaling during vertical handoff............................................................................33 14 Handoff failure comparison ( HO=50ms, 100ms)...................................................37 15 Handoff failure comparison....................................................................................38 16 Diverse 3G and 4G wireless networks....................................................................40 17 Scenario 2—Prioritized user sessions.....................................................................47 18 Cell overlay network...............................................................................................48 19 APUSR provided by RSS-based algorithm.............................................................53 20 Blocking probability of each network in RSS based algorithm..............................54

PAGE 9

ix 21 APUSR comparison................................................................................................55 22 Blocking probability comparison............................................................................56 24 Blocking probability of each network in MUSE-VDA...........................................59 25 Comparison of the RSS-based, collec tive, and prioritized MUSE-VDA algorithms................................................................................................................62 26 Comparison of throughput with changing of CBR request.....................................63 27 Comparison of the RSS-based, collec tive, and prioritized MUSE-VDA algorithms................................................................................................................64 28 3G cell overlay network..........................................................................................69 29 L3 movement estimation.........................................................................................74 30 Mobile IP handoff message timing diagram...........................................................75 31 Handoff pre-registration with layer 3 movement detection....................................76 32 Signaling cost comparison (Set I)...........................................................................80 33 Signaling cost comparison (Set II)..........................................................................81

PAGE 10

x Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy OPTIMIZATIONS FOR VERTICAL HANDOFF IN NEXT GENERATION WIRELESS NETWORKS By Fang Zhu December 2005 Chair: Janise McNair Major Department: Electrical and Computer Engineering Next generation wireless networks will be based on the coordination of different types of networks in a heterogeneous environment. A significant challenge for coordination is vertical handoff that ensures mobile usersÂ’ seamless mobility and ubiquitous access to applications across heterogeneous networks. In this dissertation, we propose a policybased framework that translates handoff policies into a network configuration to produce a satisfact ory result for both the user and the network. A case study of UMTS/WLAN interworking is provided using the policybased vertical handoff architecture, and a pe rformance analysis demonstrates that the additional handoff signaling latency is low e nough to support users of various mobility with low handoff failure rate. The dissertation next focuses on the deci sion for a mobile node to handoff between different types of networks. While traditiona l handoff is based on received signal strength comparisons, vertical handoff must evaluate a dditional factors, such as monetary cost,

PAGE 11

xi offered services, network conditions, and user preferences. In this dissertation, several optimizations are proposed for the execution of vertical handoff decision algorithms, with the goal of maximizing the quality of service experienced by each user. First, the concept of policy-based handoffs is discussed. The n, a multi-service vertical handoff decision algorithm (MUSE-VDA) and cost function are introduced to j udge target networks based on a variety of userand ne twork-valued metrics. Fina lly, a performance analysis demonstrates that significant gains in the ability to satisfy user requests for multiple simultaneous services and a more efficient us e of resources can be achieved from the MUSE-VDA optimizations. This dissertation then turns its attent ion to handoff latency and signaling cost. Wireless and mobile Internet access has brought increased significance to the evolution of 3G packet-based communications and a visi on of a future based on all-IP networking. To reduce handoff latency and signaling cost are two major issues . Certain link-layer technologies are capable of provi ding various link stat us information to the IP module. It has been identified that receiving explicit hints from the layer 2 (link layer) would expedite the layer 3 (network layer) moveme nt detection process, which in turn will reduce the handoff latency. In this dissert ation, a likelihood func tion based method for implementing link layer hints is defined. Then the function is used to estimate layer 3 movement in two levels: (1) a mobile node is moving into a new subnet and (2) a mobile node is requiring an inter-system handoff. Next, the use of the likelihood function is applied to a pre-registration inter-system handoff protocol. Finally, performance analysis demonstrates significant gains in signaling cost from the proposed method.

PAGE 12

1 CHAPTER 1 INTRODUCTION At the turn of the 21st century, the wide spread success of wireless and mobile communications has resulted in the creation of a large variety of wi reless technologies, including secondand thirdgeneration (2G and 3G) cellula r, satellite, Wi-Fi, and Bluetooth. Each technology is tailo red to reach a particular mark et, or a particular type of user with a specific service need. The advantag e to these diverse networks is that they offer many choices for increasing bandwidt h, accessing the Internet, and increasing the coverage area for the average user. Howe ver, expanding services through the use and coordination of diverse networks creates th e challenge of developing novel interoperable network protocols to manage user mobility between different types of systems — a level of interoperability curre ntly not available in 3G wireless systems [1]. The fourth generation (4G) of wirele ss communications refers to the next evolutionary step after standard ization of the 3G infrastructu re and the next revolutionary step for wireless telecommunications in genera l [2]. The evolutionary goals of 4G beyond 3G include building on packet-based code -division multiple access (CDMA) networks under such systems as the Universal Mobile Telecommunications System (UMTS). The revolutionary goals are visionary in nature , requiring an evaluati on of the technological, societal, and market developments over the ne xt 10 years. Some goals may be forecast by emerging issues, such as spectrum efficien cy, dynamic bandwidth allocation, security, quality of service (QoS), a nd transceiver technology, while other goals may arise from factors that dictate entire ly new approaches and novel infrastructure solutions.

PAGE 13

2 Revolutionary drivers for 4G include a pus h toward universal wireless access and ubiquitous computing through seamless personal and terminal mobility [1, 3]. Universal wireless access refers to the ability of a us er to connect anywhere at any time from any network. The change in connection may be initi ated by the user or ma y be initiated by the network, transparent to the user. For exampl e, a user may choose to access a wireless LAN (WLAN) to send a large data file, but ma y choose the cellular network to carry on a voice call. On the other hand, a network may decide to hand off a stationary data user to a WLAN in order to increase bandwidth availa bility for mobile users in a 3G cellular network. Ubiquitous computing refers to the ability to move seamle ssly within a network while receiving intelligent, context-aware services. Personal mobility allows a user to receive services at any terminal device, wh ile terminal mobility allows the device to receive services even as it moves between network access points. To achieve seamless mobility, network management operations must be conducted without causing degradation of services, and wit hout need for user intervention. The movement of a user within or am ong different types of networks can be referred to as intersystem or vertical mobility. One of the major challenges for seamless vertical mobility is vertical handoff, wh ere handoff (or handover) is the process of maintaining a mobile userÂ’s active connectio ns as it changes its point of attachment. Traditionally, handoff research has been ba sed on an evaluation of the signal strength received at the mobile node, followed by a ch ange in access point, if needed, and an updated routing path for the user connection. However, with a vision of a diverse multinetwork environment, and considering the goals of transparent universal access, ubiquitous computing, and seamless mobility, tr aditional signal strength comparisons are

PAGE 14

3 not sufficient to make a handoff decision, as they do not take into account the current context or the various attachment options for the mobile user. Another issue in vertical handoff is the timely and reliable transfer of a mobile userÂ’s connection(s). While traditional link transfer techniques can achieve fast handoffs, there is now a need to consider the context of the link transfer, incl uding security associations, QoS guarantees, and any special processing operations. Thus, th e vision of 4G requires investigation of a more adaptive and intelligent netw ork approach to vertical handoff. The rest of the dissertation is organi zed as follows. Chapter 2 introduces the background and related research on vertical handoffs. In Chapter 3, the concept of policybased vertical handoff is introduced, followe d by an optimization for vertical handoff decision algorithms in 3G overlay multi-netwo rk environment in Ch apter 4. In Chapter 5, a likelihood function based method for implemen ting link layer hints is defined, and is used to estimate layer 3 movement for Mobile IP handoffs. Current work is summarized in Chapter 6, along with the plan of future work.

PAGE 15

4 CHAPTER 2 BACKGROUND In this chapter, we first describe vari ous network architectures and technologies currently evolving beyond 3G, including WLAN s, cellular, and Mobile IP. We then explore the problem of vertical handoff design in the context of the envisioned environment. Finally, we describe the open research problems for achieving an interoperable and transparent handoff decisi on and detection algorithm, and a contextaware radio link transfer, respectively. 2.1 Architectures and Enabling Technologies The evolutionary architecture beyond 3G builds on a hierarchic al cellular system for wireless wide area services, and a mobile satellite network to provide GPS location services, high bandwidth pipes, and the abil ity to reach customers in rural areas [1]. However, as mentioned previously, the wi despread success of wireless communications has resulted in the addition of an even great er variety of wireless networks that must coexist. Some of the various types of networks include the following: Wireless personal area networks (WPANs ) and enabling technologies, such as Bluetooth, that provide range-limited ad hoc wireless service to users for access to a variety of personalized items WLANs, such as 802.11, that provide Ethern et access to wireless users without the costly infrastructure of 3G Wireless wide area networks (WWANs), such as UMTS, that provide global cellular service to mobile users High aeronautical altitude platforms (HAA Ps), such as unmanned air vehicles (UAVs), that use aircraft to provide fl exible wireless access without the costly infrastructure of a satellite network

PAGE 16

5 The evolution of 3G packet-based communication has brought increased significance to wireless and mobile Internet access, and a vision of a future based on allIP networking [3]. In support of the all-IP vision, the Third Ge neration Partnership Projects (3GPP and 3GPP2), which repres ent the standards of the global wireless industry, have begun to develop all-IP ve rsions of their respective 3G wireless architectures. In Pahlavan et al. [4], a study is performed to compare five different prospective architectures for implementing a vertical hando ff between networks based on the WLAN standard, IEEE 802.11, and networks ba sed on the 3G cellular data standard, General Packet Radio Service (GPRS). The two architectures found to be the most efficient, without requiring a master/slave relationship between the different networks, were the mobility gateway/proxy-based ar chitecture, which c onsists of a proxy implementation between a GPRS network a nd a WLAN, and an architecture based on Mobile Internet Prot ocol (Mobile IP). In the mid-1990s, Mobile IP was standardiz ed by the Internet Engineering Task Force (IETF) to allow mobile nodes to change their point of attachment to the Internet while still being able to maintain a connec tion to the network [5]. Under Mobile IP, a mobile node that is currently residing in its home subnetwork is served by a home agent that forwards all incoming packets to the mob ile node at its home IP address. When the mobile node moves away from its home s ubnetwork to a new location, the node must contact a foreign agent at the new subnetwork to obtain a new IP address, called a care-of address. A binding update must then be pe rformed to notify the home agent about the mobile nodeÂ’s new care-of address. The home agent then forwards all incoming packets to the mobile node using a process referred to as tunneling: the home agent encapsulates

PAGE 17

6 the incoming packets for the mobile node and fo rwards them to the foreign agent, which in turn decapsulates them and delivers them to the mobile node. Meanwhile, the mobile node can continue to transmit packets directly to the correspondent node. For Mobile IP, the binding updates and care of address exchanges that establish the mobile node at each new location cause an increas ed signaling load as well as delays that may be detrimental to the serv ice being received at the mobi le node. In some cases, these delays may not be necessary, since other t echniques can be used to resolve packet forwarding to the roaming mobile node. One such method is referred to as micromobility. 2.1.1 Micromobility Protocols Micromobility protocols reduce the need to change the foreign agent of the mobile node for intradomain mobility. As illustrated in Figure 1, a hierarchy of base stations is organized under one foreign agent, so when th e mobile node changes base stations within the same domain, location information is propagated only through the domainÂ’s local routers, transparent to the home agent. Inco ming packets arriving at the domain can be efficiently and quickly forwarded to the mob ile nodeÂ’s current loca tion without the need for another binding update or car e-of address. Note, however, that when a mobile node moves into a new domain, a traditional Mobile IP handoff is necessary. Micromobility protocols discussed in the research literature include Cellular IP[6] and HAWAII [7]. In Cellular IP, routing d ecisions in the local domain are conducted using a local gateway and base station that c ache the forwarding path of each packet that arrives from or goes to the mobile node, while HAWAII uses a crossover router to manage handoff between two base st ations within the same domain.

PAGE 18

7 Figure 1. Mobile IP and micro-mobility. 2.1.2 Mobile IP Version 6 Other improvements have been proposed a nd adopted for Mobile IP under the title of Mobile IP version 6 [5]. For example, Mobile IPv6 eliminates triangular routing and enables the correspondent node to reroute packets on a direct path to the mobile node. This process is referred to as route optimiza tion, which is not always available in Mobile IPv4. In addition, Mobile IPv4 also suffers from a lack of security constructs for authorization, authentication, and accounting, as well as for source routing. Mobile IPv6 includes embedded binding update s and care-of address conf iguration for the execution

PAGE 19

8 of location updates and processing the change in the mobile nodeÂ’s address. The newer version also includes authentication header processing to provide validation of mobile nodes. Finally, IPv6 has a fourfold increase in IP address space, which may be useful for developing new mobile node addressing schemes. Regardless of the type of network, issues such as addressing, route optimization, and authentication are part of the challengi ng overall problem of creating an efficient handoff mechanism that satisfies the seamle ss mobility needs of the user population while enabling advanced processing and op timization operations at the network. The resolution of these problems in a multi-network environment is the goal of vertical handoff research, described next. 2.2 Vertical Handoffs Any handoff operation is a threestage process that includes handoff decision , radio link transfer , and channel assignment [1]. Traditionally, handoff decision is performed based on a perception of channel quality re flected by the received signal strength and other measurements, and the availability of resources in the new cell. The base station usually measures the quality of the radio li nk channels being used by mobile nodes in its service area. This is done peri odically so that degradations in signal strength below a prescribed threshold can be de tected and handoff to another ra dio channel or cell can be initiated. Under network-controlled ha ndoff (NCHO) or mob ile-assisted handoff (MAHO), the network makes the decision for handoff, while under mobile-controlled handoff (MCHO), the mobile node takes its own signal strength measurements and makes the handoff decision on its own. While performing handoff, the mobile nodeÂ’s connection may be created at the target base station before the old base station connection is released. This is referred to as a make before break handoff. On the other hand, the new

PAGE 20

9 connection may be set up after the old connect ion has been torn down, which is referred to as a break before make handoff. In either case, the mobile node executes a hard handoff , which means that the mobile node can only communicate on a channel with one base station at time. In 3G CDMA networks , a mobile node is able to communicate on more than one coded channel, which enables it to communicate with more than one base station. Thus, CDMA networks allow a soft handoff , where the mobile node can listen to a set of candidate base stati ons at the same time before choosing one for its point of attachment. The second part of the three-stage handoff is radio link transfer. Radio link transfer refers to the responsibility of the network to form new links to the call at its new point of attachment. Handoffs to another radio channe l within the same cell, referred to as intracell handoff , require no new link transfer operat ions. However, handoff to another cell as a result of mobile node movement to a new base station is referred to as intercell handoff and requires handoff rerouting opera tions to link the mobileÂ’s existing communication path to the new cell. The third handoff stage, channel assignment, consists of the allo cation of resources to the handoff call at the new point of atta chment. Channel assignment for handoff calls is also part of the problem of resour ce management and call admission control for wireless networks, and is out of the scope of this dissertation. The traditional handoff process described above is insufficient for the challenges of the 4G system for the following reasons: Criteria. The use of the signal strength crit erion for traditional handoffs limits the ability of the network to initiate a han doff for control reasons (congestion relief, change in data traffic, etc.).

PAGE 21

10 User selection. Traditional handoff does not allow user selection of networks, and assumes that there is only one choice for access technology. In a heterogeneous environment, user choice is a desirable amenity. Context. Whereas traditional handoff link tr ansfer concerns the delivery of packets to the new point of attachment, there is now a need for the delivery of the context of the information flow between the mob ile node and the network as well. Context may include security associations, QoS gua rantees, authentication headers, and so on. Interoperability. Traditional handoff prot ocols are developed for homogeneous systems that rely on a common signaling pr otocol, routing technique, and mobility management standard. In heterogeneous environments, mobile nodes and network routers must be able to interoperate with different networks, and with the corresponding protocols and standards. In the next two sections, we define th e open problems for implementing vertical handoffs. 2.2.1 Vertical Handoff Decision The decision to perform a handoff is ba sed on the evaluation of handoff metrics, which are the measured qualities that give an indication of whether or not a handoff is needed. As stated previously, in traditiona l handoffs only received signal strength and channel availability are considered. In th e envisioned 4G system, the following new metrics have been proposed: Service type . Different types of services re quire various combinations of reliability, latency, and data rate. Monetary cost . Cost is always a major consid eration to users, as different networks may employ different billing strategi es that may affect the userÂ’s choice of handoff. Network conditions . Network-related parameters such as traffic, available bandwidth, network latency, and congestion (packet loss) may n eed to be considered for effective network usage. Use of network inform ation in the choice to hand off can also be

PAGE 22

11 useful for load balancing across different networks, possibly relieving congestion in certain systems. System performance . To guarantee the system pe rformance, a variety of parameters can be employed in the handoff decision, such as the channel propagation characteristics, path loss, interchannel interference, signal-to-noise ratio (SNR), and bit error rate (BER). In additi on, battery power may be anothe r crucial factor for certain users. When the battery level is low, the user may choose to switch to a network with lower power requirements, such as an ad hoc Bluetooth network. Mobile node conditions . Mobile node conditions incl ude dynamic factors such as velocity, moving pattern, moving hist ories, and location information. User preferences . User preferences can be used to cater to special requests for one type of system over another. The use of new metrics will increase the complexity of the handoff process, making the handoff decision more and more ambiguous. Thus the development of a cost function to simultaneously evaluate various metrics becomes crucial to th e success of a 4G handoff decision. The cost function must be ba sed on a general policy to be implemented at the network. 2.2.1.1 Handoff policy Whereas handoff decision metrics help to determine where to handoff (i.e., which network should be chosen), the handoff polic y represents the influence of the network on when the handoff occurs. The traditional hando ff policy is illustrated in Figure 2. Each vertical axis represents the signal strength received at the mobile node from each base station, represented as base station (BS) 1 and BS 2. The horizontal axis can represent either the time to hand off or the distance tr aveled from BS 1 to BS 2. The intersection

PAGE 23

12 point between the two curves represents an equal received signal strength from the two BSs, while the points labeled A, B, C, a nd D show the traditional handoff policies for cellular networks. The threshold technique requ ires that the receive d signal strength from BS 1 pass below a pre-determined level befo re handoff can occur. In the figure, the corresponding thresholds and handoff points are (T1, A), (T2, B), and (T3, D). A more robust technique uses both a threshold valu e and a consideration of the difference in signal strength between BS 1 a nd BS 2, known as the hysteresis, h . The hysteresis technique is used to prevent ping-pong ha ndoffs (i.e., unnecessarily repeated handoffs between BSs). A hysteresis handoff point is shown as point C in Figure 2. Figure 2. Traditional handoff policies. A similar handoff policy structure to Figure 2 can also be applied to soft handoffs, wherein a signal strength above a certain thre shold makes a BS a candidate for handoff, and a handoff decision is made according to a ll of the available candidates. An example of soft handoff in IS-95 is shown in Figure 3. The handoff to BS 2 is initialized when the pilot strength measurement from its pilot exceeds a pre-determined threshold value

PAGE 24

13 T_ADD. After a traffic channel is set up with BS 2, the mobile terminal is able to communicate with both BSs for a short period of time (from Point A to B), before the pilot strength of BS1 falls bellow another pr e-determined threshold value T_DROP. Then the connection with BS 1 will be torn down. The 4G multi-network environment require s new handoff policies that reflect the updated 4G criteria but continue to prevent de trimental effects that apply to all handoffs, such as ping-pong effects. Figure 3. Soft handoff policies 2.2.2 Vertical Handoff Radio Link Transfer Design Handoff radio link transfer is the process of rerouting a mobile userÂ’s connection path to the userÂ’s new point of attachment. It requires the networ k to transfer routing information about the mobile user to the new (or target) access router for the proper forwarding of packets. An example of signa l flow for a vertical handoff to a boundary cell base station is illustrated in Figure 4 [11]. When the Mobile Terminal (MT) approaches the intersystem boundary cells of network 1, MT can hear beacons from network 1 boundary cell base station (BBS1 in Figure 4). Once the handoff is initiated, the MT sends an intersystem handoff warning (ISHO_warn) message to BBS1, and T_ADD T_DROP BS1 Pilot BS2 Pilot BS 1 BS 2 A B

PAGE 25

14 BBS1 sends an acknowledgement of the warning message to MT. Next, the ISHO_warn is forwarded from BBS1 to a switch in ne twork 1 (SW1 in Figure 4). SW1 replies to BBS1 with an acknowledgement, and sends an intersystem handoff reroute ( ISHO_reroute ) message to SW2, the switch for the adjacent network 2 boundary cells, to trigger operations such as authentication and location management. In the meantime, the MT continues to handoff to BBS1 using the standard network 1 procedures. Once the handoff is complete, BBS1 sends a transceiver start ( TXRCVR_start ) message to activate a second transceiver pair at the MT using th e network 2 radio characteristics. After the corresponding format transformation, the MT is ready to handoff to network 2 when necessary. When the vertical handoff is initia ted, MT sends a vertical handoff request to BBS1. BBS1 then sends RECINFIG message to trigger the MT to switch to network 2 operations. As described previously, 4G networks are assumed to operate in an environment of multiple standards and networks. Thus, it is expected that 4G handoff link transfer will require additional context (i.e., userand ne twork-specific information) to enable the mobile node to move through different networ ks, while maintaining multiple data flows, and to choose among a variety of options for billing and service permissions, depending on the characteristics of the transmitted data and the current network. Research problems for the transfer of a mobile nodeÂ’s context to a target network include the following: Formatting and interoperability . Handoff signaling messages must be specified and formatted so as to be inte rpretable by the target network. Performance . The desired goal of transferring th e context of a mobile node to the new network is to minimize the delay in re-e stablishing the mobile nodeÂ’s traffic flows.

PAGE 26

15 However, if the context transfer delay were so large as to have the same effect of the complete re-establishment, or large enough to increase the overal l handoff call dropping rate, the advantages of contex t transfer have been removed. Figure 4. Signal flow for handoff into a boundary cell Quality of service (re)negotiation. For a mobile node being handed off (or handing off) to a new network, there may be a change in service quality for better or worse, depending on such factors as bandwidth availability, congesti on, and interference. As discussed earlier for handoff decision techniques, there may be a selection process at the network or mob ile node to give certain flow s a priority status, or to SW2 MT BBS1 SW1 ISHO_warn ISHO_warn_AC ISHO_warn ISHO_warn_AC ISHO_reroute ISHO_reroute_ACK MT handoff to BBS1 New route is determined TXRCVR_start Initiation BBS2 Format Transformation ISHO_req RECONFIG_done RECONFIG_exe MT hands off to BBS2 Initiation

PAGE 27

16 appropriately adjust the author ization and billing constraints. A mechanism is needed to allow for internetwork and/or inter-servic e-provider agreements to support fast intersystem roaming that avoids an unr easonable amount of in ternetwork signaling exchanges to validate or institute the adjustment in services. 2.2.2.1 Context transfer A context transfer protocol is designed to allow access routers to exchange state information regarding a mobile nodeÂ’s packet treatment [10]. State information affects packet handling at each access router, and in cludes, as examples, protocols for managing QoS guarantees, header compression, and au thentication, authorization, and accounting (AAA). In the absence of context transfer, there may be large delays because of the network signaling required to re-establish QoS flows, re-authenticate the mobile user at the new router, and set the header compressi on algorithms. Large handoff delays can lead to reduced QoS, disruption of TCP operati on, and dropped handoff calls. Furthermore, a context transfer between access routers reduc es the need for control signaling over the unreliable and bandwidth-limited wireless channel. 2.2.2.2 Related work on context transfer The context transfer protocol being devel oped by the IETF is i llustrated in Figure 5. As the mobile node moves from its previous access router to the new access router, the corresponding information about each of the mobile nodeÂ’s microflows is forwarded between the access routers. Each microflow is categorized into feature contexts, which allow the network to indicate and provide the particular context information needed per microflow. For example, a pa rticular mobile user may be downloading streaming video while conducting a voice transmission. The cont ext required to continue the voice call may be authentication information only, while the context needed for the video service

PAGE 28

17 may be QoS and header compression, in additi on to authentication. Th e initiation of the context transfer can be trigge red by several events, such as a handoff due to a change in mobile node location, a request by the mobile node to change services, or by a network need to relieve congestion at a certain rout er. The mechanisms and others are discussed as part of the handoff decision algorithms earlie r. We discuss here the message sequences required to initiate the context transfer. Figure 5. Context transfer protocol operation. The message sequences are illustrated in Figure 6. They are separated into two categories: proactive and reactive. For the pr oactive case, the context may be transferred to the new access router before the mobile node attaches, so the c ontext is immediately available before or during handoff. In the re active case, the new access router explicitly requests the mobile nodeÂ’s cont ext information, either as pa rt of the handoff signaling or

PAGE 29

18 after the handoff is completed. The Context Transfer Start Request message is sent from the MN to an access router to begin a context transfer. The message data consists of the mobile nodeÂ’s previous care-of address, the previous access router Â’s IP address, an authorization token for the mobile node, and a list of the requested context types. It may also include the mobile nodeÂ’s new IP address and the new access router Â’s IP address, if known. The Context Transfer Request message is sent from the new access router to the previous access router, and provides the IP addresses of the MN and the new access router, the list of feature cont exts to be transferred, and a token authorizing the transfer. The token is required to authenticate the m obile node requesting the transfer. Finally, the Context Transfer Data message is the re sponse from the previous access router, containing the relevant context data. In additi on to the data, it incl udes an authorization token that is computed over the binary context data, and provides the mobile nodeÂ’s previous care-of address and new care-of address (if known). The contents of the messages, as well as the messages themselves, are still a subject of research within the SEAMOBY working gr oup, but are provided here to give some indication of the operations required for cont ext transfer. Other open research problems include the investigation of transport laye r congestion control vs. overload risks from context transfer signaling, and the problem of interdomain signaling, which requires additional efforts to establish security relati onships from the previous to the new access router.

PAGE 30

19 Figure 6. Message sequences for proactive and reactive context transfer initiation. 2.3 Related Work Related work on vertical handoff has been presented in recent research literature. Several papers have addresse d designing a higher layer arch itecture for hybrid networks, however, handoff decision was not discussed. In Wu et al. [41] application-layer-based Session Initiation Protocol (SIP) was consider ed but was found considerable delay that might be unacceptable for real-time multimedia service due to transmission of SIP signaling messages over errone ous and width-limited wirele ss links. [42] proposed a hierarchical mobility management architectur e coupled to use multicast packets as the packet forwarding mechanism to handle the fr equent handoffs. P-ha ndoff protocol [43] complemented classical vertical handoff, redi recting traffic to the best ad-hoc link, such as BlueTooth and 802.11b, on a peer by peer basi s. P-Handoff uses link layer events and doesnÂ’t require an infrastruc ture. However, these papers focused on architecture design and did not address the handoff decision point or the vertical handoff performance issues.

PAGE 31

20 Other work considered optimizations after the vertical handoff decision has been made with respect to delay, TCP timeout, thr oughput, etc. In Bernaschi et al. [44], measurement showed that the use of multip le interfaces with layer 2 triggering can effectively reduce the handoff latency in MIP. TCP performance during vertical handoff in hybrid mobile date networks was analyzed in Huang et al. [46] and Kim et al. [47]. In Huang et al. [46], three solutions were pr oposed to solve TCP timeout problem during soft vertical handoff. In Kim et al. [47], af ter a vertical handoff, the sender was suggested to re-adjust its window size based on the capacity of the new network to improve throughput. The buffer sizes to achieve lossle ss vertical handoffs was discussed in Salamah et al. [48]. However, the vertical handoff decision did not consider multiple networks supporting multiple interfaces. The related papers that explored ve rtical handoff decision mainly focus on traditional issues, such as received signal strength (RSS) between the base station and the mobile node, and data rate, which is a very limit view on QoS. The following research explored vertical handoff based on RSS. In Zhang et al. [45], a connection manager was introduced to intelligently detect the conditions of different types of networks and the availability of multiple networks. Fast Fourier Transform (FFT)-based signal decay detection scheme was used to reduce the ping-p ong effect, and an adaptive threshold configuration approach was proposed to pr olong the time the user stays in WLAN. However, RSS was the only factor that was c onsidered for vertical handoffs. A vertical handoff algorithm was proposed, where RSS, diff erent data rates pr ovided by different systems, and packet loss due to handoff de lay were taken into account, and the handoff was controlled by dwell timer [49, 50]. Howeve r, only single service was considered. A

PAGE 32

21 vertical handoff system, called OmniCon, was proposed for seamless handoff between GPRS and WLAN, where handoffs were tri ggered based on computed background noise and signal strength [51]. Handoff decision algor ithm, WISE was proposed in Nam et al. [52] to maximize energy-efficiency without sa crifice of overall netw ork degradation. In Jung et al. [53], the signaling of QoS-ba sed handoff method be tween UMTS and WLAN was proposed, where the handoff policy pref ers WLAN than UMTS. However, the definition of QoS was not defined in the pape r. Other papers focused on mobility level and user position in the netw ork. Mobility level was proposed as a proper metric for multi-tier handoffs in Holtzman et al. [31]. Multi-network architectural issues were explored, and an advanced neural-network-based vert ical handoff algorithm was developed to satisfy user ba ndwidth requirements [4]. A ve rtical handoff algorithm based on pattern recognition was presente d in Mehbodniya et. al [54]. Several paper shave created utility functi ons to better evaluate the choice for vertical handoff. A vertical handoff d ecision function was proposed, which was a measurement of network quality based on ha ndoff metrics in Hassw a et al. [55]. The network with highest network quality va lue was the handoff target. However, no performance analysis was provided. An ac tive application oriented handoff decision algorithm was proposed for multi-interface m obile terminals to reduce the power consumption caused by unnecessary handoffs an d other unnecessary interface activation in Chen et al. [56]. A handoff target networ k was decided by QoS level provided to active application by a network. Howeve r, the situation where multip le active applications exist was not considered. In et al. [57], two ad aptive handoff decision methods were proposed to adaptively adjust the stability period based on utility functi on to avoid unnecessary

PAGE 33

22 handoffs and reduce decision time. A polic y-enabled handoff decision algorithm was proposed along with a cost function that cons iders several handoff me trics [9]. However, multi-service handoff was not fully discusse d. A cost function based vertical handoff decision algorithm was proposed for multi-service handoff. Preliminary results demonstrated significant gain in throughput in Zhu et al. [10].

PAGE 34

23 CHAPTER 3 POLICY-BASED VERTICAL HANDOFF NETWORK ARCHITECTURE The coordination of the many different type s of wireless networks has become a significant challenge for third generation (3 G) and fourth genera tion (4G) wireless networks [1, 4, 8]. The interworking between 3G cellular and data services, such as General Packet Radio Service (GPRS), and wi reless local area netw orks (WLANs), such as IEEE 802.11 and HiPerLAN, has been consid ered a suitable evolution toward next generation networks[32, 35]. The deployment of high speed WLANs in hotpot areas can well supplement 3G networks, which provide re latively large coverage but relatively low data rates. To make an adaptive and intelligent vert ical handoff decision, factors such as received signal strength (RSS), monetary cost, network conditions, mobile node conditions, user preferences, etc., must be c onsidered, as well as th e capabilities of the various networks in the vicinity of the us er. Therefore, a framework that translates handoff policies into network configuration is needed to implement vertical handoff protocols that produce a sa tisfactory result for both the user and the network. Related work on the integration of he terogeneous networks and performing seamless vertical handoff has been presented in recent research literature. Several papers have addressed higher layer ar chitecture and integration re quirements for 3G/WLAN [34, 35, 36, 38, 40, 41]. A framework for providi ng policy-based control over admission control decision is introduced by IETF [27]. However, the focus of this work is on RSVP-based admission control, the extension to mobility issues are not considered. In

PAGE 35

24 Yang et al. [37], a service level agreement based framework is presented for a service provider to perform resource manage ment in heterogeneous wireless networks. However, the function of handoff Gatewa y Resource Agent is not elaborated. In this chapter, we propose a policy-base d architecture for vertical handoffs. The framework translates handoff policies into netw ork a configuration in order to produce a satisfactory result for both mobile users and the network. Contri butions include: (1) Various vertical handoff metrics can be obtaine d and evaluated to make an intelligent and adaptive handoff decision. (2) The additional handoff signaling latency is analyzed to determine the impact of verti cal mobility on handoff failure rate. Section I introduces the UMTS and WLAN interworking architecture, and Section II intr oduces a novel concept of policy-based handoffs in network-controlled and mobile controlled scenarios that can be applied to any integrated network. It includes a case study for UMTS and WLAN integration. A performance analysis of the signaling impact is pr ovided in Section III, followed by the numerical results in Section IV. 3.1 UMTS/WLAN Integration UMTS is the successor to Global System for Mobile Communications (GSM). It addresses the growing demand of mobile and Internet applic ations for new capacity in wireless communications. The network incr eases transmission speed to 2Mbps per mobile user and establishes a global roaming standard. On the other hand, WLANs are mainly designed for the local area, supporting much higher data rates to users with low mobility. To take the advantage of both the UMTS networkÂ’s always-on wide connectivity and the WLANÂ’s higher data serv ice, the following approaches have been proposed for the integration of WLAN s and 3GPP networks [32, 33, 34].

PAGE 36

25 The European Telecommunications Standards Institute (ETSI) specifies two generic approaches for interworking: l oose coupling and tight coupling [32]. The interconnection of UMTS and WLAN is shown in Figure 7. Figure 7. UMTS and WLAN in tegration (ETSI) In tight coupling approach, the WLAN is connected to the UMTS core network via the Serving GPRS Support Node (SGSN) in the same manner as any other radio access network (RAN), such as GPRS RAN and UMTS terrestrial R AN (UTRAN). The WLAN Internet GGSN SGSN RNC WLAN Gateway HSS 3GAAA serve r Tight coupling Loose coupling N ode B AP AP AP MT HLR GGSN: Gateway GPRS service node SGSN: Serving GPRS support node RNC: Radio network controller HSS: Home subscriber server HLR: Home Location register

PAGE 37

26 gateway implements all the UMTS protocols required in UMTS terrestrial radio access network (UTRAN). Thus, the WLAN data traffic goes through the UMTS core network before reaching the external data network. As a result, the mechanisms for mobility, QoS provisioning and Authentication /Authoriza tion/Accounting (AAA) in the UMTS core network can be reused di rectly over the WLAN. On the other hand, with loose coupling the WLAN bypasses the UMTS core network, and directly co nnects to Internet, as shown in Figure 7. In this case, UMTS and WLAN use different mechanisms to handle authentication, mobili ty and billing. The WLAN is able to access the subscriber da tabases in the UMTS network for security, billing, etc, but has no data traffi c interface to UMTS core network. 3.2 Vertical Handoff Policy Architecture 3.2.1 The Internet Engineering Task Force (IETF) Guidelines Vertical handoff performed on a policy-ba sed networking archite cture requires the coordination of a wide variety of network devices within a single administrative domain to implement a set of quality of service (QoS) based services [27]. Figure 8 shows two possible conceptual architectures of polic y-based solutions that have been proposed by the IETF. The two main architectural elements for policy control are the policy enforcement point (PEP) and the policy deci sion point (PDP). These two el ements may be located in the same network node (as shown in Figure 8 (a)) or in different nodes (as shown in Figure 8 (b)). The latter is especially convenient to apply local policies. PEP is a component that runs on the policyaware network node a nd is the point at which the policies are enforced. Policy decision s are made primarily at the PDP, based on the policies extracted from the policy databa se. The PDP as specified by the IETF may

PAGE 38

27 make use of additional mechanisms and protocols to achieve additional functionality such as user authentication, accounting, policy information storage, etc. Figure 8. Two possible policy-ba sed network architectures In the case of vertical handoff, the polic y database holds info rmation regarding the metrics to be considered for a vertical ha ndoff, where handoff metrics are the measured qualities that give an indication of whethe r or not a handoff is needed. As stated previously, in traditional handoffs, only RSS and channel availability are considered. In the envisioned 4G system, the followi ng new metrics have been suggested: Service Type . Different types of services require various combinations of reliability, latency and data rate. Monetary Cost. A major consideration to users, as different networks may employ different billing strategies that may affect the userÂ’s choice to handoff. Network Conditions. Network -related parameters, such as traffic, available bandwidth, network latency, and conges tion (packet loss) may need to be considered for effective network usage. Us e of network information in the choice to handoff can also be useful for load bala ncing across different networks, possibly relieving congestion in certain systems. Network node PEP PDP Policy DB Network node PEP PDP Policy DB Policy Server (a) PEP and PDP locate in the same network node (b) PEP and PDP locate in different network nodes

PAGE 39

28 System Performance. To guarantee the system performance, a variety of parameters can be employed in the handoff decisi on, such as the channel propagation characteristics, path loss, inter-channel interference, signal-to -noise ratio (SNR), and the bit error rate (BER ). In addition, battery power may be another crucial factor for certain users. For example, when the battery level is low, the user may choose to switch to a network with lower power requirements, such as an ad hoc Bluetooth network. Mobile Terminal Conditions. MT cond ition includes dynamic factors such as velocity, moving pattern, moving hi stories and location information. User Preferences. User preference can be added to cater to special requests for users that favor one type of system over another. The use of new vertical handoff metr ics and the policy-based networking architecture increases the complexity of the handoff process and makes the handoff decision more and more ambiguous. However, th e use of an optimized cost function can simplify the handoff process and speed up the handoff decision. Then, intelligent techniques can be developed to evaluate the effectiveness of new decision algorithms, balanced against user satisf action and network efficiency. In the next chapter, a new vertical handoff cost function is introduced, that takes into account a wide variety of handoff metrics, and a novel vertical handoff policy is described for achieving a network optimized quality of service. 3.2.2 Proposed Vertical Handoff Interworking Scenariors To demonstrate the operation of the polic y-based architectures, the following two scenarios are explored: (1) Network-C ontrolled Handoff (NCHO)/Mobile-Assisted Handoff (MAHO), where the network generates a new connection, finding new resources for the handoff and performing any additi onal routing operation, and (2) MobileControlled Handoff (MCHO), where the mobile terminal must take its own measurements and make the handoff decision.

PAGE 40

29 3.2.2.1 NCHO/MAHO As shown in Figure 9, the handoff decision procedure begins with the PEP. Upon receiving a handoff trigger, the PEP formulat es a request for a po licy decision and sends it to the PDP. The request fo r policy control from the PEP to the PDP may contain one or more policy elements extracted from the MT that are necessary for handoff decision. The PDP then extracts other necessary informa tion, e.g., the userÂ’s subscriber profile and network conditions, from the database locate d in local or home network, makes the handoff decision, and returns the decision message to the PEP. (The handoff decision is made using utility function based algorithms as proposed in Zhu et al. [10].) The PEP then informs the MT about the handoff d ecision, and enforces the policy decision by handing off to the target network. Figure 9. NCHO or MAHO ha ndoff decision procedure In NCHO/MAHO, the network makes the decision for handoff. Thus we propose that the PDP point is represented by the ba se station (BS) or access point (AP). BS/AP MT Handoff Trigger PDP REQ DEC PEP DEC

PAGE 41

30 With tight coupling, WLAN traffic and UMTS traffic merged into the UMTS core network via the SGSN. A NCHO or MAHO with PDP located in SGSN can conveniently provide seamless vertical handoff. The signaling diagram for UMTS/WLAN tight coupling is shown in Figure 10. Figure 10. Signaling duri ng vertical handoff When the MT enters cell boundary, it ma y actively send SGSN (via RNC/WLAN AN) its updated handoff related personal info rmation (as in Message 4. INQ-RPL) such as battery life and changed user preferences, or passively respond when SGSN detects its location and inquires these information (in Message 1.Â’ INQ). To reduce handoff latency, registration and authentication information may also be encapsulated for pre-registration and pre-authentication. In anycase, when a handoff trigger is received, the PEP forms a REQ message (Message 1) to request a hando ff decision from the PDP. PDP may extract authentication information and subscriber profile retrieval from 3GPP AAA server (in Messages 2 and 3), if they are not yet av ailable in local database. The inquired MT SGSN (PEP) (PDP) HLR/HSS 3GPP AAA Server 1. REQ 2. INQ 3. INQ 7. DEC 6. INQ-RPL 5. INQ-RPL Making handoff decision 1Â’. INQ 4. INQ-RPL

PAGE 42

31 information is provided in Messages 5 a nd 6. In Message 7, SGSN makes the handoff decision and returns it to the MT. In Zhu et al. [10], we described a list of possible new metrics for vertical handoff which include service type, monetary cost, network conditions, system performance, MT conditions, user preferences, etc. To reduce signaling overhead, each measured metric is quantized into 7 levels, from 1 to 7. Zero is used when no value is currently available for a certain metric. For example, 7 indicates that the mobile deviceÂ’s measured channel quality is perfect, while 1 indicates a very noi sy channel. In this way, 3 bits are enough to represent each metric that is sent to PDP. For those non-measured metrics, such as user preference, number 1 to 7 is translated as a userÂ’s preferred network at PDP. For example, 1 represents UMTS, 2 to 7 represen t different types of WLANs. An example of the message format of INQ-RPL message MT sent to PEP is shown in Figure 11. 128 bits 3 bits 3 bits 3 bits 3 bits 16 bits MT ID Request bandwidth Battery level Measured channel quality User preference Reserved for other metrics Figure 11. HO trigger message format REQ and INQ Messages are short control messages. The INQ-RPL message includes the inquired metric values from the ne twork side. Their format is similar to the one shown in Figure 11. With loose coupling, there is no convenient network node, such as SGSN, that is a convergence of both WLAN and UMTS tra ffic. Thus, we only recommend MCHO, which will be discussed in the next section.

PAGE 43

32 3.2.2.2 MCHO Figure 12. MCHO handoff decision procedure In MCHO, the MT finds the new resources and the network approves. Thus, we proposed that the PDP is located at the MT. As shown in Figure 12, when the MT detects a severe QoS degradation, its PEP module triggers the handoff decision process by sending a handoff decision request message to the PDP. While some information is already available at lo cal database, the PDP may also n eed other necessary information, such as network conditions, from the network devices. Other information may not be immediately available at the BS or AP, a nd may need to be extracted from the home network. Upon receiving all handoff metrics, the PDP makes the handoff decision and returns the decision to the PEP. The PEP then informs the network the handoff decision by forwarding the DEC message, along with enforced authentication information. A Policy DB PDP Home Agent REQ DEC BS/AP PEP Handoff Trigger DEC MT

PAGE 44

33 handoff will take place once the network appr oves. The corresponding signaling diagram is shown in Figure 13. MCHO is a more practical solution for l oose coupling, since, unlike tight coupling, there is no convenient network node, such as SGSN, that is a convergence of both WLAN and UMTS traffic. Figure 13. Signaling duri ng vertical handoff For loose coupling, after detecting multiple available network interfaces, the MT sends inquiries (INQ) to RNC/WLAN AN re spectively. We assume that the WLAN can use the subscriber databases in the UMTS network for security and billing purposes. WLAN may also have local AAA agent and o ffline billing server, supporting its own network management mechanism. After collect ing all necessary information (Message 5 to 7), the PDP is able to make the ha ndoff decision and inform network devices. MT ( PEP ) ( PDP ) 1. REQ 9. DE C RNC/WLAN AN HLR/HSS 3GPP AAA Server 2. INQ 3. INQ 4. INQ 7. INQ-RPL 6. INQ-RPL 5. INQ-RPL Making handoff decision 8.DEC

PAGE 45

34 MCHO can also be implemented for tight coupling. The signaling diagram will be exactly the same as in Figure 13, except that SGSN in this case will be responsible for collecting handoff metrics for both UMTS and WLAN. 3.3 Performance Analysis 3.3.1 Policy-Based Handoff Overhead Since each metric introduces 3 bit over head, the total overhead introduced for metric inquiry is 3N, where N is the number of metrics considered. In fact, since some handoff metrics are available at PDP, e.g., netw ork parameters are available at SGSN in a NCHO or MAHO, most user information ar e available at MT in a MCHO, signaling overhead is less than 3N. The upper limit of radio link overhead for NCHO/MAHO (tight coupling for UMTS/WLAN) and MC HO (loose coupling for UMTS/WLAN) are calculated as DEC INQ NCHOO O N O 3 (1) DEC INQ MCHOO O N O 3 (2) where OINQ and ODEC are overhead by INQ and DEC messages. As we can see, the total overhead depends on the number of metric s. If a handoff decision replies more on information from MT side, then a MCHO requi res least radio link ove rhead, otherwise, a NCHO or MAHO is a better choice. 3.3.2 Policy-Based Handoff Decision Latency The handoff latency is total aggregate handoff decision signaling delay of each message and handoff time: HO m i i HOT T 1 (3)

PAGE 46

35 where Ti is the signaling delay for message I , HO is the handoff time. Ti consists of transmit time, , propagation time, , and processing time, , for message i . i i i iT (4) The handoff latency for tight c oupling is calculated as HO PD tight HOT T T T T T T T T T 7 4 ' 1 6 5 3 2 1)} ( ), max{( (5) where ) , max( b arepresents the bigger value between a and b , and TPD is the time that PDP needs to make a handoff decision. Similarly, the handoff latency fo r loose coupling is calculated as HO PD i i loose HOT T T 9 1 (6) Comparing (5) and (6), we can see that tight coupling introduces less handoff delay. 3.3.3 Probability of Policy-Based Vertical Handoff Failure We now calculate the probability that the policy-based vertical handoff will fail due to the MT leaving the cell before the handoff decision signaling completes. Let T be a random variable that takes on values of the time to the next consecutive handoff after the MTÂ’s arrival into the a cell, i.e., the time that the MT resides in the cell. Then the probability that the MT leaves th e cell before the handoff delay, THO, is ] [ PrHOT T ob P (7) If we assume that T is exponentially distributed, then HOT HOe T T ob 1 ] [ Pr (8)

PAGE 47

36 To restrict the probability to certain threshold, Pf, then THos should satisfy: f TP eHO 1 (9) where is the arrival rate of MTÂ’s into the cell. We assume an MTÂ’s direction is uniformly distributed within [0, 2 ). Then the arrival rate is: S vL (10) where v is the expected velocity of the MT, L is the length of the perimeter of the cell, and S is the cell area [39]. For cells in hexagonal shape, ) 3 / sin( 2 l v , (11) where l is length of each side of the hexagon. Substituting (11) into (9), the velocity of the MT is limited to: HO fT P l v 2 ) 1 ln( ) 3 / sin( (12) 3.4 Numerical Results We analyze the impact of the policy-ba sed vertical handoff latency on system performance using the parameters shown in Table I. Figure 14 demonstrates the handoff failure, Pf, comparison with the MTÂ’s incr easing velocities, where handoff time is assumed to be 50ms and 100 ms respectively. As we can see, the handoff failure increases with the increasing velocity of MTs. Since NCHO takes less time for handoff signaling, the handoff failure incr eases more slowly than MCHO.

PAGE 48

37 Table 1. Parameters Parameters Values Radio link bandwidth UMTS(up to 2Mbps) 802.11b (1, 2, 5.5, 11 Mbps) AAA server processing time 10ms TPD 10ms other processing time 2ms Propagation time between SGSN and AAA server 0.5ms Propagation time between AAA server and HLR/HSS 0.1 ms INQ message 5 bytes DEC message 5 bytes INQ-RPL message 20 bytes Handoff time ( HO) 50ms or 100ms l 300m 0 1 2 3 4 5 6 7 8 9 10 x 104 0 10 20 30 40 50 60 70 80 90 100 MT velocity, v(km/h)handoff failure, Pf(%) NCHO (tight coupling, HO=50ms) MCHO (loose coupling, HO=50ms) NCHO (tight coupling, HO=100ms) MCHO (loose coupling, HO=100ms) Figure 14. Handoff failure comparison ( HO=50ms, 100ms) Thus, with the same handoff failure limit, NCHO can support MTs with higher mobility. Comparing the handoff failure curves with different handoff time, we can see

PAGE 49

38 that handoff time has a greater impact on system performance. In other words, the additional signaling delay intr oduced by policy-based handoff is not a major problem. In any case, the policy-based hando ff architecture can support mob ile users of a large range of mobility with very lo w handoff dropping probability. Figure 15 shows the impact of handoff deci sion time on handoff failure rate for mobile users of high and low velocities respec tively. It can be obser ved that the handoff failure increases with the increase of polic y-based handoff decision time, and it is more dramatic for users with higher mobility. As long as the PDP has reasonably fast computation capacity, low handoff fa ilure rates can be achieved. 0 2 4 6 8 10 12 14 16 18 20 0 10 20 30 40 50 60 70 80 90 100 handoff decision time TPD(s)handoff failure rate, Pf (%) NCHO (tight coupling, Low velocity) MCHO (loose coupling, Low velocity) NCHO (tight coupling, high velocity) MCHO (loose coupling, high velocity) Figure 15. Handoff failure comparison

PAGE 50

39 A significant challenge for coordination the different types of ne twork in the next generation wireless networks is vertical handoff. In this paper, we propose a policy-based framework that translates handoff policies in to network configuration is proposed to implement vertical handoff protocols in orde r to produce a satisfactory result for both the user and the network. Various handoff metric s are able to be incorporated for an intelligent and adaptive handoff decision. A case study of UMTS/WLAN interworking is provided using the policy-based vertical ha ndoff architecture. Performance analysis demonstrates that the additional handoff si gnaling latency is low enough to support users of various mobility with low handoff failure rate.

PAGE 51

40 CHAPTER 4 MULTI-SERVICE VERTICAL HANDO FF DECISION ALGORITHM (MUSE-VDA) Future wireless networks must be able to coordinate services within a diverse network environment. For example, a widely-d eployed 3G cellular and data service, such as GPRS, may be supplemented by the local de ployment of high bandwidth wireless local area networks (WLANs), such as IEEE 802.11 and HiPerLAN. Furthermore, as shown in Figure 16, existing networks, such as satellite, cellular, and WLAN, will need to integrate with emerging networks and technologies, su ch as wireless mesh networks and Wi-Max to allow a user to transparently a nd seamlessly roam between systems. Figure 16. Diverse 3G and 4G wireless networks

PAGE 52

41 Seamless roaming involves handoff, which is the process of maintaining a mobile userÂ’s active connections as it moves within a wireless network [1]. Vertical handoff, or inter-system handoff, involves handoff between different types of networks [4, 18, 19, 21]. Traditionally, handoff research has been based on an evaluation of the received signal strength (RSS) at the mobile node. Ho wever, traditional RSS comparisons are not sufficient to make a vertical handoff decision, as they do not take in to account the various attachment options for the mobile user. Othe r factors, such as, monetary cost, network conditions, mobile node conditions , user preferences, etc., must be considered, as well as the capabilities of the various networks in the vicinity of the user. Thus, a more complex, adaptive and intelligent approach is needed to implement vertical handoff protocols to produce a satisfactory result for bot h the user and the network. Related work on vertical handoff has been presented in recent research literature. Several papers have addressed designing an architecture for hybrid networks, such as the application-layer Session Initiation Protocol (SIP) [41], the hierarchical mobility management architecture proposed in Badis [ 42], and the P-handoff protocol [43], which complemented classical vertical handoff by re directing traffic to th e best ad-hoc link, such as BlueTooth and 802.11b, on a peer by pe er basis. However, these papers focused on architecture design and did not address the handoff decision point or the vertical handoff performance issues. Other work consid ered optimizations after the vertical handoff decision has been made, measuring pe rformance with respect to handoff latency [44], TCP timeout and throughput [46, 47], and packet loss [48]. However, the vertical handoff decision did not consider multiple netw orks supporting multiple services for each user.

PAGE 53

42 The related papers that explored ve rtical handoff decision mainly focus on traditional issues, such as RSS and data rate. In Zhang et al. [45], a Fast Fourier Transform (FFT)-based signal decay detection scheme was used to reduce the ping-pong handoff effect, and an adaptive threshold c onfiguration approach was proposed to prolong the time a user stays in WLAN. In Y lianttila et al. [48, 50], a vertical handoff algorithm was proposed that took into account RSS, data rate, and packet loss due to handoff delay for a single service per user. A vertical handoff system based on computed background noise and signal strength was propos ed in Sharma et al. [51]. The WISE handoff decision algorithm was proposed to maximize energy-efficiency without sacrifice of overall network degradation in Nam et al. [52]. A QoS-based handoff method between UMTS and WLAN was proposed in Ju ng et al.[53], but the definition of QoS was not defined in the paper. Finally, severa l papers have focused on mobility level and user position in the network. Mobility level was proposed as a proper metric for multi-tier handoffs in Holtzman et al. [31]. Multi-network architectural issues were explored, and an advanced neural-network-based vertical handoff algorithm was developed to satisfy user bandwidth requirements in Pahlavan et al. [4]. A vertical handoff algorithm based on pattern recognition was pres ented in Mehbodniya et al . [54]. Although the abovementioned research addresses handoff decisi on, most address 3G/WLAN issues, and do not provide a way to incorporate a general, user-defined idea of quality of service, on which to base vertical handoff decisions. Several papers have created utility func tions to better evaluate the choice for vertical handoff. The vertical handoff deci sion function was a measurement of network quality in Hasswa et al. [55]. However, no pe rformance analysis was provided. An active

PAGE 54

43 application-oriented handoff decision algorith m was proposed in Chen et al. [56] for multi-interface mobile terminals to reduce the power consumption caused by unnecessary handoffs and other unnecessary interface activa tion, and in Wang et al. [9], a policyenabled handoff decision algorithm was pr oposed along with a cost function that considers several handoff metrics. Howeve r, multi-service handoff was not fully discussed. However, the multiple active servi ces case was not considered. Chen et al. [57] adaptively adjusted the handoff stabilit y period based on a utility function to avoid unnecessary handoffs and reduce decision time. Finally, the authors have presented a tutorial on vertical handoffs in McNair et al . [21], and in Zhu et al . [10], introduce a cost function-based vertical handoff decision algorithm for multi-services handoff. Preliminary results demonstrated significant gain in throughput. This paper extends the work in Zhu et al. [10] to examine the sy stem performance with respect to blocking probability and user satisfactions, i.e., the abilit y of the network to satisfy all of the usersÂ’ simultaneous requests. In this chapter, we propose several optim izations to the handoff decision process and make the following contributions: (1) the development of a handoff cost function that accounts for the dynamic values that are in herent to a vertical handoff, (2) the incorporation of a network elimination process in the vertical handoff metric, to potentially reduce delay and processing pow er in the handoff calculation, and (3) a constraint optimization analysis for the propos ed handoff cost function for different types of user services spread among multiple netw orks. In Section I a novel cost function is introduced to judge target ne tworks based on a variety of userand network-valued metrics. Section II provides a performance analysis. In Section III, numerical results

PAGE 55

44 demonstrate significant gains in quality of service and a more efficient use of resources can be achieved from the proposed optimizations. 4.1 MUSE-VDA Algorithm a nd Cost Function The MUSE-VDA vertical handoff cost func tion measures the benefit obtained by handing off to a particular network. It is evaluated for each network n that covers the service area of a user. The netw ork choice that results in th e lowest calculated value of the cost function is the netw ork that provides the most be nefit, where the benefit is defined by the given handoff policy. The cost function evaluated for network n includes the cost of receiving each of the userÂ’s requested serv ices from network n, and is calculated as s n s nC C, (13) where s is the index representing the user-requested services, and n sC is the per-service cost function for network n. n sC represents the QoS experi enced by choosing to receive service s from network n, and is calculated: j n j s n j s n sQ W C, ,, (14) where the n j sQ, is the normalized QoS provided by network n for parameterj for service s, and n j sW,is the weight indicating the impact of the QoS parameter on the user or the network. n sC includes both a normalized value for the QoS parameter and a weight for the impact of the parameter on either the user or the network. For an example from the userÂ’s perspective, suppose a mobile terminal requ ests a service with a specified minimum

PAGE 56

45 delay and minimum power consumption requirem ent. If the mobile terminal has a low battery life, the power consumption takes on greater importance than meeting the delay constraints. For an example of a netw ork-based QoS request and the corresponding impact, the availability of the services request ed by the user in the target network impacts the network congestion in the target network. Using the im pact factor, the network may direct users toward a less desira ble, but less congested network. The handoff decision problem thus equals to the following cons traint optimization problem: i s E t s Q W C Cn i s ssj n j s n j s n s n n , , 0 . . min, , , (15) where n i sE, is the network elimination factor, indicating whether the constraint ifor service s can be met by network n. It equals to one if constraint ican be satisfied, and equals to zero if constraint i can not be satisfied. It is introduced to reflect the inability of a network to guarantee the requested QoS constraints for a particular service s, and can be implemented as a check-list at PDP. For example, an available network may not be able to guarantee the minimu m requested delay for a real time service, and should be immediately removed from consideration as a handoff target for the requested service. The application of the vertical handoff co st function is flexible to allow for different vertical handoff policies. To dem onstrate the performance of the new cost function, two different policy scenarios are explored. It is assumed that a single user may conduct multiple communication sessions. In the first vertical handoff policy, the vertical handoff decision is optimized for all sessions collectively, i.e., all of the userÂ’s active sessions are handed off to the same target

PAGE 57

46 network at the same time. The cost function, nC, Equation (13), is determined for all sessions going to a single network. The optim al target network for handoff is determined by solving Equation (15). The second vertical handoff policy prioriti zes each service and then optimizes the vertical handoff decision individually for each session, i.e., each of the userÂ’s active sessions may be independently handed off to a different target network. In this scenario, the mobile terminal maintains a list of its current active sessions, arranged in priority order. Then, the cost function, n sC, Equation (14), is evaluated for the highest priority service. The optimal target network is chosen by minimizing the per-service cost: i E t s Q W Cn i s s n j s n j s n s n , 0 . . min, , , (16) Then, the next highest priority service is selected, the corresponding cost function is evaluated, and the target network determin ed. The process continues to the last active session. If the constraints for one session cannot be met, then the user loses the individual session only. The process for the second scenario is outlined in Figure 17. In the next section, the performance of the proposed cost function optimization is demonstrated for a typical 3G multi-network environment. 4.2 Performance Analysis For effective comparison with other techni ques, the performance analysis considers the case of 3G/WLAN handoff scenario, where signal strength, channel availability, and bandwidth are the specified constraints. Howe ver, note that any of the vertical handoff metrics listed in Section 2 can just as easily be substituted in the evaluation. The top view of a typical 3G multi-network environment is shown in Figure 18, which shows three

PAGE 58

47 networks of different data rates co-exis ting in the same wireless service area. For example, Network1 (centered at A ) and Network 2 (centered at B ) may each represent WLANs [25], while Network 3 (centered at C ) may be a GPRS network [26, 28]. The shaded circles on the left and right represen t the area where RSS from Network 1 or 2 is stronger than that from Network 3. To hi ghlight the effects of the vertical handoff procedure among the three networks, only the users within the boundary cell area, i.e., the dashed square in Figure 18, are considered. Figure 17. Scenario 2—Prioritized user sessions Begin with a list of active services Select the service with highest p riorit y End No Update resource database Any unassigned services left? Yes Evaluate n sC, Equ (16) for each possible target network Handoff to network n based on the optimal result of Equ (16)

PAGE 59

48 A user with active sessions that enters the overlay of all three networks must decide when and where to execute a vertical hando ff request. If the request is accepted, the appropriate amount of bandwidth is assigned by the serving network. If the request is denied at one network, the request can be re-a ssigned to another network, if resources are available. If the second (or thir d) network is not available, th e request is blocked from the system. Figure 18. Cell overlay network 4.2.1 Blocking Probability Each of the three networks in Figure 18 is modeled as an M / M /1/ Nn queue system [30], where Nn is the number of available channels in Network n . Nn is calculated: D B Nn n (17) D E G C F M H I J K L O P Q R S T A B SN1-N3 SN2-N3 SN3-N1 SN3-N2 SN3 SN3 SBOUND Network 1 Network 2 Network 3 SN3-N1N2

PAGE 60

49 where Bn is the total bandwidth of Network n , and D is the average data rate of each user. The traffic load within the overlay cells is , where is the arrival rate of service requests, µ is the departure rate, and arrivals and departures are modeled as Poisson distributions. Handoff ca lls are given a higher priori ty than new calls, and for simplicity, a buffer-less handoff algorithm is used. For the blocking probability of Network n , Pbn, we use the blocking probability of an M / M /1/ Nn queue when there are Nn users in system. For mo re detail on deriving the blocking probability, refer to [30]. 11 ) 1 ( n n nN n n N bnP (18) where n is the effective load experienced by Network n : n nr , (19) and rn is the percentage of total requests that will go to Network n , based on the vertical handoff decision metrics. To determine rn, all handoff requests to Network n are considered, including both original handoff requests and the handoff request s that arrive because the user has been rejected by another network. Si nce it is assumed that the user s are uniformly distributed, the service request load can be calculated according to each network’s proportion of the coverage area within the boundary area. Th e corresponding coverage is described in Table 2and labeled in Figure 18.

PAGE 61

50 For RSS based handoff algorithm, the values of rn for n =1,2,3 are calculated as follows. BOUND N N N N BOUNDS S S S r3 2 3 1 3 (20) BOUND bS P S S r3 N1 N3 N3 N1 1 (21) BOUND bS P S S r3 N2 N3 N3 N2 2 (22) where Pb3 is defined in Equation (18), Si is the geometric area of region i described in table 2, and SBOUND is the geometric area of the boundary region. Table 2. Region names and descriptions Region Number Area Description 1 SN3 (DEH, JFG) Network 3 pr ovides the only coverage. 2 SN3 -N1N2 (HIJK) Network 3 is the st rongest. In RSS based handoff algorithm, if the request is deni ed by Network 3, the user can try either Network 1 or 2 with equal probability. In the MUSE-VDA, the selection or der will be N1->N2->N3. 3 SN3-N1 (DHKJGP) Network 3 is the st rongest. In RSS based handoff algorithm, if the request is deni ed, the user can try Network 1 only. In the MUSE-VDA, the selection order will be N1>N3. 4 SN3-N2 (EHIJFS) Network 3 is the st rongest. In RSS based handoff algorithm, if the request is denied the user can try Network 2 only. In the MUSE-VDA, the selection order will be N2>N3. 5 SN1-N3 (OPQA) Network 1 is the st rongest. In RSS based handoff algorithm, if the request is deni ed by Network 1, user can try Network 3 only. In the MUSE-V DA, the selection order will be N1->N3. 6 SN2-N3 (RSTB) Network 2 is the st rongest. In RSS based handoff algorithm, if the request is de nied by Network 2, users can try Network 3 only. In the MUSE-VDA, the selection order will be N2->N3. SBOUND Boundary region

PAGE 62

51 In the MUSE-VDA, the values of rn for n =1,2,3 are calculated: BOUND N N N N N N NIS S S S r2 1 3 1 3 3 1 (23) BOUND b N N N N N N NS P S S S r1 2 1 3 2 3 3 2 2 (24) ] ) ( ) ( [ 12 2 1 3 2 3 3 2 1 1 3 3 1 3 3 b N N N N N N N b N N N N N BOUNDP S S S P S S S S r (25) 4.2.2 Mobility Model UsersÂ’ trajectories are ch aracterized by the Random Waypoint (RWP) model [29], including the adjustments for the shortcomi ngs of the model described in Yoon et al. [58]. Each user chooses uniformly at random a destination point (or waypoint) in the dashed rectangular in Figure 18. A user m oves to this destinat ion with a velocity v , which is chosen uniformly in the interval [ vmin, vmax]. The vmin and vmax are chosen as 0.3m/s and 12.5m/s, respectively. When the user reach es the waypoint, it remains static for a predefined pause time, and then moves again according to the same rule. Note that user trajectories characterized by the improved RW P model can be assumed to be uniformly distributed at any given time. 4.3 Numerical Results The 3G/WLAN overlay system was modeled using MATLAB, according to parameters shown in Table 3. We assume that each user requests up to a maximum 500kb/s data rate which includes a constant bit rate (CBR) service of 50kb/s and an available bit rate (ABR) service of up to 450kb/ s. If Network 1 or Network 2 is selected

PAGE 63

52 as the handoff target, then the userÂ’s 500kb/s request can be fully satisfied due to the large data rate available to each user in th ese networks. If Network 3 is selected, only 30% of a 500kb/s request can be satisfied. Table 3. Parameters used in the numerical results Network ( n ) Resource (Bandwidth) Type of network Mobility 1 2Mbps [25] WLAN 2 1Mbps [25] WLAN 3 up to 8 slots, 21.4kbps each [26, 28] GPRS vmin = 0.3 m/s; vmax =12.5 m/s; vthreshold =5.5m/s We estimate the average percentage of us ersÂ’ satisfied requests (APUSR) in the overlay network: i i R RR P A T Ei) (, (26) whereiRA is the APUSR for Region iin Table 2. iRT is calculated: j j i j i RN P t Ti) (, ,, (27) where ti,j is the maximum throughput that can be received from Network Ni,j in Region i, and P( Ni,j) in Equation (27) is th e probability that Network Ni,j is available and chosen by a user as the serving network. P ( Ri) is the probability that a user is located in Region i, and is found in term of the geometrical proportion of Region i to the entire boundary region: BOUND i iS S R P . (28)

PAGE 64

53 where users are assumed to be uniformly di stributed. This has been verified by a comparison of theoretical results based on th e uniform distribution and simulation results based on the improved RWP model. 4.3.1 RSS Performance First, the RSS performance is examined to provide a baseline for comparison with the MUSE-VDA results. Figure 19 shows the APUSR with the increasing network load for an RSS-based handoff algorithm. Since Netw ork 3 has the strongest transmit power, it is the preferred service provider. Thus, at the low load range, Network 3 must satisfy a large portion of the total requests. With increasing network load, the resources of Network 3 are used up earlier than the resources of the other two networks. 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 APUSR network 1 network 2 network 3 total Figure 19. APUSR provided by RSS-based algorithm

PAGE 65

54 Figure 20 demonstrates the corresponding bl ocking probability of each network for the traditional RSS algorithm. An increase in blocking probability of Network 3 earlier than Networks 1 and 2 can be observed. Mobile users thus have a gr eater chance to select Network 1 and Network 2 as service provider. Since they have a total APUSR that is higher than Network 3 by itself, a "hump" can be observed. The result that Network 3 is chosen more often as the target handoff cell, leads to two unsatisfactory effects: (1) unbalanced load assignment and (2) low overa ll achievable data ra te. Only when the resource in Network 3 is highly consumed, Networks 1 and 2 will have a greater chance to be the service provider. Thus a more in telligent handoff algorithm that can balance the usage of overlay networks is needed, and a higher overall APUSR is expected. 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 blocking probabilities network 1 network 2 network 3 Figure 20. Blocking probability of ea ch network in RSS based algorithm

PAGE 66

55 The result that Network 3 is chosen more often as the target handoff cell, leads to two unsatisfactory effects: (1) unbalanced lo ad assignment and (2) low overall achievable data rate. Only when the resource in Netw ork 3 is highly consumed, Networks 1 and 2 will have a greater chance to be the service provider. Thus a more intelligent handoff algorithm that can balance the usage of overl ay networks is needed, and a higher overall throughput is expected. 4.3.2 RSS with Mobility Performance 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 APUSR RSS RSS+mobility MUSE-VDA Figure 21. APUSR comparison Next, we compare the RSS only technique versus a mobility-level technique. Mobility level is a metric that can be co mbined with RSS based to improve system performance. For example, fast moving users ( v > vthreshold) are selected to receive service from the largest cell, while medium to slow users ( v < vthreshold) receive service from the

PAGE 67

56 small cells. Figure 21 and Figure 22 show the APUSR and blocking probability comparison of the pure RSS based algorithm and the RSS based algorithm combined with mobility level consideration. The m obility level algorithm demonstrates an improved APUSR performance. However, its ac hievable APUSR is lower than that of MUSE-VDA (which will be discussed in more details later in this section), i.e., there remains a load-balancing issue for increasing requests. 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 blocking probabilities RSS RSS+mobility Figure 22. Blocking probability comparison 4.3.3 MUSE-VDA performance We now examine the MUSE -VDA performance by c onsidering two handoff scenarios: (1) collective handoff, where all of the userÂ’s active sessions are handed off to the same target network at the same time, and (2) prioritized multi-network handoff,

PAGE 68

57 where each service is prioritized and optimal decision is made individually for each session. 4.3.3.1 Scenario 1: Collective Handoff The cost function to be considered in this scenario is: n n s n s nB W B W C C2 2 1 1 2 11 ln 1 ln (29) where n sB is the offered bandwidth for service s by network n . Note that the bandwidth is normalized by taking the logarithm of the reciprocal value. Here, s =1 corresponds to RSS, while s =2 corresponds to channel availabil ity. The handoff target network is decided by solving the optimization problem: 0 0 . . 1 ln 1 ln min1 2 1 req n th n n n n nB B R R t s B B C (30) where W1= W2=1, nR is the RSS from network n, thR is the received power threshold, and reqB is the user-requested CBR bandwidth. 4.3.3.2 Scenario 2: Prioritized Multi-Network Handoff Since there are two services being requested (CBR and ABR), there are two cost functions to be considered. The cost function for the CBR service is: n nB C1 11 ln (31) and the cost function for the ABR service is:

PAGE 69

58 n nB C2 21 ln (32) The optimization problem considered for CBR service is: 0 0 . . 1 ln min1 1 1 req n th n n n nB B R R t s B C (33) and the optimization problem considered for ABR service is: 0 . . 1 ln min2 2 th n n n nR R t s B C (34) The target network is then chosen accordi ng to the procedure de scribed in figure 17. 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 APUSR network 1 network 2 network 3 total Figure 23. APUSR provided by MUSE-VDA

PAGE 70

59 Figure 23 shows MUSE-VDA results for the APUSR provided by each of the three networks and overall achievable APUSR im plementing the collective handoff algorithm, for comparison with Figure 19, the RSS-only case. Since either ne twork 1 or network 2 provides relatively larger data rate than netw ork 3, they are the default service provider for the mobile users, depending on their locati on. Thus, at the low load range, network 1 and network 2 satisfy the most portion of th e total request. With the increasing network load, the resource of network 1 and network 2 is used up earlier than the resources of network 3. Then mobile users start to select network 3 more frequently than in low load range. The portion of requests satisfied by ne twork 3 thus starts to increase when the portion satisfied by network 1 and network 2 decreases. 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 blocking probabilities network 1 network 2 network 3 Figure 24. Blocking probability of each network in MUSE-VDA

PAGE 71

60 Figure 24 demonstrates the corresponding blocking probability of each network. An increase in blocking probability of Networ ks 1 and 2 earlier than Network 3 can be observed, which indicates that WLANs are favor ite networks due to their relative larger available bandwidth to each user. 4.3.4 Results for More Demanding CBR Services We now present results with random requests for the APUSR and blocking probability for three cases: the traditional handoff protocol based on the strongest RSS (“RSS”), the cost function with collectiv e handoff (“collective MUSE-VDA”), and the cost function with the prioritized, multi-n etwork optimization (“prioritized MUSEVDA”). In this case, a user may request up to 50 kbps CBR service, as well as an additional, varying ABR service, which is uniformly distributed between 0 and 0.95 Mbps. The user chooses the best network base d on the given constr aints and is blocked when a session cannot be supported. In the “collective MUSE-VDA” case, the user is blocked when a single session cannot be s upported. In the “prioritized MUSE-VDA” case, each session can be blocked individua lly, thus a single blocked session may not necessarily cause the us er to be blocked. Figure 25 shows the APUSR and blocking rate versus user requests, for the three cases. In Figure 25 (a), the APUSR refers to the ratio of the actual data rate over the requested rate. In Figure 25 (b), the corresponding blocki ng rates are shown. It is observed that significant gains in APUSR can be achieved without any sacrifice of the user blocking rate for the new optimized vertical handoff algorithms. In the RSS case, Network 3, is chosen more often as the targ et handoff cell, since it provides the strongest signal. This condition leads to two unsatisfact ory effects: (1) unbalanced load assignment

PAGE 72

61 and (2) lower overall achievable data ra te. On the other ha nd, with algorithm optimization, the user may choose a handoff target based on bandwidth criteria. In the “prioritized MUSE-VDA” case, APUSR can be maximized, since the mobile node’s connections can be distributed among the three networks according to bandwidth availability. Since the ABR service does not have minimum bandwidth requirement, it is blocked only when there is absolutely no resource for it. Thus the CBR service’s blocking probability is the dominant factor in the blocking probability calculation, and the three blocking probability curves demonstrate close behavior. The advantage of MUSE-VDA is not to reduce blocking probabili ty of individual services, but to better satisfy the requests of individual users with multiple services with a successful handoff. In fact, by spreading users multiple sessions into multiple networks, ABR service is able to get potentially more bandwidth in pr ioritized MUSE-VDA than the other two algorithms. Since ABR and CBR services are sharing the same total resource, the advantage of ABR service may in fact result in higher blocking probability for individual CBR services in prioritized MUSE-VDA. Expanding services through the use and coor dination of diverse networks creates the challenge of developing a more complex, adaptive and intelligent vertical handoff protocol. In this paper, new optimizations for vertical handoff decision algorithms have been developed to maximize the benefit of th e handoff for both the user and the network. The optimizations incorporate a network e limination feature to reduce the delay and processing required in the evaluation of the cost function, and a multi-network

PAGE 73

62 optimization is introduced to improve throughput for mobile terminals with multiple active sessions. A performance analysis dem onstrated significant gains in quality of 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 APUSR RSS collective MUSE-VDA prioritized MUSE-VDA (a) APUSR 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 blocking probabilities RSS collective MUSE-VDA prioritized MUSE-VDA (b) Blocking probability Figure 25. Comparison of the RSS-based, collective, and prioritized MUSE-VDA algorithms.

PAGE 74

63 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 CBR (kbps)APUSR RSS collective MUSE-VDA prioritized MUSE-VDA Figure 26. Comparison of throughput with changing of CBR request. service and a more efficient use of res ources from the proposed optimizations. Assume each user carries an ABR servi ce with a guaranteed bandwidth of 150kb/s, in addition to the 50kb/s CBR service. Figur e 27 shows APUSR and blocking probability vs. user requests for the three handoff algorithms . In this case, Network 3 is eliminated in RSS-based and collective MUSE-VDA handoff al gorithms, due to its limited data rate per user (less than 170kb/s). T hus, users will only be able to choose between Networks 1 and 2. This increases the APUSR in R SS-based algorithm in light traffic ( <1 in Figure 27), since users no longer join with Networ k 3. However, the APUSR in collective MUSE-VDA decreases, since Networks1 and 2 can not cover the whole area. On the other hand, all 3 networks can be used in prioritized MUSE-VDA, where userÂ’s two sessions can be spread into multiple networks. If the bandwidth for one session can not be

PAGE 75

64 satisfied, only one session will be blocked. This results in a higher APUSR and a lower blocking probability than the other two ha ndoff schemes. Moreover, since users moving out of the limited coverage of Networks 1 a nd 2 can not be served in RSS-based handoff algorithm and collective MUSE -VDA, a non-zero blocking pr obability can be observed all the time. Due to the flexibility prio ritized MUSE-VDA has, its performance is demonstrated similar to the performance in Figure 25. 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 APUSR RSS collective MUSE-VDA prioritized MUSE-VDA (a) APUSR Figure 27. Comparison of the RSS-based, collective, and prioritized MUSE-VDA algorithms.

PAGE 76

65 10-1 100 101 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 blocking probabilities RSS collective MUSE-VDA prioritized MUSE-VDA (b) Blocking probabilities Figure 27. Continued

PAGE 77

66 CHAPTER 5. LAYER 3 MOVEMENT ESTIMATION (L3ME) USING LAYER 2 HINTS As mentioned previously, Mobile IP, standa rdized by the Internet Engineering Task Force (IETF), allows mobile nodes (MN) to change their point of attachment to the Internet while still being able to maintain a connection to the network [14, 15]. The specific operation is as follows. Under Mobile IP, a MN that is currently residing in its home subnet is served by a home agent (HA) that forwards all incoming packets to the MN at its home IP address. When the MN moves away from its home subnet to a new location, the node must contact a Foreign Agent (FA) at the new subnet. Each FA periodically broadcasts an advertisement of the subnet prefix, so that MNs can make contact and obtain a new IP address for the ne w subnet. The new IP address is called a Care-of Address (CoA). A binding update must then be performed to notify the HA about the MNÂ’s new CoA. The HA then forwards all incoming packets to the MN using a process referred to as tunneling, i.e., the HA encapsulates the incoming packets for the MN and forwards them to the FA, which in turn encapsulates them and delivers them to the MN. Meanwhile, the MN can continue to transmit packets directly to the correspondent node. Each time a host changes its point-of attachment, it is possible that it will also have to change its IP-layer configurations, su ch as its IP address and default gateway information. In order to make these changes, the IP module has to detect the new network attachment, realize that the old configuration is no longer valid and obtain the new configuration parameters. In Mobile IP v4 and IPv6, the network detection phase

PAGE 78

67 traditionally uses network-layer movement detect ion, such as a change in the advertised subnet prefixes [14, 15]. However, general reliance on the network layer does not yield rapid detection, since these indications ar e not readily available upon a link change. In Yegin et al. [13], it has been identified that receiving explicit hints from the linklayer would expedite the detection process. The link-layer, indicating that the host has established a new connection, can be used as a hint to further probe the network for a possible configuration change. (“Possible”, si nce it might very well be the case that the host is still connected to the same IP subnet despite the link change.) In the third generation (3G) and fourth generation (4G) wireless networks, more than one type of network is expected to co exist under a Mobile IP based environment, and thus users will need to coordinate vertical (or inter-system) handoffs, which are handoffs between different types of networks [17, 18]. Under vertical handoffs, when the MN moves to the new subnet that belongs to a new type of network, not only a layer 3 (L3) handoff is triggered, but a conseque ntial inter-system handoff and inter-system registration should take place immediately as well. Thus, an accurate and rapid movement detection process can trigger the need for handoff quickly with less sacrifice in signaling cost. Related work on Mobile IP based handoffs with implementation of link hints has been presented in recent rese arch literature. In Yegin et al. [13], a non-exhaustive catalogue of link layer hints from well-know n link-layer technologi es was provided. Furthermore, a high-level abstraction is defined to categorize such hints. In Perkins et al. [14], a list of parameters to be transported in link hints was presented. In Zhu et al. [10], layer 2 (L2) hints were used for handoff deci sion optimization. In Park et al. [20], a

PAGE 79

68 mechanism that extends Mobile IPv6 protocol was presented with description of using the link events information to optimize layer 3 movement detection. It considered smooth handoffs for MNs equipped with multiple interfaces moving across different and heterogeneous links, e.g. between 802.11 and GP RS. In Malki et al. [16], pre-registration and post-registrations are proposed to achieve low-latency Mobile IP handoffs and to allow greater support for real-time services on a Mobile IPv4 network. Although the use of layer 2 hints for quick detection of laye r 3 movement was suggested, how layer 2 hints can facilitate inter-sys tem handoff was not fully developed. In this chapter, a cross layer (layer 2 and layer 3) movement detecti on technique is proposed and the following contributions are addressed: (1) the deve lopment of a layer 3 movement detection likelihood function that takes multiple layer 2 hints into account intelligently, and (2) the likelihood function is combined with a pre-registration pr otocol to reduce the handoff latency. The rest of this chapter is organized as follows. Section I presents the architecture of Mobile IP based 3G cell overlay network. Then in Section II, a likelihood function is defined for layer 3 movement detection base d on layer 2 hints. Section III provides the application of the lik elihood function in pre-registrati on inter-system handoff protocol. Finally, in Section IV, a perf ormance analysis demonstrates that significant gains in signaling cost can be achieved from the proposed cross layer solution. 5.1 L3ME Architecture A typical 3G wireless network and Wirele ss Local Area Network (WLAN) overlay structure is shown in Figure 28. The 3G wire less network, such as UMTS or GPRS, is assumed to cover a relatively larger ar ea than the WLAN. Network connectivity is offered to the MN through an Access Point (A P) that connects to an Access Router (AR)

PAGE 80

69 which belongs to either the 3G wireless ne twork or a WLAN. When a MN moves out of the coverage area of its current AP, it may attach to a new AP. The roaming of MNs between APs is managed by the link-layer protoc ol and is known as layer 2, or link layer handoff. The new AP can be connected to th e same access network, or to a different access network. If the new AP is connected to the same subnet as the old AP, the MN can continue its IP communication through the new AP through a layer 2 handoff without any configuration change at layer 3. Figure 28. 3G cell overlay network. If the new AP is connected to a different subnet, then the MN needs to configure a new IP address that is valid for the new subnet and use some additional mechanism to maintain its ongoing communication sessions, su ch as a pre/post-registration protocol [16]. In this case, the layer 2 handoff will result in a layer 3 handoff. AR2 AR3 AP1 AP2 3G network WLAN AP3 AP5 AP6 AP4 AR1

PAGE 81

70 Furthermore, if the current AP and the new AP are connected to two different systems, e.g. 3G network and WLAN as in Fi gure 28, the layer 2 handoff will result in an inter-system handoff. Because certain layer 2 tec hnologies are capable of provi ding various link status information to the IP module, such as conn ected or disconnected, th e link identifier can help the IP module make intell igent decisions regarding confi guration changes. This help is referred to as a layer 2 hint. Next we develop a likelihood function for layer 3 movement detection based on layer 2 hints. 5.2 Layer 2 Hints Two sets of parameters have been proposed to gather layer 2 information: (1) the MN set, which gathers information available at the MN, and (2) the AR set. In each set, parameters are grouped into three distinct categories: static parameters – which are related to the hardware impl ementation of the interface, configuration parameters, which are managed through interface configuration, a nd status parameters – which are highly varying in order to provide the current link e nvironment of the interface. In this research, we assume a mobile controlled handoff s cenario, only considering the parameters available at MN, i.e., the MN set. The netw ork controlled or m obile assisted handoff algorithms can be developed accordingly. 5.2.1 Categories of Layer 2 Hints To quickly detect layer 3 movement, the following categories of layer 2 hints are considered to be useful: (a) Link-type hint: Characteristics that describes the type of the technology from which the layer 2 trigger was generated. Examples include:

PAGE 82

71 MN measured bandwidth : current available bandwidth measured by the MN over the link. MN bit error rate : current measured bit error rate. MN packet error rate : current measured packet error rate. MN current data rate : current rate at which the MN link layer is transmitting/receiving packets. (b) Link identifier: For example, in GPRS networks , the relevant link-identifiers are the Transaction Identifier (TI), which includes the Network Layer Service Access Point Identifier (NSAPI). The NSAPI can be used as the link identifier since it can uniquely identify the associated Policy Deci sion Point (PDP) contex t (The soft state maintained between the MN, the SGSN a nd the GGSN for guaranteeing a negotiated quality of service in a GPRS network). In the WLAN, the link identifier used by the MN is the Basic Service Set Identification (BSSID) , where the BSSID is the MAC address of the AP. However, several Service Set Identifie rs (SSIDs) can be configured on a single AP. So it is possible that a MN can switch between two SSIDs and change its networklayer configuration while remain ing connected to the same AP. (c) IP address: IP Address Identifiers which may need to be resolved to IP addresses using methods that may be specific to the wireless network. For example, if the old FA (oFA) or MN determines that the IP address of the new FA (nFA) is equal to the oFA's address, then the layer 3 hando ff doesnÂ’t need to be initiated [16]. (d) Subnet prefix: Subsequent to a layer 2 handoff, a MN detects a change in an on-link subnet prefix that woul d require a change in the primary care-of address. For example, a change of AR typically results in a layer 3 handoff [15]. In addition to layer 2 hints, the velocity and trajectory of MN also play important roles in the layer 3 movement detection. MN moving at greater speed generally has a

PAGE 83

72 greater chance of entering a new subnet when performing a layer 2 handoff. If a MN travels in pre-determined trajectories, e. g., commuting between home and office, it may preset certain handoff operati ons upon receiving each specific link identifier to reduce handoff latency and eliminate unnecessary estimation cost. 5.2.2 L3ME Likelihood Function Layer 3 movement estimation can be based on: i i iL w E L (35) where i is an index representing the laye r 2 parameters to be considered, Li, is the likelihood factor, a norm alized difference in characteris tics (based on layer 2 hints) measured between the old AP and the new AP, and wi is a weight used to indicate the impact of Li. The likelihood factors, Li, represent the likelihood that current layer 2 handoff will result layer 3 handoff, and the li kelihood that a layer 3 handoff implies an inter-system handoff. The parameter, E , is introduced to reflect the effect of certain network parameters in layer 3 movement estimation. For exam ple, the subnet prefix is very useful information to decide if MN is moving into a new subnet. If the prefix from the new AP is exactly the same as the prefix from the ol d AP, the MN can draw the conclusion that it is not moving to a new subnet without considering other f actors described by L . In this case, E should set to be zero, so that the value of the estimation function will be zero, to eliminate the effects of L . The parameter, E , is determined by: i iE E , (36)

PAGE 84

73 where iEindicates whether an obvious indica tor exists between APs for the ith layer 2 parameter. Since a small value of the lik elihood equation will result in no layer 3 movement, the goal is to result in zero wh en an obvious layer 3 movement is observed. Thus, otherwise 1, change 3 layer obvious no if , 0iE . (37) Each time a layer 2 trigger is rece ived, a handoff decision is made, and L is calculated for the target AP to estimate the likelihood that it is part of a new subnet. When the probability is larger than a give n (horizontal) threshol d, layer 3 handoff is considered necessary, and c onfiguration of a new IP address along with pre/postregistration protocol will begin. A second threshold, called the vertical thre shold, may be set to evaluate what kind of layer 3 handoff will occur when more than one type of network is involved. When the second threshold is achieved, it indicates th at not only is the MN moving to a new subnet, it is also switching from one networ k to another, e.g., th e MN is moving from WLAN to GPRS. The equivalent Layer 3 movement estimation is shown in Figure 29. The layer 3 movement estimation starts w ith layer 2 hints collection. When any parameter, Ei, is available, a quick judgment will be deployed to decide if an layer 3 handoff will be triggered. Otherwise, further steps will be implemented to decide if the layer 3 handoff is an intra or inter-system handoff based on the value of Equation (35). If no crucial parameters are available, the layer 3 movement estimation will be based only on likelihood factors Li (e.g., the bandwidth and BER measur ed at MN), with the value of

PAGE 85

74 E set to 1. When the value of Equation (35) is larger than the first (horizontal handoff) threshold, layer 3 handoff with the same acce ss network is considered to be necessary, and if the value is also larger than the second (vertical handoff) threshold, it indicates that the MN is moving into a new s ubnet of different access network. Figure 29. L3 movement estimation No Any E i, e.g., IP address indicator, available? Judgment, e.g., same prefix? Calculate the value of (35) Any Li, e.g., bandwidth indicator, available? No Yes Yes Begin with L2 hints collection No L3 handoff > the horizontal threshold? Yes No Inter-system L3 handoff Yes > the vertical theshold? Yes No Intra-system L3 handoff No

PAGE 86

75 5.3 Application in Pre-Registration Handoff The Mobile IP handoff message timing diagram is shown in Figure 30. Figure 30. Mobile IP handoff message timing diagram Mobile IP uses two types of addresse s to manage a MNÂ’s movement: home IP address and CoA. A CoA is used when a MN connects to a foreign network, which is a temporary address obtained by exchanging Rout er Solicitation and Router Advertisement messages with the FA. MN informs its HA by a registration process in which the Registration Request and Registration Repl y are exchanged between the MN and HA, relayed by the FA. The HA then will be able to forward all incomi ng packets to the MN using tunneling. Since Mobile IP was originally designed to ope rate at layer 3, with no consideration of layer 2, it imposes pos sibly unnecessary handoff latencies. In pre-registration handoff, the network a ssists the MN to perform a layer 3 handoff prior to the layer 2 handoff. Accurate a nd rapid layer 3 movement detections are important to a timely trigger of the pre-regist ration process. The pr e-registration handoff message timing diagram with layer 3 movement detecti on is shown in Figure 31. MN nFA HA Router Solicitation Registration Request Router Advertisement Registration reply L2 handoff

PAGE 87

76 Figure 31. Handoff pre-registration w ith layer 3 movement detection In order to expedite the handoff, Router Solicitation and Router Advertisement exchange are performed in advance of a layer 2 handoff, once a layer 2 trigger is received. Layer 3 movement estimation will be based on the Router Advertisement received from FA as well as the layer 2 hi nts collected by MN. If a layer 3 movement is needed, the pre-registration procedure will be triggered immediately. When an intersystem handoff is expected to occur, the co rresponding inter-system pre-registration and pre-authentication will be initiated before layer 2 handoff completes. 5.3.1 Handoff Latency Analysis In Mobile IP handoff, the handoff latency c onsists of the delay for layer 2 handoff, router solicitation and advertisement exchange , and registration proce dure. Thus, the total handoff latency for Mobile IP is calculated as HA FA FA MN HO L MIPT T T HO 2 42 (38) L2 hints collection MNnFA HA Router Solicitation Pre-Registration Request Router Advertisement Pre-Registration reply L2 handoff L3 movement detection

PAGE 88

77 where Ta-b represents the li nk delay between node a and node b , and TL2HO represents the link layer 2 handoff delay. During the whole duration HOMIP, MN will not be able to connect to the new FA, which will ca use significant loss in throughput. In the pre-registration handoff, the time for the pre-registration process is calculated as HA FA FA MN Est L preT T T T 2 4. 3 (39) where TL3Est is the time to estimate if a layer 3 handoff is needed. The total time for pre-registration hando ff is decided by both the layer 2 handoff latency and the pre-registration procedure. If the pre-registration procedure can be completed before the completion of layer 2 ha ndoff, then the layer 2 latency is the only factor for the total handoff latency. Thus, th e total handoff latency for pre-registration is calculated as } 0 ), max{(2 2 HO L pre HO L reg preT T T HO (40) A rapid layer 3 movement detection will ensure that the pre-registration process completes in advance of layer 2 handoff, resulting: HO L reg preT HO2 (41) and a reduction in handoff latency by pre-regist ration with layer 3 m ovement detection. 5.3.2 Signaling Cost Analysis The signaling cost of Mobile IP handoff, CMIP is calculated: HA FA FA MN MIPC C C 2 4 (42) where Ca-b represents the signaling co st parameter between node a and node b .

PAGE 89

78 In addition to the signaling cost required for Mobile IP handoff, an extra signaling cost is needed for layer 3 movement dete ction, including the co st for layer 2 hints collection and computation cost for layer 3 movement estimation. The total cost Cpre is: . 3 . 32 4Est L HA FA FA MN Est L MIP preC C C C C C (43) 5.3.3 Signaling Efficiency Analysis Although link up and down information is suggested by IETF for pre-registration handoff trigger, no accurate layer 3 moveme nt estimation based on layer 2 hints is combined with the procedure. Thus, unneces sary pre-registration procedures may be triggered because not every layer 2 moveme nt will result in a layer 3 movement. Moreover, in a heterogeneous network envi ronment, the registration information for inter-system handoffs should bear additional information for inter-system authentication and registration, compared with intra-system handoffs. Without accurate layer 3 movement detection, a MN w ill not be able to recognize the difference between intra and inter-system handoffs, thus unnecessary information for inter-system has to be sent each time a pre-registration is triggered. The average signaling cost of pre-regist ration handoff with layer 3 movement detection, ' preC is calculated: . 3 3 3 3 '2 ) 2 2 (Est L vertical V L HA FA L FA MN L preC C p C p C p C (44) where pL3 represents the probability that a layer 2 handoff results in a layer 3 handoff and among which pL3V is the probability that the laye r 3 handoff requires a vertical handoff with additional cost Cvertical. In the overlay architecture a gr eater number of APs within a

PAGE 90

79 single subnet means a greater coverage, wh ich makes layer 2 handoffs more likely. pL3V is in inverse proportion of the average number of APs, n , within certain subnet: n pL 3 (45) where is a constant. pL3V depends on the overlay areas of the coordinating networks. For pre-registration handoffs without accurate layer 3 movement detection, pL3= pL3V=1, i.e., the averag e signaling cast is: . 3 ' '2 4Est L vertical HA FA FA MN preC C C C C (46) In the traditional Mobile IP , the signaling cost is: HA FA V L FA MN L MIPC p C p C 3 3 '2 ) 2 2 ( (47) 5.4 Performance Analysis Since the signaling cost depe nds a variety of factors, su ch as network topology and location of entities, the perfor mance analysis is based on the assumptions shown in Table 4. Two sets of signaling cost parameters are gi ven in Table 4. In se t 1, the signaling cost is the same as the summation of the numb er of exchanged signaling messages during signaling procedure. Set 2 denotes the case when the signaling cost between foreign network and home network is high, considering the distan ce between these networks. Table 4. Sets of signaling cost parameters Signaling cost parameters Set 1 Set 2 CMN-FA 1 1 CFA-HA 1 varies Cvertical 1 1

PAGE 91

80 Figure 32 shows the comparison of signaling co st with parameters in the first set. As we can see from the figure, the aver age signaling cost with layer 3 movement detection decreases with the increasing numb er of APs much faster than the average signaling cost of the method without layer 3 movement detection, which implies a decreasing probability of MNÂ’s moving into a new subnet, and it is close to the signaling cost of the traditional Mobile IP handoffs. 1 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 average number of APssignaling cost MIP without L3 movement detection with L3 movement detection Figure 32. Signaling cost comparison (Set I) Figure 33. Signaling cost comparison (Set II)shows the comparison of signaling cost with varying signaling cost from FA to HA with a fixed average number of APs within a subnet. The advantage of layer 3 m ovement detection can be observed. With the increasing of the distance between FA and HA, the cost for pre-registration is relatively high compared to other signaling cost, which re sults in a higher increase of signaling cost

PAGE 92

81 over the pre-registration handoff without laye r 3 movement detection. Thus, accurate layer 3 movement detection can significantly reduce the unnecessary signaling cost of pre-registration. 1 1.5 2 2.5 3 3.5 4 4.5 5 2 4 6 8 10 12 14 16 Distance between FA and HAsignaling cost MIP without L3 movement detection with L3 movement detection Figure 33. Signaling cost comparison (Set II) In Mobile IP networks, layer 2 hints pl ay important role in layer 3 movement detection, and are going to be standardized to enhance interact ions between link and network layers. This will allow a given node to integrate and effici ently manage one or even several network interfaces. Intelligent use of explicit hints from the link-layer would expedite the layer 3 movement detecti on process and reduce handoff latency with reasonable signaling cost. In this chap ter, a likelihood function based method implementing layer 2 hints was proposed to effectively detect layer 3 movement. An application of the lik elihood function in pre-registrati on inter-system handoff protocol

PAGE 93

82 was presented. Performance analysis demons trated significant gains in signaling cost from the proposed method.

PAGE 94

83 CHAPTER 6 SUMMARY AND FUTURE WORK 6.1 Summary This dissertation presents an adaptive and intelligent ne twork approach to vertical handoff. We summarize the research presen ted in this dissertation as follow: Chapter 1 introduced and motivated th e problem. Chapter 2 introduced the background and related research on vertical handoffs. In Chapter 3, we introduced the concept of policy-based framework that translates handoff policies into network configuration to implement vertical handoff protocols in order to produce a satisfactory result for both the user and the network. Various handoff metrics are able to be incorporated for an intelligent and adaptive handoff decision. A case study of UMTS/WLAN interworking is pr ovided using the policy-based vertical handoff architecture. Performance analysis demonstrates that the additional handoff signaling latency is low enough to support user s of various mobility with low handoff failure rate. In Chapter 4, we proposed the Multi-se rvice vertical handoff decision algorithm and analyzed the optimization for vertical handoff decision algorithms in 3G overlay multi-network environment. The optimizations incorporate a network elimination feature to reduce the delay and processing required in the evaluation of the cost function, and a multi-network optimization is introduced to improve throughput for mobile terminals with multiple active sessions. A performance an alysis demonstrated significant gains in quality of service and a more efficient use of resources from the proposed optimizations.

PAGE 95

84 In Chapter 5, a new likelihood function is introduced for implementing low-cost layer 3 movement detection for Mobile IP handoffs. An application of the likelihood function in pre-registration inter-system ha ndoff protocol was pres ented. Performance analysis demonstrated signi ficant gains in signaling cost from the proposed method. 6.2 Future Work In this section, we point out the future research directions. Although the inter-opera tion issues of 3G wireless ne twork and WLAN have been studied in the past several years, there is not a general architectur e for the envisioned future network. The next generation wireless network will be an integration of a larger variety of wireless and mobile networks, which might even include ad hoc networks. Factors such as scalability, reliability, seamle ss mobility support, service level agreement, economics, compatibility and inter-operability need to be considered when we design the architecture for next genera tion wireless network. I plan to propose a novel network architecture for global ubiqu itous communications that considers the metrics mentioned above. Wireless mesh networks (WMN) is a pr omising wireless technology for various applications, e.g., home networking, campus ne tworking, enterprise networking, and hot spot networking. They provide network acce ss for both mesh and conventional clients. The integration of WMN with other networks such as Internet, cellular, IEEE 802.n, and sensor networks, can be accomplished thr ough the gateway and bridging functions in mesh routers. Cross-layer design between la yers 2 and 3 is one of the open research issues for WMN. There has been a clear cu t between layer 2 and layer 3. By adopting performance metrics from layer 2, I plan to build a more intelligent routing protocol.

PAGE 96

85 Security is always a critical issue in wireless network development. Security technologies and procedures al ready in place are very like ly to be insufficient when businesses open up to remote access and mob ile connectivity. Secu rity in terms of authentication and authorization might not be a problem for any single network. However, in the future, centralized scheme is not efficient or scalable in the heterogeneous network environment. Security thus becomes a very important metrics for vertical handoff. A novel practic al secure handoff scheme is needed to fight against all kinds of possible malicious attacks and intrusions.

PAGE 97

85 LIST OF REFERENCES [1] I. F. Akyildiz, J. McNair, J.S.M. Ho, H. Uzunalioglu, and W. Wang, “Mobility Management in Next Generation Wireless Systems,” Proceedings of the IEEE , vol. 87, no. 8, Aug. 1999, pp. 1347–84. [2] W. W. Lu, “Fourth-Generation Mob ile Initiatives and Technologies,” IEEE Communications Magazine , vol. 40, no. 3, Mar. 2002, pp. 104–05. [3] R. Berezdivin, R. Breinig and R. Topp, “Next-Generation Wireless Communications Concepts and Technologies,” IEEE Communications Magazine , vol. 40, no. 3, Mar. 2002, pp. 108–16. [4] K. Pahlavan, P. Krishnamurthy, A. Hatami, M. Ylianttila, J.P. Makela, R. Pichna, J.Vallstron, “Handoff in Hybrid Mobile Data Networks,” IEEE Personal Communications Magazine , vol. 7, no. 2, Apr. 2000, pp. 34–47. [5] D. B. Johnson, C. Perkins and J. Arkko, “Mobility Support in IPv6,” IETF, May 2002, http://www.ietf.org/ [6] A. T. Campbell, J. Gomez, S. Kim, A. G. Valko, W. Chieh-Yih, Z. R. Turanyi, “Design, Implementation, and Evaluation of Cellular IP,” IEEE Personal Communications Magazine , vol. 7, no. 4, Aug. 2000, pp. 42–49. [7] R. Ramjee, K. Varadhan, L. Salgarelli, S. R. Thuel, S.-Y. Wang, T. La Porta, “HAWAII: A Domain-Based Approach for Supporting Mobility in Wide-Area Wireless Networks,” IEEE/ACM Transactions on Networking , vol. 10, no. 3, June 2002, pp. 396–410. [8] J. Makela, M. Ylianttila, and K. Pahl avan, “Handoff Decisi on in Multi-service Networks,” IEEE Personal, Indoor and Mobile Radio Communications (PIMRC) , London, vol. 1, Sept. 2000, pp. 655–59. [9] H. J. Wang, R. H. Katz and J. Gi ese, “Policy-enabled Handoffs Across Heterogeneous Wireless Networks,” IEEE Mobile Computing Systems and Applications (WMCSA '99) , New Orleans, LA, Feb. 1999, pp. 51–60. [10] F. Zhu and J. McNair, “Optimizations for Vertical Handoff Decision Algorithms,” IEEE Wireless Communications and Networking Conference (WCNC’04) , Atlanta, GA, March. 2004, pp. 867-872.

PAGE 98

87 [11] Context Transfer, Handoff Ca ndidate Discovery, and Dormant Mode Host Alerting (seamoby) Working Group, IETF, Apr. 2003, http://www.ietf.org/ [12] J. McNair, I. F. Akyildiz and M. Be nder “An Intersystem Handoff Technique for the IMT-2000 Systems,” IEEE INFOCOM 2000 , Tel-Aviv, Israel, Mar. 2000, vol. 1, pp. 208–16. [13] A. Yegin, E. Njedjou, S. Veerepalli, T. Noel, “Link-layer Hints for Detecting Network Attachments,” IETF draft draft-yegin-dna-12-hints-00.txt , October 2003. [14] C. Perkins "IP Mobility Support for IPv4", IETF RFC3344 , August 2002. [15] C. Perkins and J. Arko, “Mobility Support in IPv6,” IETF draft-ietf-mobileip-ipv624.txt , June 2003. [16] K. El Malki, “Low latency Handoff in Mobile IPv4,” IETF draft raft-ietf-mobileiplowlatency-handovers-v4-09.txt , June 2004. [17] K. Pahlavan, P. Krishnamurthy, A. Hatami, M. Yliantillai, I.P. Makela, R. Pichna, J. Vallstron, “Handoff in Hybr id Mobile Data Networks,” IEEE Personal Communications Magazine , vol. 7, no. 2, pp. 34-47, April 2000. [18] M. Stemm and R. Katz, “Vertical Hando ffs in Wireless Overlay Networks,” ACM Mobil Networking (MONET), Special Issu e on Mobile Networking in the Internet , Summer 1998, pp. 335-350. [19] R. Berezdivin, R. Breinig and R. top, “Next-generation wireless communications concepts and technologies,” IEEE Communication Magazine , vol. 40, no. 3, March 2002, pp. 104-105. [20] S. D. Park, E. Njedjou, and N. Monta vont , “L2 Triggers Optimized Mobile IPv6 Vertical Handoff: The 802.11/GPRS Exam ple,” IETF draft draft-daniel-mip6optimized-vertical-handover00.txt, January 31, 2004. [21] J. McNair, F. Zhu, “Migration toward 4G wireless communications Vertical handoffs in fourth-generation multinetwork environments,” IEEE Wireless Communications , vol. 11, issue 3, pp. 8-15, June 2004. [22] F. Zhu and J. McNair, "Cross-Layer Design for Mobile IP Handoff Hints," IEEE Vehicular Technology Conference (VTC) Spring 2005 , Stockholm, Sweden, June 2005, pp. 2255 2259. [23] F. Zhu, J. McNair and S. Wang, “On Load Balancing Issues for Vertical Handoffs in Next Generation Multi-Tier Networks,” submitted. [24] F. Zhu and J. McNair, “Increasing throughput in wireless overlay networks using smart vertical handoffs,” submitted.

PAGE 99

88 [25] K. Pahlavan, P. Krishnamurthy, Principl es of Wireless Networks: A Unified Approach, Prentice-Hall, Inc, Upper Saddle River, NJ, 2002. [26] Y.-B. Lin, I. Chlamtac, Wireless and M obile Network Architectures, John Wiley and Sons, Inc, New York, NY, 2000. [27] R. Yavatkar, D. Pendarakis, and Gu erin, “A Framework for Policy-Based Admission Control,” IETF Request for Comments (RFC) 2753 , January 2000. [28] T.S. Rappaport, Wireless Communicati ons: Principles a nd Practice, 2nd Ed, Prentice-Hall, Inc, Upper Saddle River, NJ, 1996. [29] D.B. Johnson and D.A. Maltz, “Dynami c Source Routing in Ad Hoc Wireless Networks,” Mobile Computing , pp. 153-181, 1996. [30] D. Bertsekas and R. Gallager, Data Netw orks, 2nd Ed, Prentice-Hall, Inc, Upper Saddle River, NJ, 1992. [31] J.M. Holtzman and A. Sampath, “Ada ptive Averaging Methodology for Handoffs in Cellular Systems,” IEEE Transactions on Vechcular Technology , vol. 44, no. 1, pp. 59-66, Feburary 1995. [32] ETSI, “Requirements and Architectures for Interworking between HIPERLAN/3 and 3rd Generation Cellular Systems,” Tech. rep. ETSI TR 101 957, Aug. 2001. [33] A. K. Salkintzis, C. Fors, and R. Paz hyannur, “WLAN-GPRS In tegration for Nextgeneration Mobile Data Networks,” IEEE Wireless Communications , vol. 9, no. 5, Oct. 2002. pp. 112–24. [34] M. Buddhikot, G. Chandranmenon, S. Han, Y.W. Lee, S. Miller, L. Salgarelli, “Integration of 802.11 and Third Gene ration Wireless Data Networks,” IEEE INFOCOM ’03 , San Francisco, CA, 30 March-3 April 2003, vol 1, pp. 503 – 512. [35] 3GPP Ts 23.234 V6.0.0, “3GPP system to Wireless Local Area Network (WLAN) interworking; System descripti on (Release 6),” March, 2004. [36] A. K. Salkintzis, “Interworking Tech niques and Architectures for WLAN/3G integration Towards 4G Mobile Data Networks,” IEEE Wireless Communications , vol. 11, no. 3, June 2004. [37] X. Yang, J. Bigham and L. Cuthbert, “R esource Management for Service Providers in Heterogeneous Wireless Networks,” IEEE Wireless Communications and Networking Conference (WCNC’05) , New Orleans, LA, 13-17 March, 2005, pp. 1305-1310. [38] C. Liu, C. Zhou, “An improved interw orking architecture fo r UMTS-WLAN tight coupling,” IEEE Wireless Communications and Networking Conference (WCNC’05) , New Orleans, LA, 2005, vol. 3, 13-17 March, 2005, pp.1690 – 1695.

PAGE 100

89 [39] H. Thomas, H. Gilbert and G. Mazziotto, “Influence of the moving of the mobile stations on the performance of a radio mobile cel lular network,” the 3rd Nordic Seminar , Copenhagen, Denmark, 12–15 September, 1988, pp.1-9. [40] F. G. Marquez, M. G. Rodriguez, T. R. Valladares, T. de Miguel, L. A. Galindo, L.A, “Interworking of IP multimedia core networks between 3GPP and WLAN,” IEEE Wireless Communications, vol. 12, no. 3, June 2005, pp. 58 – 65. [41] W. Wu, N. Banerjee, K. Ba su, S. K. Das, “SIP-based vertical handoff between WWANs and WLANs,” IEEE Wireless Commu nications, Vol. 12(3), June 2005, pp. 66 – 72. [42] H. Badis, K. Al Agha, “Fast and efficien t vertical handoffs in wireless overlay networks,” IEEE Personal, Indoor and Mobile Radio Communications (PIMRC) , vol. 3, pp. 1968 1972, Barcelona, Spain, 5-8 Sept. 2004. [43] J. Tourrilhes, C. Carter, “P-Handoff: a protocol for fine gr ained peer-to-peer vertical handoff,” IEEE Personal, Indoor and Mob ile Radio Communications (PIMRC) , vol. 2, pp. 966 – 971, Lisboa, Portugal, 15-18 Sept. 2002. [44] M. Bernaschi, F. Cacace, G. Iannello , “Vertical handoff performance in heterogeneous networks,” IEEE International Conference on Parallel Processing (ICPP) Workshops , pp. 100 – 107, Montreal, Quebec, Canada, 15-18 August, 2004. [45] Q. Zhang, C. Guo, Z. Guo, W. Zhu, “E fficient mobility management for vertical handoff between WWAN and WLAN,” IEEE Communications Magazine , vol. 41. no. 11, pp. 102 – 108, Nov. 2003. [46] H. Huang, J. Cai, “Improving TCP perf ormance during soft vertical handoff,” IEEE Advanced Information Networ king and Applications (AINA) , vol. 2, pp. 329 – 332, Taiwan, 28-30 March 2005. [47] S. E. Kim, J. A. Copeland, “TCP for seamless vertical handoff in hybrid mobile data networks,” IEEE Global Telecommunicatio ns Conference (GLOBECOM) , San Francisco, USA, vol. 2, pp. 661 – 665, 1-5 Dec. 2003. [48] M. Salamah, F. Tansu, N. Khalil, “Buffering requirements for lossless vertical handoffs in wireless overlay networks,” IEEE Vehicular Technology Conference , vol. 3, pp. 1984 – 1987, Jeju, Korea, 22-25 April 2003. [49] M. Ylianttila, M. Pande, J. Makela, P. Mahonen, “Optimization scheme for mobile users performing vertical handoffs between IEEE 802.11 and GPRS/EDGE networks,” IEEE Global Telecommunications Conference (GLOBECOM) , vol. 6, pp. 3439 – 3443, San Antonio, Texas, 25-29 Nov. 2001.

PAGE 101

90 [50] M. Ylianttila, J. Makela, P. Mahonen, “S upporting resource allocation with vertical handoffs in multiple radio network environment,” IEEE Personal, Indoor and Mobile Radio Communications (PIMRC) , Lisboa, Portugal, vol. 1, pp. 64 – 68, 1518 Sept. 2002. [51] S. Sharma, I. Baek, Y. Dodia, T. Chiu eh, “Omnicon:a mobile ip-based vertical handoff system for wireless LAN and GPRS links,” IEEE International Conference on Parallel Processing (ICPP) Workshops , Montreal, Quebec, Canada, pp. 330 – 337, 2004. [52] M. Nam, N. Choi, Y. Seok, Y. Choi, “W ISE: energy-efficient interface selection on vertical handoff between 3G networks and WLANs,” IEEE Personal, Indoor and Mobile Radio Communications (PIMRC) , Barcelona, Spain, vol. 1, pp. 692 – 698, 5-8 Sept. 2004. [53] S. Jung, D. Cho, O. Song, “QoS base d vertical handoff method between UMTS systems and wireless LAN networks,” IEEE Vehicular Technology Conference (VTC) , vol. 6, pp. 4451 – 4455, Los Angeles, CA, 26-29 Sept. 2004. [54] A. Mehbodniya, J. Chitizadeh, “An intellig ent vertical handoff algorithm for next generation wireless networks,” IEEE Wireless and Optic al Communications Networks (WOCN) , pp. 244 – 249, Dubai, United Arab Emirates, March 6 8, 2005. [55] A. Hasswa, N. Nasser, H. Hassanein, “Gen eric vertical han doff decision function for heterogeneous wireless networks,” IEEE Wireless and Optical Communications Networks (WOCN) , pp. 239 – 243, Dubai, United Arab Emirates, March 6 8, 2005. [56] W.-Y. Chen, Y.-Y. Shu, “Ac tive application oriented vertical handoff in nextgeneration wireless networks,” IEEE Wireless Communica tions and Networking Conference (WCNC) , vol. 3, pp. 1383 1388, New Orleans, LA, 13-17 March, 2005. [57] W.-T. Chen, J.-C. Liu, H.-K. Huang, “An adaptive scheme for vertical handoff in wireless overlay networks,” IEEE Parallel and Distributed Systems , pp. 541 – 548, Newport Beach, CA, 7-9 July 2004. [58] J. Yoon, M. Liu, B. Noble, “Ra ndom waypoint considered harmful,” IEEE INFOCOM , vol. 2, pp. 1312 – 1321, San Franci sco, CA, 30 March-3 April, 2003.

PAGE 102

91 BIOGRAPHICAL SKETCH Fang Zhu received the B.S. degree fro m Beijing University of Posts and Telecommunications, Beijing, China, in 1999 and the M.S. degree from Colorado State University, Fort Collins, Colorado, in 2002, and worked as an engineer with China Telecom, Beijing, China, from 1999 to 2000. She is currently pursuing the Ph.D. degree in the Department of Electr ical and Computer Engineeri ng, University of Florida, Gainesville, Florida. She is a research a ssistant in the Wireless and Mobile Systems Laboratory. Her research interests include mobility management, quality of service and resource management for the ne xt generation wireless systems.