International Journal of Wireless Information Networks, Vol. 5 , No. 1 , 1998
A Survey of Handover Techniques for Wireless ATM Networks C.-K. Toh 1,3 and Bora Akyol 2
This paper provides a review and comprehensive comparison of existing handover schemes proposed for wireless ATM networks. Existing schemes include those proposed by NEC, Olivetti, Bell Labs, Cambridge, Stanford, CMU, Berkeley, Michigan, VTT, and Columbia. We highlight the methodologies employed by the various schemes and reveal their differences. We discuss issues related to the effect of ATM switch architectures on mobility implementation and on future handover paradigms. KEY WORDS: Wireless ATM; mobile handover protocols.
1. INTRODUCTION
acteristics and hence the user-perceived quality of service (QoS). Existing handover schemes can be classi® ed in accordance with the path rebuild strategy employed. They include the following handover schemes: (a) connection extension [31], (b) full reestablishment [16], (c) incremental / partial re-restablishment [16], (d) multicast join [16], and (e) multicast group [16, 8, 7]. In the connection extension scheme, a mobile connection is extended from the previous base station (BS) to the new BS. In the full reestablishment scheme, a new connection has to be established from the new BS to the remote end host. In the incremental reestablishment approach, a partial connection is established from the new BS to a node residing on the connection path prior to handover. In the multicast-join scheme, a partial path is established from an intermediate switch to the new BS, using the feature of ATM multicast. This is quite similar to the problem of handling dynamic host membership in multicasting [30]. Finally, in the m ulticast-group scheme, multiple connections are always established to multiple wireless cells surrounding the wireless cell where the MH is currently residing. Hence, this scheme is sometimes referred to as the `torch’ or `footprint’ approach. Instead of reserving communication resources to all neighboring cells, an alternative approach is to allow only
Several local- and wide-area wireless ATM architectures have evolved over the past few years. The reasons for their evolution are diverse. Some architectures aim at providing seamless integration with broadband ISDN, while others aim to support mobile multimedia. Nonetheless, the fundamental aim of wireless ATM is to provide mobility support for mobile ATM users as they roam from one location to another. The term `mobility support’ in a wireless ATM environment refers to the ability to establish and tear down mobile connections and to provide a mechanism for mobile handovers, location tracking, and queries. Several handover mechanisms have been proposed and studied, and some have been implemented. Depending on the handover strategy employed, the amount of traf® c disruption, cell loss, cell delay variation, and signaling traf® c generated can vary considerably. Since wireless ATM aims to provide multimedia services to mobile hosts (MHs), the performance of the handover protocol thus has a signi® cant impact on the traf® c char1 Hughes
Research Laboratories, Malibu, California 90265 ; e-mail:
[email protected] om [formerly with Cambridge University]. 2 BBN Systems and Technologies, Cambridge, Massachusetts 02140 ; e-mail:
[email protected] [formerly with Stanford University]. 3 To whom correspondence should be addressed.
43 1068-9605 / 98 / 0100-0043$15.00
/0 Ó
1998 Plenum Publishing Corporation
44 one such connection to be active while others remain passive. Handover strategies may also be categorized according to the direction of handover invocation. The forward handover scheme usually refers to the case where a handover is initiated by sending a handover message from the MH to the new BS. On the other hand, the backward handover scheme initiates a handover via the previous BS. Coincidentally, in the connected (i.e., where a MH has good radio link connectivity with its current BS) handover case, handover is initiated via the old BS. However, in the disconnected handover case, handovers are initiated via the new BS as a result of loss of radio link connectivity with old BS. Radio hint derived from received beacon signal strength may also be used to trigger handovers. Hence, handovers can also be classi® ed into those with and without radio hint. Both forward and backward handovers belong to the group of horizontal handovers, where the cellular network uses only a single layer of microcells. In a hierarchical multilayered cellular network, vertical handovers occur [28]. MHs roaming in picocell regions can be gradually handed off to microcell and macrocell regions, hence resulting in vertical handovers. Picocells may be found inside buildings, while microcells are deployed to cover roads and towns. Bigger wireless cells are provided by satellites in order to attain national coverage. Vertical handovers ensure universal roaming between different wireless networks operated by different network service providers. Several names are also used to characterize handovers, based on the buffering strategy and new path establishment strategy employed. Soft handovers involve two active paths where a MH can communicate with the old and new BSs. Such a scheme uses the CDMA media access technique to remain simultaneously in connectivity with both the old and new BSs. Hard handovers, however, require the MH to switch to a new frequency channel so that it is connected to the new BS. Hence, a short traf® c ¯ ow disruption is expected as a result of the frequency switching process. This is found commonly in analog cellular systems, such as AMPS (advanced mobile phone systems). Finally, in a seamless handover case, a new path is established in parallel while the old path is still in use. The MH concerned has to operate on two carriers in time division. The connection switching operation is initiated when proper cell stream synchronization has been performed. In mobile-controlled handovers, the MH is responsible for initiating a handover. It does this by evaluating the signal strength and traf® c load conditions and
Toh and Akyol detecting the presence of neighboring BSs. When a MH decides to initiate a handover, it sends an explicit message to a mobility management node residing in the network. In a network-controlled handover, however, connection rerouting is performed by the network. The network collects statistics related to signal strength, traf® c load, and other information to decide when to initiate a handover and which new BS is the target BS. In contrast, in a mobile-assisted handover, the MH monitors the signal strength and the presence of neighboring BSs and conveys this information to the network controller. The network controller then uses such information to make handover decisions. There are several factors to be considered in the design of a handover protocol [33]. These factors are related to handover performance issues and constraints imposed by the operating environment. There is a need for fast handovers due to the time constraint imposed by the MH’ s migration speed and the wireless cell overlapping width. An ef® cient handover scheme is desirable so that the resultant signaling traf® c is a small proportion compared to user data traf® c. Finally, a continuous handover protocol is necessary so that there will be minimum interruptions to traf® c ¯ ows during handovers. Continuity, in the context of ATM, implies that ATM cells should continue to arrive in sequence. The handover protocol must also be scalable to the number of mobile connections and MHs. It must also be mutually exclusive such that source MH handovers are independent of the destination MH (for a MH-to-MH connection). Lastly, it is desirable for the protocol to support handovers of single and multiple VCs, for both unicast and multicast mobile connections. While quality of service (QoS) parameters for ATM networks are currently well de® ned by the ATM Forum and ITU, QoS for mobile ATM networks are still in its early stage of developments. Traditionally, handover blocking probability is regarded as an important performance measure of a mobile network system. Although this may seem suf® cient for cellular voice systems, highspeed wireless networks aiming to support multimedia traf® c will demand more stringent and diverse QoS requirements. Instead of allowing a call to be blocked or dropped during a handover, some forms of application and network adaptation techniques have been introduced in [37]. These techniques require applications to operate over a permissible range of QoS variations, trading off QoS for lower handover blocking. A summary of handover classi® cation is illustrated by Fig. 1. In this paper, we examine the handover strate-
Handover Techniques for Wireless ATM Networks
45
Fig. 1 . Classi® cation of mobile handovers in wireless ATM networks.
gies employed by various organizations. Section 2 discusses the handover methodologies proposed by CMU, Stanford, Olivetti, NEC, VTT, Colum bia, Bell Labs (Holmdel), Bell Labs (Murray Hill), Michigan, Berkeley, and Cambridge. Section 3 presents a detailed comparison of these schemes; a discussion on a future handover paradigm is provided in Section 4.
2. REVIEW OF HANDOVER STRATEGIES 2.1. CMU Handover Scheme Proposed in 1995 [17], Kam at CMU introduced a handover scheme which employs ATM multicasting. Prior to a handover, branch connections (equivalent to connection segments) are established in advance from one or more intermediate or `crossover’ nodes in the connection path to the wireless cells surrounding the cell where the MH is currently residing, as shown in Fig. 2a. However, there are no procedures for identifying these crossover nodes. The established branch connections consume no resources at this stage. During a handover, resources are only allocated to the ® nal branch connection that serves the MH. After a handoff, new branch connections are again established in the new neighborhood and all the old branch connections are then torn down. This is illustrated in Fig. 2c, d.
Fig. 2 . CMU handover scheme based on cell neighborhood.
This handover strategy is similar to the `footprint’ concept proposed in ref. 8. Since all neighborin g wireless cells are activated in preparation for handoff, the handoff process can be very fast. In addition, since communication resources are only allocated on con® rmation of handoff, there is no wastage of resources during handoffs. However, because the current BS has to initiate a multicast to all its neighboring wireless cells to initiate branch connection setup and teardown, the signaling traf® c generated can be substantial. In addition to a handover strategy, Kam has also proposed a reservation control protocol that works in harmony with the handover protocol to resolve QoS consistency issues before and after a handover. As shown in Fig. 3, a MH initiates a handover by sending a GREET message to the current BS. The new BS then sends a GRAFT message toward the crossover node to allocate communication resources. The crossover node then performs connection rerouting on reception of the GRAFT message. This message is also forwarded to the remote
Fig. 3 . Reservation control during handovers.
46
Toh and Akyol
end host and QoS adaptation will occur over the remaining links and the end host. The adaptation process may result in either QoS upgrade or downgrade. QoS refresh is performed on a node-by-node basis over the links affected. When the adaptation process is completed, an ACCEPT message is then returned to the new BS. The new BS then inform s the MH about the completion of adaptation via an UPDATE message. The reservation control protocol, however, is only applicable for MH to ® xed host connections. Consequently, further work is necessary to extend this reservation control concept for MH-to-MH connections. 2.2. Stanford Handover Scheme Proposed in 1995 [3, 2], Akyol and Cox at Stanford introduced a wireless ATM network architecture employing the `zone’ concept. As shown in Fig. 4, within a zone, several radio port controllers are connected to a wireless-to-ATM network interface. Hence, handovers are now separated into intra- and interzones. Handovers are mobile-initiate d and can be invoked via the new or old BS. If invoked via the old BS, 10 handover signaling messages are required. The handover signaling involves the following network components:
· MH to old zone manager
· MH to new zone manager
· Old and new zone manager
· Old zone manager and end point
· New zone manager and end point However, if the handover is initiated via the new BS, a total of 15 handover signaling messages are required. The rerouting strategy employed is based on Nearest Common Node Rerouting (NCNR) concept, where the nearest common intermediate node is used to reroute mobile ATM connections, hence attaining route
optim ization. The nearest common node is determined by means of network topology or in terms of number of hops to the end point. However, no speci® c procedures have been proposed for the discovery of the nearest common node in wireless ATM networks with different network topologies. The proposed NCNR scheme takes into consideration the types of traf® c transported. For time-sensitive traf® c, if two zone managers are not connected directly, a handover start message is sent by the old zone manager. This message transits through the old path, in a hop-byhop manner. The ® rst node encountered that could reach both old and new zone managers is the nearest common node. The NCNR scheme is similar to the backward tracking crosspoint location process proposed in ref. 16. It allows maximal reuse of existing user connection, and the a priori probability of not meeting the QoS of the connection is minimized since there is a lower number of nodes involved. However, the disadvantage of this scheme is that the ® rst common point located is not necessarily the best common node, since it may not result in the shortest convergence path (i.e., the path between the common node and the new BS) or a resultant path that ful® ls the end-to-end delay requirements among other possible CXs. The NCNR scheme for throughput-de pendent traf® c has similar handover procedures as the timesensitive scheme except that speci® c cell forwarding and buffering schemes are proposed to ensure that there is no cell loss. In this proposal, multiple connections are rerouted using a virtual path concept. NCNR utilizes the point-to-m ultipoint (where two active transmission paths exist during a handover) to provide robustness to the `ping-pong’ effect of MHs. The protocol procedures for interzone handovers are illustrated by Fig. 5. For intrazone handovers, connection rerouting is performed at the wireless-to-ATM network interface unit by updating the ATM VC translation tables within the zone. Therefore, intrazone handovers do not require wide-area network signaling. An alternative proposal using migratory signaling instead of overlay signaling which is not covered in this paper can be found in refs. 4 and 6. 2.3. Olivetti Handover Scheme
Fig. 4 . The zone architecture of a wireless ATM network.
De® ned in 1996 [15], the Olivetti wireless ATM architecture exploits the advantage of locality by dividing a wireless network into domains. The architectural components consist of three switching elements: (a) mo-
Handover Techniques for Wireless ATM Networks
47
Fig. 5 . Stanford handover protocol based on zone grouping.
bile switching point (MSP), (b) base station switching point (BSP), and (c) ® xed switching point (FSP). A FSP is allocated to a mobile via the DLS (domain location server). The MSP maintains information related to link qualities and the occupancy status of neighboring BSs. This therefore results in substantial signaling traf® c between BSs, which is undesirable. The BSP, on the other hand, provides virtual path management between BSs and MHs. Finally, the FSP is the point residing in the ATM network where all virtual circuits to the mobile are routed. As shown in Fig. 6, in the `connected’ handover mode, a minimum of 18 handover messages are required.
These signaling involve the MSP, old BSP, new BSP, and FSP. If a mobile has lost its radio connectivity with its current BS, a `disconnected’ handover procedure is invoked. This procedure again involves signaling among the MSP, old BSP, new BSP, and FSP. A total of 11 handover signaling messages are incurred. Since all connections are routed via the FSP, routes so formed after handovers are not optim ized. In addition, the FSP will have to be migrated when handovers occur between domains. Meta-signaling is used to establish signaling channels between the MH and BSs. The upstream and downstream disruption times cannot be evaluated due to lack of detailed inform ation on the protocol control and data ¯ ows. Another shortcoming is that
Fig. 6 . The Olivetti handover protocol based on FSP.
48
Toh and Akyol
a FSP is allocated by the domain location server. However, when a MH migrates, the FSP may no longer be in the optimum location, resulting in inef® cient routes. 2.4. NEC Handover Scheme As discussed in refs. 26 and 27, the NEC handover strategy is based on the concept of a handover switch (HOS). Mechanisms are provided in the handover protocol to support cell resequencing, which is achieved by transferring wireless data-link state information from one BS to another. End-to-end VCs are separated into two segments, (a) ® xed and (b) dynam ic segments. Sequential cell delivery is guaranteed via the usage of special inband OA&M (operation, administration, and maintenance) cells to explicitly mark the end of a cell stream during handoffs. The NEC wireless ATM network architecture [29] is organized in a hierarchical fashion and it consists of radio BSs, BS multiplexor, mobile-enabled ATM switches, and normal ATM switches. The ITU Q.2931 signaling protocol is employed for signaling between switches, while an enhanced Q.2931+ UNI protocol is used among the mobile device, BS, and mobile-enabled ATM switch. Associated with each mobile-enabled ATM switch is a mobile service unit which contains the necessary switch control software to perform connection, handover, and location management functions. Exploitation of locality is achieved by connecting multiple BSs to BS multiplexors which are subsequently connected to ATM switches. The procedures for intraswitch handovers are similar to intrazone handovers proposed by Akyol and Cox. During interswitch handovers, a HOS is selected to reroute mobile connections. As shown in Fig. 7, handovers are mobile-initiated and a
total of 14 signaling messages are required. The handover process consists of three phases: (a) new connection segment setup, (b) route change, and (c) old connection segment teardown. Procedures to locate a handover switch have neither been speci® ed nor proposed. A distinctive feature of the handover strategy is the transfer of data-link state inform ation between BSs, allowing the MAC layer protocol to better schedule wireless resources. The protocol procedures for interswitch handovers is illustrated by Fig. 7 and it reveals that upstream and downstream disruption times are different. 2.5. VTT Handover Scheme Proposed in 1996 [11], the VTT handover scheme emphasizes no cell loss (be it a result of cell duplication or misordering) during handovers. Cell streams are split into segments, with marker cells inserted at speci® c points in the cell stream to indicate the start and end of a connection segment. Therefore, upstream and downstream cell markers are required. Handovers can be initiated in the forward or backward direction. Connection re-outing for uplink and downlink traf® c is treated differently during handovers. A total of 15 signaling messages are incurred for both `backward’ and `forward’ handovers. Cell forwarding from the old BS to the new BS is also employed to ensure cell sequencing is achieved. As shown in Fig. 8, uplink and downlink connection re-routing occur at different times. A MH initiates a handover by sending a message to the old BS. This handover REQUEST message is then propagated to the handover switch. The handover switch performs the downlink rerouting, and downlink data are then buffered at the new BS. A handover RESPONSE
Fig. 7 . The NEC handover protocol based on handover switch.
Handover Techniques for Wireless ATM Networks
49
Fig. 8 . The VTT handover protocol initiated at the old BS.
message is then propagated back to the MH via the old BS. The old BS is also informed about the completion of downlink rerouting. This completes the downlink handover process. The MH then informs the old BS about the last upstream cell transmitted. Thereafter, upstream data are buffered at the MH. The old BS sends a message to the handover switch to invoke the uplink rerouting process. The MH is subsequently informed when this rerouting process is completed. From the protocol outline shown in Fig. 8, the downlink data are ® rst rerouted from the old BS to the new BS. Three sets of buffers are required at the BSs. They are the upstream, downstream, and forwarding cell buffers. Cell forwarding has been employed in order to ¯ ush remaining cells in the old BS that have not been delivered to the MH as a result of the MH releasing its old radio link. The protocol is fairly complicated, as downlink and uplink handovers are performed independently. It is unclear whether uplink rerouting will continue to occur when the downlink rerouting process fails. In the without-hint or forward handover protocol, the uplink and downlink reroutings are performed together, since cell sequencing can no longer be guaranteed when an abrupt loss of radio connectivity exists. Hence, this implies that the switch must support different types of rerouting during handover of upstream and downstream traf® c. The complexity and buffering
requirements can be high and severe bottlenecks may be present as the number of mobile connections to be handed over increases. The VTT proposal has only considered intraswitch handovers. Interswitch handovers, therefore, remain to be investigated. 2.6. Columbia Handover Scheme The Columbia handover scheme is based on the concept of the virtual connection tree (VCT) [21] scheme. In this scheme, a hierarchical network topology approach is adopted. BSs are connected to ATM switches and these switches are connected to common switches higher up in the network hierarchy. A root switch is a switch that oversees a group of BSs, also referred to as the leaves of the multicast tree. The VCT concept is similar to the `footprint’ approach where multiple connections are always preestablished to all the neighboring wireless cells. However, only one such connection will serve the MH after a handover. Hence, establishing multiple connections is inef® cient, as only one connection will eventually be utilized. The VCT scheme requires no call processer intervention in the processing of handovers occurring within a connection tree. When a MH admitted to a VCT wishes to hand over to another BS in the same VCT, it simply
50
Toh and Akyol
transmits ATM cells containing the connection number assigned for use between itself and the new BS. By using the preestablished path between the new BS and the root switch, upstream cells now ¯ ow to the root switch. These cells then update the virtual channel translator (VCN) residing in the root switch, which then ensures that downstream cells are properly rerouted. Hence, if a mobile connection is purely downstream, the root switch will not be updated about the handover. When a MH migrates outside the coverage of the current VCT, intertree handovers occur. When this happens, the network call processor is invoked to transfer the call to the appropriate VCT. Because handovers are handled solely at the root switch, potential bottlenecks can exist at the root switch, especially when the number of mobile connections and MHs increases. Compared to Akyol and Cox’ s scheme, the root switch is not necessarily the nearest common switch. Compared to the Cambridge scheme [33], the root switch is ® xed for a speci® c VCT. Because no speci® c handover protocol has been proposed, we cannot evaluate the performance of the protocol further.
2.7. Bell Labs (Holmdel) Handover Scheme A distributed control strategy for wireless ATM networks has been proposed in ref. 19. The cluster-based network architecture de® nes multiple clusters within a network, with each network comprised of call, home, and visitor’ s location servers. Within each ATM switch, a channel server is present to manage radio channels and VC / VP translation tables and to support handovers. Mobile location, call control, and connection control functions are separated and parallel connection setup is proposed instead of the traditional hop-by-hop call setup approach. The handover scheme employed in the
distributed wireless ATM architecture is also based on establishing new partial connections. Handovers are performed by establishing partial connections from the new BS to a handover switch. A brief outline of the handover process is presented, but there is no detailed description of the handover protocol and procedures for discovering the handover switch. This therefore inhibits us from evaluating the protocol further.
2.8. Bell Labs (Murray Hill) Handover Scheme The handover scheme proposed by Mishra and Srivastava [32] of Bell Labs is based on the concept of `connection extension,’ `route rebuild,’ and `loop removal’ . The handover scheme was developed as part of the SWAN wireless ATM project at Bell Labs [22, 23]. The main argument behind connection extension is that local ATM paths are cheap, while global ATM paths maintained by network operators are expensive. Hence, mobile ATM connections are extended from the old BS to the new BS when a MH migrates within a local ATM network domain. When a MH migrates between network domains, a route rebuild process is initiated in order to optim ize the route. Route rebuild is achieved by establishing a new partial path from the new BS to a common switch in the network. The disadvantage of the connection extension scheme is that the route path is lengthened (meaning end-to-end delay is increased) and routes so formed are not optim ized (meaning inef® cient usage of resources). In addition, if MHs move in a zigzag fashion, route loops can occur and route loop removal procedures are required. As shown in Fig. 9, BSs now have the ability to perform connection extension and rerouting. Handovers are initiated by the MH and the extended connection
Fig. 9 . Connection extension handover protocol proposed by Bell Labs (Murray Hill).
Handover Techniques for Wireless ATM Networks is established via messages 5 and 6. The connection rerouting process is completed when the old BS has received messages 9 and 10. Ignoring messages 1 and 2, which are concerned with MH registration, a total of eight handover signaling messages are required. Let T mhup and T mhdow n be the upstream and downstream disruption times experienced by the MH, respectively. Let T w ireless and T w ired be the times required to transmit a message over the wireless and wired channels, respectively. Let ` V ’ represent the time span from one event to another. The handover protocol analysis is
= T (3) V T (9) = 2(T w ireless ) + 2(T w ired ) T mhdow n = T (5) V T (10) T mhdow n = 3(T w ired ) T mhup
(1)
T mhup
(2)
51 to the MH. BSs in the loop chain will then deallocate reserved VCs during the loop removal process, in a hopby-hop fashion. When a MH migrates across local fabric domains, a partial route rebuild process is initiated. The rebuild is achieved by backtracking from the original path prior to handovers in search for a common node that can reach both old and new BSs to perform connection rerouting. However, the common node process does not take into consideration shorter resultant paths and resource utilization ef® ciency. Again, similar to Akyol and Cox, the ® rst encountered node is chosen with no regard to other possible common nodes.
(3)
2.9. Michigan Handover Scheme
(4)
A virtual path hierarchical handoff scheme has been proposed by Lin [10] of Michigan State University. The hierarchical organization is such that a group of BSs is connected to a cluster coordinator (CC) and these coordinators are then connected to a gateway (GW). The GW provides connectivity to the backbone ATM network. BSs, CCs, and the GW are interconnected via virtual paths. In addition, each BS has a preestablished handover VP to each of its neighboring BSs. Cell arrival sequence is preserved through the use of a special type of OA&M cell known as a `beforehandover ’ cell. The arguments behind adopting a strict hierarchical approach are:
The analysis reveals that the upstream and downstream traf® c disruption experienced by the MH is asymmetric, since the downstream disruption time is lower than the upstream time. As shown in Fig. 10, the VC loop removal process is nontrivial. Multiple VC loops can arise when a MH moves in a zigzag direction. Such VC loops can result in excessive usage of communication resources and cause route instability, as data are now forwarded multiple times. The loop removal process is explicitly initiated by the MH, and arriving cells are buffered at the current BS with an end-of-cell-stream marker appended. When the remote end BS receives this message, it ensures that it has forwarded all the remaining cells to the source BS prior to appending another end-of-cell-forward marker. When the source BS receives this marker, the cell streams are merged into a single stream where ATM cells will now ¯ ow from the source BS directly
Fig. 10 . Route loop removal procedures proposed by Bell Labs (Murray Hill).
· The ATM protocol standards do not allow the alteration of a mobile connection.
· To allow isolation of connection rerouting functions from the standard-based ATM network.
· To allow the elimination of searching for common node to perform handover. The hierarchical organization has resulted in four types of handovers: (a) intracluster uplink, (b) intracluster downlink, (c) intercluster uplink, and (d) intercluster downlink handovers. Hence, handover procedures for uplink and downlink traf® c are different. As shown in Fig. 11, a total of nine handover signaling messages are required to complete the handover. The handover is initiated by sending a handover HINT message from the MH to the old BS. This message contains the MH’ s current connections and identity of the target BS. The MH proceeds to acquire a new wireless channel in the target BS and sends a GREET message to the new BS. This results in the new BS returning an acknowledgm ent message back to the MH. Thereafter,
52
Toh and Akyol
Fig. 11 . The uplink handover protocol proposed by Michigan State.
uplink transmission continues via the new BS. The old BS determines if the handover is intercluster. If af® rmative, it forwards the handover HINT message to the old CC. The old CC then forwards it to the GW. The GW then performs authentication checking and route recon® guration. A con® rmation message is then propagated back to the new CC and the new BS. Based on the understanding from ref. 10, the GW will have to buffer all upstream data until the route recon® guration is completed. Explanation of the downstream handover procedure is, however, not presented and therefore evaluation of the upstream and downstream traf® c disruption times cannot be performed. Since the GW is responsible for all signaling and route recon® guring functions for all intercluster handovers, it can become a potential bottleneck, especially when the number of active mobile connections to be handed over increases. Hence, there is a need to evaluate its scalability. In addition, the Michigan proposal has not addressed issues related to handovers occurring outside the coverage of the GW.
2.10. Berkeley Handover Scheme The handover schemes [16] proposed by workers at UC Berkeley are not speci® c to wireless ATM networks, but are applicable to other connection-oriented wireless networks. Three types of handover schemes have been proposed: (a) full reestablishment (FR), (b) partial reestablishment (PR), and (c) multicast-based reestablishment (MCR). Two sets of protocols are introduced for situations with and without radio hint. In the FR scheme, an entirely new connection is established during handovers. In the PR scheme, the existing connection is modi® ed by establishing only a partial connection from a node in the existing path to the
new BS. In the multicast-based scheme, a new branch is originated from an intermediate node to the new BS during a handover. This is in fact the multicast join scheme discussed in Section 1. Under the scenario where radio hint is employed, a total of 11, 12, and 9 messages are required for the FR, IR, and MCR schemes, respectively. Performance evaluation of the handover schemes using an analytically derived model showed that using radio hint reduces the amount of buffering required at the BSs. In addition, network topology is regarded as important since it is advantageous when a network is constructed in a such way that the paths between BSs and the crossover node are short, thereby reducing communication resources required for handovers.
2.11. Cambridge Handover Scheme 2.11.1. Wireless Cell Clustering Architecture The Cambridge handover scheme evolved after a set of handover protocol design issues had been de® ned and studied. The handover architecture partitions handover management functions into intra- and interswitch types. Several BSs are connected to a common switch, known as the cluster ATM switch (CLS). MH migrations occurring under the coverage of the CLS result in intracluster handovers, while those migrating between CLSs result in intercluster handovers. The wireless cell clustering methodology employed has resulted in a distributed handover management, location management, and call admission control architecture. The protocol design issues addressed are: (a) cell delay variation, (b) exploitation of locality, (c) exploitation of mobility pro® les, (d) cell loss, (e) service disruption time, (f) circuit reuse ef® ciency, (g) data looping, (h) protocol robustness, (i) traf® c disruption symmetry,
Handover Techniques for Wireless ATM Networks ( j ) handovers of mobile multicast connections, (k) roaming across wireless ATM LANs, and (l) consideration of mobile QoS. These design issues have been elaborated in detail in ref. 38. 2.11.2. Handover Protocol Outline The Cambridge handover protocol is based on the concept of `handover via convergence.’ For intracluster handovers, the CLS is responsible for performing handovers for all MHs under its coverage. Hence, it acts as an anchor point where mobile connections are rerouted. Based on mobility pro® les gathered from the Active Badge system [12] installed in the Cambridge Computer Laboratory, it was discovered that mobile handovers are likely to occur locally. Hence, by allowing frequently occurring handovers to be handled by the CLS, handovers can be performed quickly. For intercluster handovers, a crossover switch (CX) has to be discovered and be held responsible for perform ing connection rerouting. Several CX discovery schemes have been proposed and studied in refs. 35 and 36. These discovery schemes can be initiated at the old or new BS and they utilize routing and QoS inform ation to derive an appropriate CX. A suitable CX is one that best ful® ls QoS requirements. By de® nition, a CX must be a node in the existing path that can reach both old and new BSs. Two versions of the handover protocol are proposed. One is based on radio hint while the other with-
53 out. As shown in Fig. 12, the with-hint handover protocol requires 11 handover signaling messages. Handover signaling channels are established through metasignaling, using reserved VCs. Essentially, the handover process involves three phases. The ® rst phase involves establishing a new partial path in advance. This requires the invocation of a CX discovery mechanism at the new BS during intercluster handovers. The second phase is to perform connection rerouting. The third phase is to tear down the old partial path. During handovers, both the upstream and downstream traf® c are handed off simultaneously. This gives rise to traf® c disruption symmetry and the advantage that only a single version of the handover protocol is required for both uplink and downlink traf® c. During cases when radio hint is not available, an explicit handover message is sent by the MH to the new BS (i.e., a backward handover) to initiate a new partial path setup and to perform connection rerouting. Although some ATM cells will be lost during this period, the ongoing connection will not be forced to terminate. In this scenario, seven handover signaling messages are required. 2.11.3. Handovers of Mobile Multicast Connections One of the oustanding features of ATM (other than QoS speci® cation) is multicasting. Multicasting is regarded as being potentially useful for video conferencing applications. Recently, multicasting has also
Fig. 12 . The with-hint handover protocol.
54
Toh and Akyol
been used for propagating control messages. In a wireless ATM network environm ent, in addition to unicast, mobile multicast connections can also exist. Hence, handovers of mobile multicast connections become a concern. While conventional multicast join and leave operations occur over the same path, this is not the case in a wireless ATM network. The Cambridge handover architecture is designed to support handovers of both mobile unicast and multicast connections [34, 39], regardless of whether the multicast tree is source-, server-, or cored-based. In addition, source and receiver handovers are performed independently, in a manner without reestablishing a new multicast tree and without affecting the existing direction of traf® c ¯ ows. Using the same handover protocol, handovers of multicast connections are supported by the concept of conditional deletion of the old partial path. This is necessary when the same VP / VC is used for MHs residing in the same wireless cell. Consider the scenario when a MH migrates from BS OLD to BS N E W . If more than one valid multicast group members exist, the MH migration would cause the CLS to BS OLD path to be deleted. This is undesirable, as other valid multicast group members residing in the old wireless cell are pruned out of the multicast tree. However, if different VP / VCs are assigned for MHs residing in the same wireless cell, then the partial path teardown process is no longer conditional. Hence, this conditional deletion of the old partial path is necessary in order to prevent shunting away valid multicast group members of a multicast tree.
In addition, the same crossover switch discovery mechanism is used during intercluster handovers. Instead of considering all possible nodes in the existing path as possible CXs (as in the case of mobile unicast connections) [36], only those nodes from the currently associated BS to the branching node (DP x )4 are considered for CX convergence. In this manner, connection rerouting at the selected CX can be performed without tearing down other branches of the tree which may have valid multicast group members. In addition, both connection rerouting and teardown operations are centered at a common CX. This is illustrated by Fig. 13. 2.11.4. Roaming Across Wireless ATM LANs The Cambridge handover protocol and CX discovery scheme can also be used to support handovers when a MH crosses wireless ATM LAN boundaries. According to Toh [40], each wireless ATM LAN is responsible for routing, location, and connection management within its own network. Boundary nodes are present in each LAN to support routing. Routing and QoS inform ation is summarized at each border node. These border nodes then exchange information among themselves in a manner as de® ned by the ATM P-NNI routing protocol [24]. Hence, only ® ne-grain routing and QoS information are known at each node within a wireless ATM LAN. For a wire4A
branching node is a switch in the multicast tree which performs a minimum of dualcast action.
Fig. 13 . Handovers of multicast connections without affecting direction of traf® c ¯ ows.
Handover Techniques for Wireless ATM Networks less ATM LAN employing a centralized connection management scheme, a connection server is present. When a connection spans across multiple wireless ATM LANs, each connection server will be responsible for establishing parts of the mobile connection. The important aspects to be highlighted are that CX discovery is only con® ned to the wireless ATM LAN concerned and the nodes to be considered for possible crossover node convergence are a portion of the connection path, not the entire path as in the case of handovers of unicast mobile connections. In addition, when a MH migrates from a wireless ATM LAN into another, the same handover protocol and CX discovery scheme applies. This ensures that extended roaming throughout homogeneous wireless ATM LANs is possible.
3. COMPARISONS OF HANDOVER STRATEGIES Although some comparison work has been performed and reported [5, 9], this paper aims to provide a more comprehensive review and comparison of existing handover schemes. As shown in Table I, in terms of signaling messages, we are more concerned with the number of handover messages occurring over the wireless and wired media and the number of multicast mes-
55 sages invoked. This is because multicast messages (MM) consume excessive communication resources. Among the handover schemes discussed, both the CMU and Columbia multicast-group schemes are multicast messages to support handover. These schemes, therefore, result in substantial handover signaling traf® c and they may not scale well with an increasing number of MHs and mobile connections. In terms of QoS adaptation during handovers, only those schemes proposed by CMU and Cambridge have such provisions. Among the handover protocols mentioned, only Cambridge, Olivetti, Bell (Murray Hill), NEC, and Michigan have performed some implementation work. With regard to handover protocol performance evaluation, only Cambridge, Michigan, and NEC have provided implementation results. Although many schemes have employed the concept of a common handover node, most do not have a detailed common node discovery procedure. Cambridge and Stanford have adopted a dynamic selection approach for deriving a common handover node. Michigan and Columbia use a ® xed common node concept. Bell (Murray Hill) uses a bracktrack common node discovery process, which is similar to Berkeley’ s proposal. NEC and Bell (Holmdel) have no means of discovering the most appropriate handover node. Only the Cambridge scheme considers all possible common nodes and selects the one
Table I. Comparisons of Various Handover Techniques for Wireless ATM Networks (Part 1 )
Existing handover schemes CMU Stanford Cambridge Bell Labs (Holmdel) Bell Labs (Murray Hill) Olivetti VTT Research Columbia NEC Research Michigan State Berkeley
Number of handover signaling messages
Support for QoS adaptation
Implemented?
Common node discovery
Exploitation of locality
6 MM + 5 (with hint) 7 (no hint) 10 (backward) 15 (forward) 11 (with hint) 7 (no hint) Not applicable 10 (VC Extension) 16 (with hint) 11 (no hint) 15 (backward) 15 (forward) Not applicable 14 (with hint) 9 (backward) FR (hint) = 11 IR (hint) = 12 MCR (hint) = 9
Yes
No
No
No
Multicast multicell
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No No No
No Doing Yes
No No No
No Yes Yes
No
No
No
Yes
Zone-based partial rebuild Cluster-based partial rebuild Partial rebuild Connection extension Domain-based partial rebuild Partial rebuild
No No No No
No Yes Yes No
No No No Yes
Yes Yes Yes No
Handover strategy
Multicast multicell Partial rebuild Pre-build VP-based Full rebuild Partial rebuild Multicast
56
Toh and Akyol
that best meets the QoS requirements. The VTT scheme, however, is unusual, as it does not consider MH migrations across the coverage of a cluster switch. In terms of localized handovers, all schemes except Bell (Holmdel), Berkeley, and CMU employ the concept of cell clustering. In addition, not all the schemes discussed have taken into consideration route optim ization after a handover. This refers to forming new paths that are either equal to or shorter in length (hop counts) than the previous path after a handover. Table I gives a summary of handover strategies employed by these schemes. CMU and Columbia schemes are multi-cellbased; Stanford’ s scheme is zone-based; Cambridge’ s scheme is cluster-based; Olivetti’ s scheme is domainbased; and Michigan’ s scheme is VP-based. Stanford, Cambridge, Bell (Holmdel), Olivetti, NEC, and Berkeley (PR) use the partial path rebuild methodology. Table II compares the robustness of the protocols. By robustness, we refer to the ability of the handover protocol to handle situations where a MH migrates in a `zigzag’ or `ping-pong’ manner, resulting in multiple handovers for the same connection. Another measure of robustness is the provision of timeout mechanisms in the handover protocol to handle situations when expected handover signaling messages are not received. Other than the Cambridge, Stanford, and Bell (Murray Hill) schemes, all the other schemes have not addressed robustness in the above-mentioned aspects. Support for handovers of mobile multicast connec-
tions is provided only by the Cambridge scheme. While others have proposed methods to perform grouped VC rerouting of multiple unicast connections, no consideration has been given to handovers of mobile multicast connections. The cell redirection mechanism employed by the various schemes is either based on real-time VC remapping performed at the ATM layer or by using double threads suck-spit buffers [10] performed at the transport interface layer. Some proposals, however, contain no information with regard to the cell redirection mechanism used. Preserving cell arrival sequence is another attribute for comparison. Many schemes have resorted to the use of special indicator or marker cells that are inserted into existing cell streams to signify the end of a cell segment transported over the old path just before this path is rerouted. Handover schemes based on the path rebuild strategy seem to preserve cell arrival sequence by synchronizing cell transmission at the old and new BSs. However, for handover schemes employing multicasting, cell duplication can occur unless BSs have synchronized to ensure that only a single copy of each cell is transmitted to the MH. The ATM forum was previously unconcerned with signaling support for wireless ATM networks. Hence, existing ATM signaling standards have no provision for mobility. However, a Wireless ATM Working Group [1] within the ATM Forum was recently formed to examine issues related to mobility support in a wireless ATM
Table II. Comparisons of Various Handover Techniques for Wireless ATM Networks (Part 2 ) Existing handover schemes
Robustness
Mobile multicast support
Cell sequencing support
No No Yes
No Yes Yes
No No Yes
No Yes Yes
Yes
No
No
No
Not always
Some
No
Yes
No Yes No Yes No
No No No No No
No No No No No
No Yes No Yes Yes
No
No
No
No
Route optimization
CMU Stanford Cambridge Bell Labs (Holmdel) Bell Labs (Murray Hill) Olivetti VTT Research Columbia NEC Research Michigan State Berkeley FR, IR, & NCR)
Redirection mechanism
Signaling coexistence?
Traf® c symmetry
Not available Not available Real-time VC remapping Not available
No Yes Yes
No No Yes
No
No
Real-time VC remapping Not available Not available Not available Not available Double threads suck spit Not speci® ed
No
No
No No No Yes No
No No No No No
Not speci® ed
No
Handover Techniques for Wireless ATM Networks environment [18, 25]. One attribute for comparison is the ability of a handover scheme to coexist with the existing signaling scheme for wired ATM networks. As shown in Table II, the NEC, Stanford, and Cambridge proposals can support such signaling coexistence. An earlier section revealed that some proposals provide different handover procedures for upstream and downstream traf® c of a mobile connection. This results in asymmetric traf® c disruption times for both upstream and downstream. For interactive multimedia applications, this may be undesirable. So far, only Cambridge’ s scheme provides traf® c disruption symmetryÐ a property veri® ed through both protocol analysis and implementation [33]. Table III further compares the various handover schemes by considering other attributes. All handover schemes except Cambridge and Olivetti are proposed for wide-area wireless ATM networks. However, prototype efforts will be much easier for LAN proposals. In terms of routing requirements, many schemes have not speci® ed which type of routing is being employed in their wireless ATM networks. Only Cambridge and NEC schemes mention ATM Forum’ s P-NNI routing [24]. The implementation complexity associated with a handover protocol proposal is another important attribute, as it governs the practicality of the proposal. The CMU scheme has high complexity since it has to initiate multicast setup and teardown to neighboring wireless cells during handovers. This scheme will be even more
57 complicated if mobile multicast connections have to be handed over. This also applies to the Columbia scheme. It is speculated that Stanford, Cambridge, NEC, and Berkeley (PR) proposals have low implementation complexity, since only partial paths are rebuilt. The Olivetti and VTT schemes are highly complicated because the former uses up to 16 messages to perform a handover, while the latter has separate handover procedures for uplink and downlink. It is unclear how upstream and downstream traf® c can be identi® ed prior to handovers. Although the handover schemes discussed so far are based on wireless ATM networks with a wired ATM backbone network and a wireless last-hop extension from the BSs, it is necessary to know how dependent these handover schemes are on the topology of the backbone ATM network. The Stanford, Columbia, NEC, and Michigan schemes are topology dependent. In terms of reliance on the connection management scheme employed, most schemes assume a distributed approach compared to either the distributed or centralized approach in the Cambridge scheme. Moreover, only Cambridge, Berkeley, and VTT have studied the buffering requirements during handovers. Finally, in terms of ¯ exibility associated with the assignment of a handover switch, most schemes do not provide the ¯ exibility of reselecting another handover switch during times of switch failure, except for the Cambridge proposal.
Table III. Comparisons of Various Handover Techniques for Wireless ATM Networks (Part 3 ) Existing handover schemes CMU Stanford Cambridge Bell Labs (Holmdel) Bell Labs (Murray Hill) Olivetti VTT Research Columbia NEC Research Michigan State Berkeley (FR, IR, & NCR)
HOS ¯ exibility
Buffering requirements studied?
No No Yes
No No Yes
No
Distributed Distributed Centralized or distributed Centralized
No
No
High
No
Distributed
No
No
High High High Low High Low
Not speci® ed Not speci® ed Yes Yes Yes No
Distributed Distributed Distributed Distributed Distributed Distributed
No No No No No No
No Yes No No No Yes
LAN (L) or WAN (W)
Routing requirement
Implementation complexity
Topology dependent?
W W, L L
Not speci® ed Not speci® ed P-NNI
High Low Low
No Some No
W
Not speci® ed
Not speci® ed
W
Not speci® ed
L W W W W W
Not speci® Not speci® Not speci® P-NNI Not speci® Not speci®
ed ed ed ed ed
Connection management scheme
58 4. FUTURE HANDOVER PARADIGMS The handover strategies covered in this paper are based on wireless ATM networks employing a single layering of wireless cells. However, to enhance wireless throughput and to counteract radio interference, multilayer, commonly known as `overlay,’ wireless cell architectures have been proposed [13]. With such an architecture, the lowest layer of wireless cells consists of picocells, while the higher layers consist of micro and macrocells. Handovers can therefore be transferred to a switch entity which is managing a higher layer of wireless cells. The ability to change from one congested wireless cell to a less congested one increases the overall system robustness and enhances its capability to provide quality-ofservice assurance. Instead of implementing mobility functions into switches, suggestions have been proposed to move such functions out of the switches, perhaps into a switch controller [14, 20]. The external switch controller can then communicate with the switch via some application programming interface (API) such that various VC rerouting strategies based on rebuild, multicast, extension, etc., can be implemented [31]. With such an interface, VCI / VPI tables can be altered to support mobility. To preserve the ATM cell arrival sequence during a mobile handover, several schemes have proposed the use of `marker’ ATM cells that are inserted at speci® c points in a cell stream, i.e., inband signaling. This, however, will require a switch that has the ability to enable and disable data ¯ ow on a particular VC. However, if outof-band signaling is used, cell stream will be resumed when the marker cell arrives via a dedicated signaling channel used for handover signaling [37]. While handovers are currently initiated in advance with the aid of radio hint, prediction of handovers using arti® cial neural nets has not been seriously and thoroughly investigated. While arti® cial intelligence is often associated with a knowledge database and intelligent programs, neural networks are self-learning networks. A neural network is comprised of numerous processing units, known as neurons, that learn from experience without the presence of a mathematical model of how the results depend upon inputs. With historical data, a neural net shapes and program s itself to model the data. Hence, by applying local and wide-area migration pro® les of mobile users to a suitable neural net, predictions about a user’ s next moment can be generated. These predicted outputs, along with the results obtained through signal level detections, may then be used to trigger accurately
Toh and Akyol handovers well in advance. This is an area that demands further research. Finally, the importance of de® ning mobile quality of service (QoS) during handovers has been neglected. While mobile QoS has been de® ned and QoS adaptation strategies have been proposed [38], further attention should be directed by the ATM Forum Wireless ATM Group and other researchers in this ® eld. A challenge in the future is for standardization bodies and commercial and academic researchers to broaden their scope in the pursuit of de® ning a handover paradigm that can ful® ll a variety of mobility requirements.
5. CONCLUSION In this paper, we have provided a description of various existing handover schemes proposed for wireless ATM networks. We have also de® ned the terms that are commonly used in mobile handover work. We have revealed that existing handover schemes are either based on single- or multi-cell-based schemes. They support mobility by either fully or partially rebuilding or extending mobile connections or establishing multiple connections during mobile handovers. We have also revealed that certain schemes exploit the advantage of locality by partitioning and dedicating the handover management load to different entities in the network. When a MH migrates beyond the coverage of the handover management entity, some protocols allow the transfer of handover control to another appropriate handover management entity. Finally, we have provided a detailed comparison of the features and attributes of the various handover schemes along with a discussion on future handover paradigms.
REFERENCES 1 . A. Ayyagari, J. Harrang, and S. Ray, Extensions to proposed charter, scope & work plan for Wireless ATM Working Group, ATM Forum Document, Number 96 -0672 , June 1996 . 2 . B. A. Akyol and Donald C. Cox, Handling mobility in a wireless ATM network. In Proceedings of Infocom 96 , March 1996 . 3 . B. A. Akyol and Donald C. Cox, Rerouting for handoff in a wireless ATM network. In Proceedings of International Conference on Universal Personal Communications (ICUPC’ 96 ), September 1996 . 4 . B. A. Akyol and D. C. Cox, Signaling alternatives in a wireless ATM network: Migratory signaling. In Proceedings of ATM’ 96 Workshop , August 1996 . 5 . B. A. Akyol and D. Cox, Re-routing for handoff in a wireless ATM network, IEEE Personal Communications Magazine , Vol. 3 , No. 5 , 26 ±33 . October 1996 .
Handover Techniques for Wireless ATM Networks 6 . B. A. Akyol and D. C. Cox, Signaling alternatives in a wireless ATM network, IEEE Journal on Selected Areas in Communications, Vol. 15 , No. 1 , 35 ±49 , January 1997 . 7 . D. A. Levine, I. F. Akyildiz, and M. Naghshimeh, The shadow cluster concept for resource allocation and call admission in ATMbased wireless networks. In Proceedings of ACM First International Conference on Mobile Computing and Networking (MOBICOM’ 95 ), November 1995 , pp. 142 ±150 . 8 . R. Earnshaw, Footprints for mobile communications. In Proceedings of the 8 th IEE UK Teletraf® c Symposium , April 1991 . 9 . E. Ayanogly, K. Eng, and M. Karol, Wireless ATM: Limits, challenges and proposals, IEEE Personal Communications Magazine , Vol. 3 , No. 4 , August 1996 . 10 . H.-T. Lin, A handoff scheme for wireless ATM networks. In Proceedings of International Conference on Universal Personal Communications (ICUPC’ 96 ), September 1996 . 11 . H. Mitts, H. Hansen, J. Immonen, and S. Veikkolainen, Lossless handovers for wireless ATM. In Proceedings of Second ACM International Conference on Mobile Networking and Computing , November 1996 . 12 . R. Want and A. Hopper, The active badge location system, ACM Transactions on Information Systems, Vol. 10 , No. 1 , pp. 91 ±102 , January 1992 . 13 . I. Chih-Lin, A microcell macrocell cellular architecture for low and high mobility wireless users, IEEE Journal on Selected Areas in Communications , Vol. 11 , No. 6 , 885 ±891 , August 1993 . 14 . I. Cidon, T. Hsiao, et al., The OPENET Architecture, Technical report, Sun Microsystems, TR- 95 -37 , December 1995 . 15 . J. Porter, A. Hopper, et al., The ORL Radio ATM System, Technical report, TR 96 .5 , Olivetti Research Lab, 1996 . 16 . K. Keeton, Providing connection oriented network services to mobile hosts. In Proceedings of the USENIX Symposium on Mobile and Location Independent Computing , August 1993 , pp. 83 ±103 . 17 . K. Lee, Adaptive network support for mobile multi-media. In Proceedings of ACM First International Conference on Mobile Computing and Networking (MOBICOM’ 95 ), November 1995 , pp. 62 ±72 . 18 . L. Dellaverson, Proposed charter, work plan and schedule for a Wireless ATM Working Group, ATM Forum Document, Number 96 -0712 , June 1996 . 19 . M. Veerarghavan, T. F. La Porta, and R. Ramjee, A distributed control strategy for wireless ATM networks. In Proceedings of 9 th IEEE Computer Communications Workshop , December 1994 . 20 . K. Merwe, Switchlets and dynamic virtual ATM networks, Presentation Slides at Cambridge SRG Talk, March 1996 . 21 . A. S. Acampora and M. Naghshineh, An architecture and methodology for mobile-executed handoff in cellular ATM networks, IEEE Journal of Selected Area in Communications , Vol. 12 , No. 8 , pp. 1365 ±1375 , October 1994 . 22 . P. Agrawal, E. Hyden, P. Krzyzanowsji, M. B. Srivastava, and J. Trotter, Multimedia information processing in the SWAN mobile networked computing screen. In Proceedings of Multimedia Computing and Networking (MMCN’ 96 ), January 1996 . 23 . P. Agrawal, E. Hyden, et al., SWAN: A mobile multimedia wireless network, IEEE Personal Communications Magazine, Vol. 3 , No. 2 , 18 ±33 , April 1996 . 24 . R. Coltun, E. Hoffman, et al., A P-NNI routing protocol proposal, ATM Forum Document, Number 94 -0492 , May 1994 . 25 . R. Yuan, K. Chao, T. Chen, and B. Khasnabish, An extended reference model for wireless mobile ATM networks, ATM Forum Document, Number 96 -0688 , June 1996 . 26 . R. Yuan, S. Biswas, and D. Raychaudhuri, Mobility support in a wireless ATM network. In Proceedings of 5 th WINLAB Workshop on Third Generation Wireless Information Networks, April 1995 , pp. 335 ±345 . 27 . R. Yuan, S. Biwas, L. French, J. Li, and S. Veikkolainen, A.
59
28 . 29 .
30 . 31 .
32 .
33 .
34 .
/
/
35 .
36 .
37 . 38 . 39 .
40 .
signaling and control architecture for mobility support in wireless ATM networks, ACM Journal on Mobile Networks and ApplicationsÐ Special Issue on `Wireless ATM’ , Vol. 1 , No. 3 , January 1997 . R. Katz, The case for wireless overlay networks. In Proceedings of Multimedia Computing and Networking (MCN’ 96 ), January 1996 . D. Raychaudhuri, ATM based transport architecture for multi-services wireless PCN. In Proceedings of the IEEE Supercomm / ICC, May 1994 , pp. 559 ±565 . R. Bubenik and P. Flugstad, Leaf initiated join (LIF) extensions, ATM Forum Document, 94 -0325 R 2 , September 1994 . P. P. Mishra and Mani B. Srivasatava, Programmable ATM switches for mobility support. In Proceedings of OPENSI G (Open Signalling for Middleware and Services Creation ), April 1996 . P. P. Mishra and M. B. Srivastava, Call Establishment and Rerouting in Mobile Computing Networks, Technical report, AT&T Bell Laboratories, September 1994 ; TM 11384 -940906 - 13 . C.-K. Toh, The design and implementation of a hybrid handover protocol for multimedia wireless LANs. In Proceedings of the First ACM International Conference on Mobile Computing and Networking (MOBICOM’ 95 ), November 1995 , pp. 49 ±61 . C.-K. Toh, A handover paradigm for wireless ATM LANs. In Proceedings of the ACM Symposium on Applied Computing (SAC’ 96 )Ð Special Track on Mobile Computing Applications and Systems, February 1996 . C.-K. Toh, Crossover switch discovery schemes for fast handovers in wireless ATM LANs, ACM Journal on Mobile Networks & ApplicationsÐ Special Issue on `Routing In Mobile Communication Networks’ , Vol. 1 , No. 2 , November 1996 . C.-K. Toh, Performance evaluation of crossover switch discovery algorithms for wireless ATM LANs. In Proceedings of the IEEE International Conference on Computer Communication s (InfoCom’ 96 ), March 1996 , pp. 1380 ±1387 . C.-K. Toh, Protocol aspects of mobile radio networks. PhD thesis, Cambridge University Computer Laboratory, August 1996 . C.-K. Toh, Wireless ATM and Ad-Hoc Networks: Protocols and Architectures , Kluwer, Dordrecht, 1996 . C.-K. Toh, A unifying methodology for handovers of heterogeneous connections in wireless ATM networks, ACM Computer Communication Review , January 1997 . C.-K. Toh, Extending mobility support for roaming across wireless ATM LANs. In IEEE SICON’ 97 , April 1997 .
Chai-Keong Toh was born in 1965 ; he received his Diploma in electronics and communication engineering from the Singapore Polytechnic with a certi® cate of merit award in 1986 , his B.Eng. degree in electronics engineering with ® rst class honors from the University of Manchester Institute of Science and Technology in 1991 , and his Ph.D. degree in computer science from the Computer Laboratory, University of Cambridge, England, in 1996 . Prior to Cambridge, he was an
60 R&D Engineer, Network Specialist, and Member of Technical Staff. While at Cambridge, he was an Honorary Cambridge Commonwealth Trust Scholar and a King’ s College Cambridge Research Scholar. He founded the Cambridge Mobile Special Interest Group in 1994 and he authored, Wireless ATM and Ad-Hoc Networks: Protocols and Architectures, which was published by Kluwer Academic Press in 1996 . Dr. Toh is the inventor of the Cambridge Ad-Hoc Mobile Routing Protocol based on the concept of associativity, which is pending a U.S. patent. He is an Area Editor for IEEE Journal on Communications Survey, a Feature Editor for ACM Mobile Computing and Communications Review , and serves on the editorial board for IEEE Network Magazine and Personal Technologies Journal (Springer-Verlag). He is session chair on Wireless ATM for IEEE PIMRC’ 97 , MoMuC’ 97 , and IEEE SICON’ 97 . He is a member of Sigma Xi, ACM, IEEE, IEE, USENIX, and the New York Academy of Sciences. He is currently with Hughes Research Labs, leading research efforts on ad hoc networks and wireless / satellite ATM.
Bora Akyol received his B.S. degree in electrical engineering from Bilkent University, Ankara, Turkey, in 1992 , and his M.S. degree in electrical engineering from Stanford University, Stanford, California, in 1993 . He is currently a Ph.D. candidate at Stanford University working on wireless ATM networking. His research interests include network signaling, media access control, and resource allocation for wireless communication networks. Bora will be joining BBN after receiving his Ph.D.
Toh and Akyol