Computer Networks 55 (2011) 147–158

Contents lists available at ScienceDirect

Computer Networks

journal homepage: www.elsevier .com/ locate/comnet

An enhanced information server for seamless vertical handover inIEEE 802.21 MIH networks q

Younghyun Kim a, Sangheon Pack a,⇑, Chung Gu Kang a, Soonjun Park b

a School of Electrical Engineering, Korea University, Seoul, Republic of Koreab Mobile Communication Technology Research Laboratory, LG Electronics Inc., Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history:Received 8 February 2010Received in revised form 12 June 2010Accepted 17 August 2010Available online 25 August 2010Responsible Editor: L. Jiang Xie

Keywords:Vertical handoverIEEE 802.21 MIHSpatial localityTemporal localityEnhanced information server

1389-1286/$ - see front matter � 2010 Elsevier B.Vdoi:10.1016/j.comnet.2010.08.005

q A preliminary version of this paper was preseSymposium on Communication and Information TecSeptember 2009.⇑ Corresponding author. Tel.: +82 2 3290 4825; fa

E-mail address: shpack@korea.ac.kr (S. Pack).

Seamless vertical handover is a key requirement in heterogeneous wireless networkswhere different networks are integrated. In this paper, we introduce an enhanced informa-tion server (EIS) to accelerate vertical handover procedures in IEEE 802.21 media indepen-dent handover (MIH) networks. Based on the EIS, we propose an improved verticalhandover procedure in which wireless channel conditions are estimated by exploiting spa-tial and temporal locality at the EIS, and therefore time consuming channel scanning pro-cedures can be skipped. Simulation results demonstrate that the proposed scheme canreduce the vertical handover latency under diverse environments.

� 2010 Elsevier B.V. All rights reserved.

1. Introduction

In future wireless/mobile networks, heterogeneouswireless networks will be available, e.g., IEEE 802.11a/b/gWIreless FIdelity (WiFi), IEEE 802.11p Wireless Access inVehicular Environments (WAVE), IEEE 802.16 World Inter-operability for Microwave Access (WiMAX), UniversalMobile Telecommunications Systems (UMTS), High-SpeedDownlink/Uplink Packet Access (HSDPA/HSUPA), LongTerm Evolution (LTE), and so on. In such heterogeneouswireless networks, vertical handover (VHO) is a criticalchallenge to achieve always best connectivity (ABC)services, and extensive research has been carried out[1–3].

For VHO signaling framework, IEEE 802.21 mediaindependent handover (MIH) has been introduced [4]. To

. All rights reserved.

nted at Internationalhnology (ISCIT) 2009,

x: +82 2 872 2045.

facilitate vertical handover in heterogeneous networks,IEEE 802.21 MIH defines three services: Media Indepen-dent Event Service (MIES), Media Independent CommandService (MICS), and Media Independent Information Ser-vice (MIIS). MIES provides event reporting, event filtering,and event classification services depending on the linkdynamics. On the other hand, MICS supports methods tosend commands from higher layers to lower layers. MIISdefines a mechanism for an MIH entity to discover avail-able neighboring network information within a geograph-ical area.

In the current 802.21 MIIS specification, a mobile node(MN) gets the neighborhood information by requestinginformation elements (IEs) from the information server(IS). The IS can provide both static and dynamic informa-tion. The names of service providers, medium accesscontrol (MAC) addresses, and channel information of theMN’s current network neighborhood are examples of staticinformation. On the contrary, dynamic informationincludes link-layer parameters such as data rate, through-put, and other higher layer service information to make anintelligent handover decision. However, the currently

148 Y. Kim et al. / Computer Networks 55 (2011) 147–158

defined IEs focus on the static information and thereforehow to define and make use of dynamic information inthe IE is a still remaining issue.

In this paper, we introduce an enhanced IS (EIS) archi-tecture where spatial and temporal localities are exploitedfor seamless vertical handover. Specifically, an MN period-ically reports its location, link conditions (e.g., received sig-nal strength (RSS)), and the timing information to the EIS.The update procedures are performed by all MNs and theaccumulated information can be utilized by the EIS in acollaborative manner. That is, for a handover trigger event,the EIS estimates the MN’s current link condition by meansof the information at the EIS, and determines the best pointof attachment (PoA) for the MN. Consequently, it is possi-ble to skip time-consuming channel scanning proceduresfor a handover trigger event. Through extensive simula-tions, we evaluate the hit probability, which is the proba-bility that the selected PoA by the proposed scheme isthe same as the one decided by the full scanning proce-dure, and vertical handover latency. Simulation resultsdemonstrate that the EIS architecture guarantees higherhit probability with remarkably lower vertical handover la-tency under different situations.

The remainder of this paper is organized as follows. Sec-tion 2 summarizes related works and Section 3 presentsthe overview of IEEE 802.21 MIH. Section 4 and Section 5describe the enhanced IS architecture and propose verticalhandover with the EIS, respectively. Finally, simulation re-sults and concluding remarks are given in Sections 6 and 7,respectively.

Upper Layers (Layer 3 +)

Lower Layers (Layer 1 + Layer 2)

MIH Function

Local Stack

Remote MIH Event

Link Events

MIH Events

Upper Layers (Layer 3 +)

Lower Layers (Layer 1 + Layer 2)

MIH Function

Local Stack

Remote MIH Command

Link Commands

MIH Commands

SEIMSCIM

Fig. 1. MICS & MIES models.

2. Related works

Unlike horizontal handover, since vertical handover is aprocess of performing the handover between differentwireless technologies, it involves the following threephases: network discovery, handover decision, and hand-over execution. In the network discovery phase, an MNgets the neighbor network information such as cost, net-work security, jitter, bit error rate (BER), and so on. Usingthe obtained neighbor network information, the MN (orthe IS) decides the target network which will be connectedin the handover decision phase. After that, during thehandover execution phase, the MN does handover to thetarget network. Among the three phases, IEEE 802.21MIH can be used for both system discovery and handoverdecision. Hence, related works on IEEE 802.21 MIH canbe classified into two categories: 1) how to select an appro-priate target network [5–8] and 2) how to reduce handoverdelay in IEEE 802.21 MIH [9–11].

Several network selection schemes with QoS provisionwere proposed in [5–8]. [5] proposed an intelligent net-work selection scheme in multihomed session initiationprotocol (SIP) based network mobility (NEMO) environ-ments. To support QoS continuity between UMTS/802.16enetworks, [6] proposed a novel network-initiated hand-over scheme which includes QoS measurement setup, pas-sive reservation, and activation steps. [7] suggested a loadbalancing scheme for reducing the queueing delay in MIH-based Proxy MIPv6 (PMIPv6) networks. [8] proposed an

improved multiple attribute decision making (MADM)method, which utilizes IS information.

On the other hand, to reduce handover delay in IEEE802.21 MIH, [9] improved Fast Handover for Mobile IPv6(FMIPv6) by employing MIH in vehicular environments.[10] introduced a pre-binding update scheme, which usesthe target network information from the IS. In [11], exper-iment results on PMIPv6 handover delay in IEEE 802.21networks were reported.

However, in these previous works, little attention hasbeen paid to the design of the IS architecture and thedevelopment relevant algorithms for seamless verticalhandover, which are main contributions of this work.

3. IEEE 802.21 media independent handover (MIH)

IEEE 802.21 MIH is an evolution to support verticalhandover by providing capabilities to detect and initiatehandover from one network to another [4]. Also, MIH pro-vides services to assist handover between two 802 net-works (e.g., from 802.3 to 802.11) or an 802 network anda non-802 network (e.g., from 802.11 to HSDPA). To sup-port vertical handover, MIH defines a logical functional en-tity called MIH function (MIHF), which provides threeservices: media independent event service (MIES), mediaindependent command service (MICS), and media inde-pendent information service (MIIS).

Fig. 1 shows MICS and MIES models. MICS enables MIHusers to manage and control link behavior relevant tohandover. As shown in Fig. 1, MICS is initiated by higherlayers, and MICS commands are sent to lower layersthrough the MIHF. ‘‘MIH Scan”, ‘‘MIH Configure”, and‘‘MIH Switch” are typical examples of MICS. On the otherhand, MIES defines the functions of event classification,event filtering, and event reporting to upper layers. Specif-ically, different events such as ‘‘Link Up”, ‘‘Link Down”,‘‘MIH Link Up”, and ‘‘MIH Link Down” are defined. ‘‘LinkUp” and ‘‘Link Down” events are generated by lower layers(layer 1 or layer 2), and these events are notified to theMIHF. Then, the MIHF reports this situations to upper lay-ers by triggering ‘‘MIH Link Up” and ‘‘MIH Link Down”events.

MIIS provides information about neighboring networksfor MIH user. This information, such as neighbor maps, link

Mobile node IS

Handover trigger

Neighbor list query

Vertical handover to Mobile WiMAX PoA

Neighbor list response ( HSDPA PoA, Mobile WiMAX PoA )

scanning

scanning

Network selection Hand

over

late

ncy

Target list query (with the results of scanning )

Target list response ( Mobile WiMAX PoA > HSDPA PoA )

Mobile WiMAX PoAHSDPA PoA

Fig. 2. IEEE 802.21 vertical handover procedure.

1 Of course, the MN can maintain the connection with the old PoA duringscanning procedures and thus no disruption occurs. However, since the linkquality with the old PoA will decrease significantly after the handovertrigger, it is important to minimize the latency from the time of handovertrigger to the time of actual handover to a new PoA. Therefore, we considerthis latency as the handover latency.

Y. Kim et al. / Computer Networks 55 (2011) 147–158 149

layer information, and availability of services, can be usedto select the target network when vertical handover isneeded. To maintain neighboring network informationand offer it to the MNs, IEEE 802.21 MIH has defined aninformation server (IS). However, the design and imple-mentation of the IS are beyond scope of the standard.Hence, in this work, we introduce a novel design of the ISto decide an appropriate target network for the MN with-out any channel scanning procedure. More details will beelaborated in Section 5.

Fig. 2 illustrates the vertical handover procedure in IEEE802.21 MIH [12]. Detailed procedures are as follows.

1. When a handover is triggered, an MN sends a neighborlist query message, which requests the neighboring net-work list from the IS. Then, the IS sends a neighbor listresponse message containing the neighboring networkinformation such as network type, subnet prefix, pointof attachment (PoA) address, and so on.

2. After receiving the response message, the MN scans fornearby PoAs to measure the up-to-date channel states.

3. Once the MN completes the scanning procedure, theMN sends a target list query message containing themeasured channel state values to the IS. Then, the ISevaluates the suitability of each network as the hand-over target network by means of vertical handoverdecision functions. After completing the evaluation,the IS sends a target list response message to the MN.

4. Finally, the MN executes a handover to the target net-work with the highest priority, i.e., Mobile WiMAX inFig. 2, in the target list.

As illustrated in Fig. 2, the MN should perform multiplescanning procedures to measure the channel states from

neighboring networks. As reported in [13], these scanningprocedures lead to significant handover latency1. Especiallywhen the number of available networks is large, drasticallyincreased handover latency can be observed.

4. Enhanced IS architecture

Although the IS provides the neighboring networkinformation, the scanning procedure occupies a significantportion in the handover latency. Hence, we introduce a no-vel IS architecture, i.e., enhanced information server (EIS)where the channel state can be estimated by exploitingtemporal and spatial localities in the channel state. Conse-quently, vertical handover delay can be reduced by the EISsince the MN does not need to perform time-consumingscanning procedures.

To describe the EIS architecture, we have severalassumptions. First, an MN can measure its current locationby means of localizing techniques or GPS. Note that spe-cific localization techniques are beyond the scope of thispaper. Then, an extended channel state information (ECSI)can be described as four-tuple,

ECSI ¼ id; t; ðx; yÞ; dð Þ; ð1Þ

where id is the PoA identifier, t is the link quality measure-ment time, and (x,y) is the MN’s location. d represents the

MN

Mobile WiMAX PoA

HSDPA PoA

CDMA PoA

Internet

EIS

MN

Fig. 3. Network model using enhanced IS.

150 Y. Kim et al. / Computer Networks 55 (2011) 147–158

link quality (i.e., RSS) to the PoA id2. Hereinafter, ECSI.id, EC-SI.t, ECSI. (x,y), and ECSI.d refer to id, t, (x,y), and d of a ECSI,respectively.

To construct the EIS, every MN should measure its loca-tion and the RSS to available PoAs, and then it should notifythe information to the EIS by sending a message with ECSI3.Then, as time goes, the EIS keeps more ECSIs measured atvarious locations and times. By collecting this information,without any channel scanning procedure, the RSS with aPoA can be estimated. How to estimate the channel statewill be given in the next section.

Fig. 3 shows the network model with the EIS. Assumethat an MN is currently connected to Mobile WiMAXPoA. At the same time, the MN lies on in the service areasof HSDPA and CDMA networks. Hence, the MN reports EC-SIs on the Mobile WiMAX PoA, HSDPA PoA, and CDMA PoAto the EIS. On the other hand, the MN can receive the sig-nals from the CDMA PoA as well as the Mobile WiMAX PoA,when the MN moves to another area as the dotted arrow.Therefore, the MN reports ECSIs on the Mobile WiMAXPoA and CDMA PoA to the EIS.

2 This paper focuses on how to estimate link quality by means of EISinformation, i.e., which link quality metric is used is beyond the scope ofthis paper. In other words, the proposed scheme can be applied to differentlink quality metrics such as signal-to-noise ratio (SNR) and signal-to-interference plus noise ratio (SINR). In this paper, since RSS is widely usedas a vertical handover parameter as in [14–16], we assume the use of RSS asa link quality metric.

3 This notification can be performed periodically or can be piggybackedwith other messages. Therefore, the additional overhead due to thenotification can be controlled and minimized.

5. Improved VHO procedure with EIS

As mentioned in Section 4, MNs report the ECSIs to theEIS periodically. Based on ECSIs, the EIS estimates the chan-nel condition with PoAs nearby the MN, and therefore ver-tical handover latency can be reduced. However, verticalhandover latency may be high if accuracy of channel esti-mation is not sufficiently high. Hence, it is important forthe EIS to estimate the RSS with high accuracy. In this sec-tion, we present a RSS estimation algorithm and an im-proved VHO procedure.

Let (x0,y0) and t0 represent the location and time whereand when the MN triggers a vertical handover, respec-tively. When a vertical handover is triggered, the MN sendsa target list query message including (x0,y0) and t0 to theEIS. Then, the EIS estimates the RSS by exploiting spatialand temporal localities in the wireless channel state.

First, for a ECSI, the elapsed time is computed ast0 � ECSI.t. By considering the time correlation in the chan-nel state (i.e., recent channel state information is more clo-sely related to the current channel state information thanthe old one), only ECSIs within a time period (i.e.,t0 � ECSI.t < W, where W is the time window) are consid-ered in the RSS estimation (temporal locality). For efficientmanagement of the EIS, out-of-date ECSIs will be evictedfrom the EIS.

In a similar context, the spatial locality is investigated.Since the RSS information in a location far away from(x0,y0) has less impact on the RSS estimation (spatial local-ity), only ECSIs within a radius RM are selected (i.e.,ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðx0 � ECSI:xÞ2 þ ðy0 � ECSI:yÞ2

q< RM). Moreover, different

Y. Kim et al. / Computer Networks 55 (2011) 147–158 151

weights are assigned depending on the locations by meansof an exponentially weighted moving average (EWMA)scheme.

Detailed procedure for estimating the RSS from ECSIs inthe EIS is illustrated in Algorithm 1. Let N and ni be thenumber of available PoAs and the number of ECSIs of theith PoA (i.e., PoAi), respectively. Also, ECSIi(j) denotes thejth ECSI of PoAi. Then, the EIS selects ECSIi(j) s satisfyingtwo conditions on spatial and temporal localities, andstores the chosen ECSIi(j) s to a buffer h (see lines 7–8). Ifthere are no ECSIi(j) s satisfying the above conditions, theexisting vertical handover procedure is performed, asshown in Fig. 2 (see lines 13–14). Otherwise, the EIS sortsthe buffer h in an ascending order (see line 16). Note sort(a,b) function sorts a depending on b. Finally, the expectedRSSes of PoAi, dE

i , can be obtained by the EWMA scheme(see lines 19–22).

Algorithm 1. RSS estimation

1: initiate RM, W, N;2: i 0;3: while i < N do4: j 0;5: m 0;6: while j < ni do7: if t0 � ESCIi(j).t < W & &ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðx0 � ECSIiðjÞ:xÞ2 þ ðy0 � ECSIiðjÞ:yÞ2q

< RMthen

8: h(m) ESCIi(j)9: m ++ ;

10: end if11: j ++ ;12: end while13: if m == 0 then14: exit;15: end if16:

sort h;ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðx0 � hðkÞ:xÞ2 þ ðy0 � hðkÞ:yÞ2

q� �;

for 0 6 k 6m � 1;

17: dEi ¼ hð0Þ:d;

18: j 1; do19: while j < m do20: dE

i ¼ a � dEi þ ð1� aÞ � hðjÞ:d;

21: j ++ ;22: end while23: i ++ ;24: end while

After estimating the RSS, the EIS should determine thetarget network list by considering the MN’s vertical hand-over criteria, which include the channel state or RSS. Let xi

be the weighting factor of criteria i. NotePK

i¼1xi ¼ 1 whereK is the number of vertical handover criteria. Since our fo-cus is given to the RSS estimation by the EIS, we considerthe channel state as a sole handover criteria; however,the proposed scheme can be easily extended to any typesof vertical handover decision algorithms.

As shown in Fig. 4, the overall vertical handover proce-dure in the proposed scheme can be described as follows:

1. When a vertical handover is triggered, an MN sends atarget list query message that includes the MN’s loca-tion (x0,y0) information and the current time t0 to theEIS.

2. On the receipt of the query, the EIS estimates the RSSesfor each PoA by using Algorithm 1. After that, based onthe vertical handover decision criteria, the EIS con-structs a target network list sorted by the priority.

3. Then, the EIS sends a target list response message to theMN.

4. Finally, the MN executes a vertical handover to the net-work with the highest priority. If the handover is failed,another network with the second highest priority istried.

6. Simulation results

In this section, we describe performance evaluation re-sults. We have developed an event-driven simulator andperformed extensive simulations. For credible simulations,we have conducted ten simulation runs with independentand identically distributed parameters (e.g., location,velocity, pause time, etc.) using random seed values.

6.1. Simulation environment

For a wireless channel model, COST-231 Hata model ischosen since it is widely used for predicting path loss inmobile wireless system [17]. Although COST-231 Hatamodel is designed for the frequency band from 500 MHzto 2000 MHz, it can be used up to 3 GHz. Under COST-231 Hata model, the path loss is given by

PLðdÞdB ¼ 46:3þ 33:9log10ðf Þ � 13:82log10ðhbÞ � aðhrÞþ 44:9� 6:55log10ðhbÞð Þlog10ðdÞ þ cm; ð2Þ

where f is the carrier frequency, hb is the antenna height ofa PoA, and d is the distance between the PoA and MN. a(hr)is the MN’s antenna height correction factor and, for urbanenvironments, it can be obtained from

aðhrÞ ¼ 3:20 log10ð11:75hrÞð Þ2 � 4:97: ð3Þ

On the other hand, for suburban environments, a(hr) can eobtained as

aðhrÞ ¼ 1:1log10ðf Þ � 0:7ð Þhr � 1:56log10ðf Þ � 0:8ð Þ; ð4Þ

where hr is the MN’s antenna height. The parameter cm is3 dB and 0 dB for urban and suburban environments,respectively. Then, from (2)–(4), the RSS c at a distance dcan be computed as

cðdÞdB ¼ PtdB � PLðdÞdB; ð5Þ

where PtdB is a transmitting power in dB.In terms of mobility model, we use the random way-

point mobility model where an MN chooses two parame-ters, i.e., direction and speed, before moving from an initiallocation. The direction and the speed are uniformly distrib-uted 0 � 2p(radian) and 0 � VMAX (m/s), respectively. The

Mobile node EIS

.

.

.

Handover trigger

Target list query ( )( )0 0 0, , x y t

Target list response ( Mobile WiMAX PoA > HSDPA PoA )

Network selection

Vertical handover to Mobile WiMAX PoA

T

T

HSDPA PoA Mobile WiMAX PoA

Hand

over

late

ncy

( )( ), , , , ECSI id t x y δ

( )( ), , , , ECSI id t x y δ

( )( ), , , , ECSI id t x y δ

Fig. 4. Vertical handover procedure with EIS.

152 Y. Kim et al. / Computer Networks 55 (2011) 147–158

MN moves from an initial location to the selected directionwith the chosen constant speed during TM, where TM is a

Table 1Simulation parameters.

Parameter Value

Mobile WiMAX’s frequency 2.3 GHzHSDPA frequency 2 GHzMobile WiMAX’s PoA transmitting power 20 dBmHSDPA’s PoA transmitting power 40 dBmRH, RW 4 km, 1 kmhr, hb 1.5 m, 32 mTM, TP [1,60] s, [1,10] sVMAX,TI 20 m/s, 60 sNumber of MNs 100Simulation time 100 min

Fig. 5. Effect of

random moving time. When the MN reaches at the desti-nation, it waits for a pause time TP, and then the MNchooses another direction and speed. These processes areiterative. In addition, each MN sends a ECSI to the EIS withan interval TI.

Table 1 summarizes simulation parameters based on[18,19]. In Table 1, TM and TP are uniformly distributed in[1,60] s and [1,10] s. Also, RH and RW which represent theradius of HSDPA and Mobile WiMAX networks are set to4 km and 1 km, respectively.

6.2. Hit probability

As mentioned before, we define the hit probability PH asthe probability that the PoA selected by the EIS is the sameas the PoA determined by the full scanning procedure.

RM on PH.

Y. Kim et al. / Computer Networks 55 (2011) 147–158 153

Fig. 5 shows the hit probability under different RM. Asthe simulation time goes, more updates to the EIS are per-formed and thus the EIS can increase PH. Also, Fig. 5 showsthat PH becomes almost 100% after some time, e.g., about60 min when RM P 30 m. This result reveals that the pro-posed vertical handover scheme can work well after someconvergence time. However, if RM is too small, PH has poorperformance because there are few ECSIs satisfying spatiallocality.

From Fig. 6, the effect of velocity VMAX can be observed.Higher velocity indicates that the channel information canbe collected from more diverse locations for a given time.Hence, higher velocity will increase PH. However, it can

Fig. 6. Effect of V

Fig. 7. Effect of

be found that there is no significant difference betweenthe cases of VMAX = 10 m/s and VMAX = 20 m/s.

Fig. 7 illustrates PH as a function of RM. It can be foundthat PH drastically increases as RM increases untilRM = 50 m, since the number of ECSIs satisfying temporaland spatial localities increases. However, since a large RM

may include more ECSIs measured at far away locations,the efficiency of the RSS estimation based on the spatiallocality can be lowered. Therefore, PH decreases slightlywhen RM exceeds 150 m. Also, it can be seen that the tem-poral locality affected by W has minor impact on PH.

Table 2 shows the standard deviation of PH under differ-ent RM. When RM is 100 m, all standard deviations are less

MAX on PH.

W on PH.

Table 2Standard deviation of PH (Unit:%).

Simulation time (min) 100 m 30 m 10 m

10 2.0028 6.1065 2.780920 1.5635 4.7246 2.263230 2.044 3.559 3.212840 1.5951 3.4577 6.022250 1.6364 2.3664 3.831260 1.5239 2.5408 5.966670 1.3166 1.7127 5.440680 1.1738 1.4142 4.472190 1.4757 1.7638 4.2701100 1.6193 1.4298 5.7048

154 Y. Kim et al. / Computer Networks 55 (2011) 147–158

than 2.044%, which reveals that our simulations producequite stable results. On the other hand, for RM of 30 m, itcan be shown that results become stable as the simulationtime goes. When RM is 10 m, larger standard deviations areobtained because sufficient ECSIs are not accumulated atthe EIS. However, it is expected that the simulation resultswill become stable after longer simulation time.

6.3. Vertical handover latency

To derive the expected vertical handover latency, wehave the following notations based on [20]:

� Delay between the MN and the Node B/radio access sta-tion (RAS) is tmr.� Delay between the Node B/RAS and the gateway GPRS

support node (GGSN)/access control router (ACR) is tra.� Delay between the GGSN/ACR and the IS/EIS is tai.

From Fig. 2, the expected vertical handover delay with theexisting IS, DIS

VHO, can be expressed as

DISVHO ¼ 4ðtmr þ tra þ taiÞ þ SH þ SW þ VHOEXEC ; ð6Þ

Fig. 8. Effect of RM on

where SH and SW are the average scanning delay for HSDPAand Mobile WiMAX, respectively. Also, VHOEXEC is the aver-age vertical handover execution time.

In the proposed vertical handover with the EIS, if thereare no ECSIs satisfying the conditions for spatial and tem-poral localities (see line 7 in Algorithm 1), the EIS sendsthe neighbor list response message instead of the target listresponse message to the MN. In other words, in this case,the MN follows the existing vertical handover procedure.On the other hand, the chosen PoA may not be the bestPoA due to the inaccurate estimation from the EIS. Underthis situation (with probability 1 � PH), the MN should per-form vertical handover execution steps multiple times. Ifthere are M PoAs, in such a case, M vertical handover exe-cution steps may be carried out. Consequently, the ex-pected vertical handover delay in the proposed scheme,DEIS

VHO, can be obtained as Eq. (7), where PE is the probabilitythat there is at least one ECSI within RM and it can be ob-tained as Appendix A.

DEISVHO ¼ ð1� PEÞ � DIS

VHO þ PE � 2ðtmr þ tra þ taiÞðþ PH � VHOEXEC þ ð1� PHÞ � ð2VHOEXECÞÞ: ð7Þ

For numerical analysis, we assume that tmr = 60 ms,tra = 2 ms, and tai = 20 ms. Also, SW, SH and VHOEXEC are setto 100 ms, 50 ms, and 2000 ms, respectively [20–22].Fig. 8 shows the vertical handover latency under differentRM. From Fig. 8, it can be observed that the expected verti-cal handover latency decreases as the simulation time goesbecause of the increased PH (as indicated in Fig. 5). Notethat the vertical handover latency of the proposed schemedoes not exceed that of the existing scheme even at thebeginning of simulation. This is because PE is almost 0 atthe beginning of simulation and therefore Eq. (7) can beapproximated as DIS

VHO (i.e., the vertical handover latencyof the existing scheme). In particular, the proposed schemeoutperforms the existing scheme regardless of simulationtime if RM is less than 100 m.

VHO latency.

Fig. 9. Effect of VMAX on VHO latency.

Fig. 10. Effect of W on VHO latency.

Y. Kim et al. / Computer Networks 55 (2011) 147–158 155

Fig. 9 illustrates the effect of VMAX on the vertical hand-over latency when RM is set to 30 m. The expected VHO de-lay of VMAX = 1 m/s is larger than the cases of VMAX = 10 m/sand VMAX = 20 m/s because it has lower PH as shown inFig. 6. However, even when VMAX = 1 m/s, the expected ver-tical handover latency of proposed scheme is less than thedelay of the previous scheme.

On the other hand, Fig. 10 demonstrates the effect of RM.It can be found that the expected vertical handover latencyexceeds that of previous scheme when RM > 250 m. Thiscan be explained as follows. As RM increases, PE convergesto 100% and DEIS

VHO becomes 2(tmr + tra + tai) + PH � VHOEXEC +(1 � PH) � (2VHOEXEC) (see Eq. (7)). On the other hand, asshown in Fig. 7, PH decreases as RM exceeds 150 m. Hence,

DEISVHO increases if RM is larger than 150 m due to the reduced

PH. That is, multiple VHO procedures are performed whenRM is too large, which result in the increased latency.

6.4. Effect of number of PoAs

Because of popularity of WiFi technologies, extensiveworks on opportunistic usage of WiFi have been con-ducted. Therefore, we have conducted additional simula-tions with three PoAs: HSDPA, Mobile WiMAX, and WiFi.

To analyze the effect of WiFi PoA, the WiFi coverage isset to 200 m and two hit probabilities are defined: 1) thePoA first selected by the EIS is the same as the onedetermined by the full scanning procedure, P1st

H and 2)

0 20 40 60 80 100 1200

10

20

30

40

50

60

70

80

90

100

RM (meter)

P H (%

)

W = 60 minutes (P1stH )

W = 60 minutes (P2ndH )

W = 30 minutes (P1stH )

W = 10 minutes (P1stH )

Fig. 11. Effect of RM on PH (with WiFi PoA).

0 20 40 60 80 100 1202000

2200

2400

2600

2800

3000

3200

RM (meter)

Tota

l VH

O d

elay

(ms)

Previous SchemeW = 60 minutesW = 30 minutesW = 10 minutes

Fig. 12. Effect of RM on VHO latency (with WiFi PoA).

156 Y. Kim et al. / Computer Networks 55 (2011) 147–158

the second selected PoA by the EIS is the one by the fullscanning procedure, P2nd

H .Fig. 11 illustrates PH under different W. From Fig. 11, it

can be observed that P2ndH is larger than P1st

H . This is reason-able since P2nd

H is a conditional probability given the failureat the first selection (i.e., the probability of 1� P1st

H ). On theother hand, unlike Fig. 6, P1st

H decreases when RM exceedsaround 40 m. A similar trend can be observed in Fig. 12,which shows the vertical handover delay when the averagescanning delay for WiFi is assumed as 250 ms. That is, theexpected vertical handover latency decreases as RM reachesto a certain point, i.e., about 40 m. On the contrary, the ver-tical handover latency increases drastically when RM ex-

ceeds the point. This is because fine-grained RM isrequired for accurate RSS estimation in WiFi with a smallercoverage. In conclusion, the optimal value of RM should beset depending on the network coverage (i.e., the smallestnetwork coverage).

7. Conclusion

In this paper, we proposed the enhanced informationserver (EIS) and the EIS-based vertical handover procedureto reduce the vertical handover latency by eliminatingtime-consuming channel scanning procedure. From exten-

Y. Kim et al. / Computer Networks 55 (2011) 147–158 157

sive simulation results, it can be shown that the proposedscheme achieves the reduced vertical handover latency un-der the stable EIS and high mobility. It can be also foundthat the channel estimation algorithm should consider net-work characteristics and mobility.

Acknowledgements

This work was supported in part by BLS project fundedby Seoul Metropolitan City (Seoul R& BD Program:WR080951), in part by NRF grants (2009–0064397 andR33–2008-000–10044-0), and in part by ITRC program byNIPA (NIPA-2010-C1090–1011-0004).

HSDPA Network

Mobile WiMAX Network

WR

MR

mrtrat

Fig. 13. Simulatio

Fig. 14. PE under d

Appendix A. Derivation of PE

The key idea of the proposed vertical handover is toeliminate the scanning delay by estimating the wirelesschannel condition. However, if there are no ECSIs for adja-cent areas (see lines 13–14 in Algorithm 1), the MN shouldperform the full scanning procedure, which will increasethe vertical handover latency. Hence, we first derive theprobability PE that there is at least one ECSI within RM.

To derive PE, we consider a network topology wherethere are two PoAs: HSDPA PoA with a larger coverageand Mobile WiMAX PoA with a smaller coverage (seeFig. 13). Definitely, a vertical handover can be triggeredat the Mobile WiMAX area in which MNs can receive the

ait

n topology.

ifferent RM.

158 Y. Kim et al. / Computer Networks 55 (2011) 147–158

signals from the Mobile WiMAX PoA as well as the HSDPAPoA, and they can send ECSIs for both the HSDPA PoA andthe Mobile WiMAX PoA to the EIS. Form Fig. 13, the MobileWiMAX area is pR2

W and thus the probability that an ECSIhas been reported outside the radius RM is given by 1 � pRM

2/pRW2. If the total number of ECSIs is N, then PE can

be obtained as

PE ¼ 1� 1� pR2M

pR2W

!N

: ðA:1Þ

Fig. 14 shows the existence probability PE where analyticaland simulation results are plotted by the dotted and solidlines, respectively. As shown in Fig. 14, PE increases withthe simulation time and PE converges to 100% after sometime. Intuitively, the convergence time for a large RM is fastsince more ECSIs are evaluated for deciding PE. It can alsobe found that analytical and simulation results areconsistent.

References

[1] P. Taaghol, A. Salkintzis, J. Iyer, Seamless integration of mobileWiMAX in 3GPP networks, IEEE Communications Magazine 46 (10)(2008) 74–85.

[2] Q. Nguyen-Vuong, N. Agoulmine, Y. Ghamri-Doudane, Terminal-controlled mobility management in heterogeneous wirelessnetworks, IEEE Communications Magazine 45 (4) (2007) 122–129.

[3] A. Sgora, D. Vergados, Handoff prioritization and decision schemes inwireless cellular networks: a survey, IEEE Communications Surveys& Tutorials 11 (4) (2009) 57–77. 4th Quarter.

[4] IEEE P802.21/D14.0, Draft Standard for Local and Metropolitan AreaNetworks: Media Independent Handover Services, 2008.

[5] C. Huang, C. Lee, and P. Tseng, Multihomed SIP-Based NetworkMobility Using IEEE 802.21 Media Independent Handover, in:Proceedings of the IEEE ICC 2009 (2009).

[6] J. Baek, D. Kim, Y. Suh, E. Hwang, and Y. Chung, Network-InitiatedHandover Based on IEEE 802.21 Framework for QoS ServiceContinuity in UMTS/802.16e Networks, in: Proceedings of the IEEEVTC 2008, May 2008.

[7] M. Kim and S. Lee, ‘‘Load Balancing based on Layer 3 and IEEE 802.21Frameworks in PMIPv6 Networks,” in: Proceedings of the IEEE PIMRC2009, September 2009.

[8] F. Jiadi, J. Hong, and L. Xi, ‘‘User-Adaptive Vertical Handover SchemeBased on MIH for Heterogeneous Wireless Networks,” in:Proceedings of the IEEE WiCom 2009, September 2009.

[9] Q. Mussabbir, W. Yao, Z. Niu, X. Fu, Optimized FMIPv6 using IEEE80221 MIH services in vehicular networks, IEEE Transactions onVehicular Technology 56 (6) (2007) 3397–3407.

[10] H. Huang, J. Wu, and S. Yang, Pre-binding Update Scheme Using802.21 over IEEE 80216e Networks, in: Proceedings of the IEEEWOCN 2008, May 2008.

[11] L. Magagula and H. Chan, ‘‘IEEE 802.21 Optimized Handover Delayfor Proxy Mobile IPV6,” in: Proceedings of the IEEE MILCOM 2008,November 2008.

[12] A. Oliva, A. Banchs, I. Soto, T. Melia, A. Vidal, An overview of IEEE802.21: media-independent handover services, IEEE WirelessCommunications 15 (4) (2008) 96–103.

[13] A. Mishra, M. Shin, W. Arbaugh, An empirical analysis of the IEEE80211 Mac layer handoff process, ACM Computer CommunicationReview 33 (2) (2003) 93–102.

[14] M. Liu, Z. Li, X. Guo, E. Dutkiewicz, and M. Wang, ‘‘SAVA: A NovelSelf-Adaptive Vertical Handoff Algorithm for HeterogeneousWireless Networks,” in: Proceedings of the IEEE GLOBECOM 2006(2006).

[15] H. Bing, C. He, and L. Jiang, ‘‘Performance Analysis of VerticalHandover in a UMTS-WLAN Integrated Network,” in: Proceedings ofthe IEEE PIMRC 2003, September 2003.

[16] B. Jung, M. Choi, H. Youn, and O. Song, ‘‘Vertical Handover based onthe Prediction of Mobility of Mobile Node,” in: Proceedings of theIEEE PERCOM Workshops 2010, March 2010.

[17] COST Action 231, ‘‘Digital Mobile Radio Towards Future GenerationSystems,” Technical Report, European Communities, EUR 18957,1999.

[18] H. Holma, A. Toskala, WCDMA for UMTS, John Wiley, New York,1996.

[19] TTA, TTAS.KO-06.0064, Specifications for 2.3 GHz band PortableInternet Service-Physical Layer.

[20] H. Fathi, R. Prasad, S. Chakraborty, Mobility management for VoIP in3G systems: evaluation of low-latency handoff schemes, IEEEWireless Communications 12 (2) (2005) 96–104.

[21] S. Choi, G. Hwang, T. kwon, A. Lim, and D. Cho, Fast HandoverScheme for Real-Time Downlink Services in IEEE 802.16e BWASystem, in: Proceedings of the IEEE VTC 2005, May 2005.

[22] B. Liu, P. Martins, A. Samhat, and P. Bertin. ‘‘A Layer 2 Scheme forInter-RAT Handover Between UMTS and WiMAX in Tight CouplingArchitecture,” in: Proceedings of the IEEE PIMRC 2008, September2008.

Younghyun Kim received the B.S. and M.S.degrees both in computer engineering fromSoongsil University, Seoul, Korea, in 2005 and2007, respectively. From September 2008, hehas been working toward the Ph.D. degree inthe School of Electrical Engineering, KoreaUniversity, Seoul. From 2007 to 2008, he wasa software engineer at Celrun Inc., Seoul,Korea. His research interests include mobilitymanagement and quality-of-service provisionissues in next-generation wireless/mobilenetworks.

Sangheon Pack received the B.S. (magna cumlaude) and Ph.D. degrees from Seoul NationalUniversity, Seoul, Korea, in 2000 and 2005,respectively, both in computer engineering.Since March 2007, he has been an AssistantProfessor with the School of Electrical Engi-neering, Korea University, Seoul. From 2005 to2006, he was a Postdoctoral Fellow with theBroadband Communications Research Group,University of Waterloo, Waterloo, ON, Canada.From 2002 to 2005, he was a recipient of theKorea Foundation for Advanced Studies

Computer Science and Information Technology Scholarship. In 2003, hewas a Visiting Researcher at Fraunhofer Institute for Open Communica-tion Systems (FOKUS), Berlin, Germany. His research interests include

mobility management, multimedia transmission, and quality-of-serviceprovision issues in next-generation wireless/mobile networks. Dr. Packwas the recipient of a Student Travel Grant Award at the 2003 IFIP Per-sonal Wireless Conference (PWC).

Chung Gu Kang received the B.S. degree inElectrical Engineering from the University ofCalifornia, San Diego in 1987 and the M.S. andPh.D. degrees both in Electrical and ComputerEngineering from the University of California,Irvine, in 1989 and 1993, respectively. SinceMarch 1994, he has been with the departmentof radio communication engineering at theKorea University, Seoul, Republic of Korea,where he is currently a full professor. Hisresearch interests include next generationmobile radio communication system and

broadband wireless networks, with special emphasis on physical layer/medium access control layer design and performance analysis. He is amember of IEEE COMSOC, IT, and VT, and a member of KICS and KITE.

Soonjoon Park received the B.S. and M.S. degrees from Korea University,Seoul, Korea, in 1987 and 1989, respectively, both in electronic engi-neering. He is a chief research engineer with LG Electronics Inc., Anyang,Korea. His research interests include 4G wireless/mobile communicationsand MIMO systems.