Page 1

Signal Threshold Adaptation for Vertical

Handoff in Heterogeneous Wireless Networks

Ahmed H. Zahran

Department of Electrical and Computer Engineering

University of Toronto

10 King’s College Road

Toronto, Ontario, M5S 3G4, Canada

Email: zahran@comm.utoronto.ca

Tel: 1-416-978-4829, Fax: 1-416-978-4425

Ben Liang

Department of Electrical and Computer Engineering

University of Toronto

10 King’s College Road

Toronto, Ontario, M5S 3G4, Canada

Email: liang@comm.utoronto.ca

Tel: 1-416-946-8614, Fax: 1-416-978-4425

Aladin Saleh

Bell Canada/ Wireless Technology

Floor 5N - 5099 Creekbank Road

Mississauga, Ontario, L4W 5N2, Canada

Email: aladdin.saleh@bell.ca

Tel: 1-905-282-3264, Fax: 1-905-282-3337

*1

1To appear in ACM/Springer Mobile Networks and Applications (MONET) journal, this article

is the extended version of a paper presented in IFIP Networking 2005.

1

Page 2

Abstract

Theconvergenceofheterogeneouswirelessaccesstechnologieshasbeen

envisioned to characterize the next generation wireless networks. In such

converged systems, the seamless and efficient handoff between different

access technologies (vertical handoff) is essential and remains a challeng-

ing problem. The heterogeneous co-existence of access technologies with

largely different characteristics results in handoff asymmetry that differs

from the traditional intra-network handoff (horizontal handoff) problem. In

the case where one network is preferred, the vertical handoff decision should

be carefully executed, based on the wireless channel state, network layer

characteristics, as well as application requirements. In this paper, we study

the performance of vertical handoff using the integration of 3G cellular and

wireless local area networks as an example. In particular, we investigate

the effect of an application-based signal strength threshold on an adaptive

preferred-network lifetime-based handoff strategy, in terms of the signalling

load, available bandwidth, and packet delay for an inter-network roaming

mobile. We present an analytical framework to evaluate the converged sys-

tem performance, which is validated by computer simulation. We show how

the proposed analytical model can be used to provide design guidelines for

the optimization of vertical handoff in the next generation integrated wire-

less networks.

Keywords: heterogeneous wireless networks, seamless integration, vertical

handoff, application signal strength threshold, 3G cellular, wireless LAN

1 Introduction

Wireless technologies are evolving toward broadband information access across

multiple networking platforms, in order to provide ubiquitous availability of mul-

timedia applications. Recent trends indicate that wide-area cellular networks

based on the 3G standards and wireless Local Area Networks (WLANs) will co-

exist to offer multimedia services to end users. These two wireless access tech-

nologies have characteristics that perfectly complement each other. By strategi-

cally combining these technologies, a converged system can provide both univer-

sal coverage and broadband access. Therefore, the integration of heterogeneous

networks is expected to become a main focus in the development toward the next

generation wireless networks [1–3].

2

Page 3

MO

MI

AP

Horizontal Handoff

Vertical Handoff

AP

AP

Base Station

Access Point

BS

BS

Figure 1: Mobile handoff in heterogeneous wireless system

Mobility management is a main challenge in the converged network. It ad-

dresses two main problems: location management and handoff management [4,5].

Location management tracks the Mobile Terminals (MT) for successful informa-

tion delivery. For this purpose, Mobile IP (MIP) enables seamless roaming and

is expected to be the main engine for location management in the next generation

networks. Handoffmanagementmaintainstheactiveconnectionsforroamingmo-

bile terminals as they change their point of attachment to the network. Handoff

management is the main concern of this paper.

In the converged network, both intra-technology handoff and inter-technology

handoff take place as illustrated in Figure 1. Intra-technology handoff is the tradi-

tional Horizontal Handoff (HHO) process in which the mobile terminal hands-off

between two Access Points (AP) or Base Stations (BS) using the same access

technology. On the other hand, inter-technology handoff, or Vertical Handoff

(VHO), occurs when the MT roams between different access technologies. The

main distinction between VHO and HHO is symmetry. While HHO is a sym-

metric process, VHO is an asymmetric process in which the MT moves between

two different networks with different characteristics. This introduces the concept

of a preferred network, which is usually the underlay WLAN that provides better

throughput performance at lower cost, even if both networks are available and in

good condition for the user.

There are two main scenarios in VHO: moving out of the preferred network

(MO) and moving into the preferred network (MI) [6]. In the converged model,

3

Page 4

it is highly desirable to associate the MT with the preferred network, as long as

the preferred network satisfies the user application. This can improve the resource

utilization of both access networks, as well as improving the user perceived QoS.

Furthermore, this handoff should be seamless with minimum user intervention,

while dynamically adapting to the wireless channel state, network layer charac-

teristics, and application requirements.

In this work, we present an adaptive lifetime-based VHO (ALIVE-HO) algo-

rithm which takes into consideration the wireless signal strength, handoff latency,

and application QoS and delay tolerance. It can satisfy the system handoff sig-

nalling load, as well as different application requirements by the tuning of an

application-based signal strength threshold (ASST). We further propose an ana-

lytical model to evaluate the performance of adaptive VHO. This analytical frame-

work is then applied to show how the VHO decision and the ASST choice can be

optimized based on multiple conflicting criteria including vertical handoff signal-

ing, user available bandwidth, and encountered packet delay. Hence, the optimal

ASST value is determined for different QoS requirements.

The rest of this paper is organized as follows. Section 2 provides an overview

for the handoff algorithms in wireless heterogeneous networks and the related

literature work. In Section 3, we present a vertical handoff algorithm that incor-

porates cross-layer adaptation to terminal mobility, channel state, and application

demand. In Section 4, we propose an analytical framework to study the effect

of cross-layer adaptation. Numerical and simulation results are provided in Sec-

tion 5, where we show how the ASST can be tuned to optimize VHO decision.

Concluding remarks are presented in Section 6.

2 Related Work

The traditional HHO problem has been studied extensively in the past. Several

approaches have been considered in cellular networks using the Received Sig-

nal Strength (RSS) as an indicator for service availability from a certain point of

attachment. Additionally, several handoff initiation strategies have been defined

based on the comparison between the current attachment point RSS and that of

the candidate attachment points as shown in [7]:

• RSS: handoff takes place if the candidate attachment point RSS is higher

than the current attachment point RSS (RSSnew> RSScur).

4

Page 5

• RSS plus threshold: handoff takes place if the candidate attachment point

RSS is higher than the current attachment point RSS and the current attach-

ment point RSS is less than a pre-defined threshold T (RSSnew> RSScur

and RSScur< T).

• RSS plus hysteresis: handoff takes place if the candidate attachment point

RSS is higher than the current attachment point RSS with a pre-defined

hysteresis margin H. (RSSnew> RSScur+ H).

• A dwell timer can be added to any of the above algorithms. In this case,

the timer is started when one of the above conditions is satisfied, and the

MT performs a handoff if the condition is satisfied for the entire dwell timer

interval.

In VHO, the RSSs are incomparable due to VHO’s asymmetrical nature. How-

ever, they can be used to determine the availability as well as the condition of

different networks. If the MI decision is based only on the preferred network

availability, the MT should start the MI process as it discovers the WLAN. In ad-

dition, if more than one WLAN APs are available, the MT should associate itself

with the one having the strongest RSS as it does in HHO2. When the MT is asso-

ciated with the preferred network, it enjoys all the preferred network advantages

before moving out. Therefore, in the ideal MO scenario, the MT performs no

more than one handoff at the WLAN edge when the network is expected to be

unavailable. This ideal MO decision usually cannot be achieved. Thus, the main

design requirements of a VHO algorithm are

• minimizing the number of unnecessary handoffs to avoid overloading the

network with signaling traffic,

• maximizing the underlay network utilization,

• providing active application with the required degree of QoS,

• prioritizing handoff to the underlay network over MO to the overlay net-

work,

• avoiding MI to a congested network, and

2If other criteria such as available bandwidth are considered, the MT may not move instanta-

neously to a WLAN, but may consider other factors such as QoS, user preference, cost, and power

consumption.

5

Page 6

• keeping fast users connected to the overlay network.

As far as we are aware, there exist very few works dealing with VHO beyond

simple extensions to the common techniques for HHO. Three main directions for

VHO algorithms are recorded in the literature.

The first approach is based on the traditional strategies of using the RSS that

may be combined with other parameters such as network loading. In [8], Hatami

et. al use the dwelling timer as a handoff initiation criterion to increase the WLAN

utilization. They combine simulation andanalysis to showthat associatingthe MT

with the WLAN for the longest possible duration improves user throughput even

during the transition period in which the RSS oscillates around the receiver sen-

sitivity level. However, they did not define a clear mechanism for choosing the

dwelling timer value. In [9], Ylianttila et al. present an algorithm to compute an

optimization policy for the dwelling timer according to the available data rates in

both networks. The main result is that the optimal value for the dwelling timer is

highly dependent on the difference between the available rates in both networks.

In [10], Ylianttila et al. extend the same analytical framework of [8] to include

multiple radio network environments. Their main results show that the handoff

delay effect seems to be dominant even with the dwelling timer optimal choice as

in [9]. In [11], Park et al. propose using a similar dwelling timer-typed approach

for both MI and MO by performing the VHO if a specific number of the received

beacons exceed or go below a predefined MI or MO threshold respectively. Addi-

tionally, the authors propose adapting the performance of their algorithm based on

the application requirements by using two different numbers of beacons for real-

time and non-real-time services. Although the dwelling timer approach seems to

be an attractive approach for the VHO in order to maximize the underlay net-

work usage, the proper dwelling timer choice is a critical decision because a large

dwelling timer may result in undesirable service interruption periods for real-time

applications. In our approach, interruptions are avoided by the proper choice of

an application specific signal threshold to satisfy the requirements of the applica-

tions.

The second approach uses artificial intelligence techniques combining several

parameters such as network conditions and MT mobility in the handoff decision.

In [12], Ylianttila et al. present a general framework for the vertical handoff pro-

cess based on fuzzy logic and neural networks. In [13], Pahlavan et al. present a

neural network-based approach to detect signal decay and making handoff deci-

sion. In [14], Majlesi and Khalaj present a fuzzy logic based adaptive algorithm

that varies the hysteresis margin and averaging window size based on MT velocity

6

Page 7

and WLAN traffic. It is worth mentioning that some of these artificial intelligence

based algorithms are complex and may be difficult to implement in practical sys-

tems. It is possible to extend our work to include improvement using similar

artificial intelligence approaches. However, this is outside the scope of this paper

and will be left for future work.

The third direction combines several metrics such as access cost, power con-

sumption, and bandwidth in a cost function estimated for the available access

networks, which is then used in the MT handoff decision. Wang et al. intro-

duce the policy enabled handoff in [15], which was followed by several papers on

similar approaches. In [15], the authors proposes policies considering different

parameters such as monetary cost, power consumption, network available band-

width, and other parameters that differ among different heterogeneous networks.

For each policy, a cost function is defined as a weighted sum of normalized policy

parameters. These weights varies according to user preferences and the MT status

(e.g., power reserve). In this scheme, the MT periodically compares the cost of

different networks and then is handed off to the one with the minimum cost. Addi-

tionally, the authors introduce the programming model and software architecture

of their solution. In [16], Zhu and McNair present cost functions that account for

the dynamic values that are inherent to vertical handoff and incorporate a network

elimination factor to potentially reduce delay and processing power in the handoff

calculation. They introduce two cost-based policies for VHO decision consider-

ing the available bandwidth and RSS of the available networks. The collective

handoff policy estimates one cost for all the services, while the prioritized multi-

network handoff policy estimates the cost for each service independently. Also,

Chen et. al. [17] introduce a smart decision model using a handoff control center

module in the MT. This module monitors the available interfaces and the system

resources to collect information required for the handoff decision. This decision

is based on a score function that considers the usage expenses, link capacity, and

power consumption for the available access technologies. The MT uses the net-

work that achieves the largest score. One main difficulty of the cost approach is its

dependence on some parameters that are difficult to estimate, especially in large

cellular networks, such as the available bandwidth, the channel condition, and the

network user density, all of which change dynamically.

In ALIVE-HO, we adopt the first approach of using the RSS as a unique input

for the algorithm, to estimate the duration through which the WLAN usage will

be beneficial for the active applications.

7

Page 8

3 Application Life-Time Adaptation

3.1System Model

We study the overlapping of 3G cellular and WLAN networks. The cellular net-

work is assumed to provide universal coverage, while WLAN availability is indi-

cated by the presence of the WLAN beacons [13] that are periodically transmitted

by the WLAN APs. Mobile-IP is assumed for mobility management.

The European Telecommunication Standards Institute (ETSI) proposed two

mobility management architectures for the next generation wireless networks:

tightly coupled and loosely coupled [18]. In tight coupling, a WLAN gateway

emulates the functions of a cellular Radio Network (RN), while in loose coupling

the WLAN gateway helps authenticate the users, obtains their service profile at

the session beginning only, and then uses its own resources to route the subscriber

data. The latter approach is preferred for several reasons, including flexibility that

enables the integration of the third party wireless Internet service providers and

independent implementation of WLAN and 3G networks.

We assume that WLAN hotspots implement loosely coupled connection with

the 3G network using WLAN gateways. These gateways perform several tasks

including serving as Mobile-IP agents and possibly providing QoS in the form of

multiple service classes defined within the WLAN. However, it is worth mention-

ing that end-to-end QoS support requires other mechanisms such as differentiated

services to be implemented over the entire network path. The details of such im-

plementation is unimportant to the proposed VHO algorithm and mathematical

analysis. We are mainly concerned about the resultant VHO delay values.

The MT is equipped with dual interfaces that allow it to communicate with

both networks. However, since Mobile-IP provides only one IP tunnel, the MT

can connect to only one network at a time. In addition, multi-interface mobility

client software is installed on the MT. This software performs Mobile-IP signaling

withtheforeignandhomeagents. Itperiodicallyscanstheavailableinterfacesand

measures the observed RSS. Then it intelligently selects the best access network

according to the predefined VHO algorithm.

Within the WLAN, a log-linear path loss channel propagation model with

shadow fading is used [19]. The RSS is expressed in dBm as

RSS = PT− L − 10nlog(d) + f(μ,σ) ,

where PTis the transmitted power, L is a constant power loss, n is the path loss

exponent and usually has values between 2 - 4, d represents the distance between

(1)

8

Page 9

the MT and the WLAN AP, and f(μ,σ) represents shadow fading which is mod-

elled as Gaussian with mean μ = 0 and standard deviation σ with values between

6-12 dB depending on the environment. We assume that when the RSS is below

a certain interface sensitivity level, α, the MT is unable to communicate with the

AP.

3.2 Adaptive Preferred-Network Life-Time Vertical Handoff

(ALIVE-HO)

For the MO scenario when the MT is within a WLAN, we use the RSS to estimate

the expected duration after which the MT is unable to maintain its connection with

the WLAN. We take into consideration the handoff delay due to MIP tunnelling,

authentication, and service initiation. We further consider an Application Signal

Strength Threshold (ASST), which is the required level of RSS for the active

application to perform satisfactorily.

The ASST is an application dependent parameter which represents a compos-

ite of the channel bit error rate, application error resilience, and application QoS

requirements. We present here how the ASST can be incorporated into the VHO

decision. We further discuss in the next section how the ASST can be adjusted to

optimize the overall system performance.

In discrete time, the RSS is expressed as

RSS[k] = μRSS[k] + N[k] ,

(2)

where k is the time index, μRSS[k] = PT−L−10nlog(d[k]), and N[k] = f(μ,σ).

The averaged RSS, RSS[k], can be estimated using a moving average

RSS[k] =

1

Wav

Wav−1

?

i=0

RSS[k − i] .

(3)

The RSS rate of change, S[k], can be obtained by

S[k] =M1[k] − M2[k]

WSTS

,

(4)

where

M1[k] =

2

WS

WS

2−1

?

i=0

RSS[k − WS+ 1 + i] ,

(5)

9

Page 10

Slope Estimation

Averaging

Lifetime Estimation

MO Handoff Algorithm

ASST

MO?

RSS

Slope

average

Compare Available RSS’s

NO

YES

Handoff to another network

RSS Measurement

Figure 2: MO handoff algorithm

M2[k] =

2

WS

WS−1

?

i=WS

2

RSS[k − WS+ 1 + i] ,

(6)

and WSand TSdenote the slope estimator window size and the RSS sampling

interval respectively.

Then, we estimate the MT lifetime within the WLAN, EL[k], as follows.

EL[k] =RSS[k] − γ

S[k]

,

(7)

where γ denotes the ASST. Thus, EL[k] represents the application specific time

period in which the WLAN is likely to remain usable to the MT. Figure 2 depicts

the MO scenario block diagram.

Once the VHO decision is taken, the available cellular RSS from different

base stations are compared to determine the base station with which the MT will

associate itself.

Based on the measured and estimated parameters, the MT will initiate the MO

handoff at time k if the averaged received signal strength is less or equal to a

predefined MO threshold, MOTWLAN, and the estimated lifetime is less than or

equal to the handoff delay threshold, THO. The first condition prevents unnec-

essary handoffs near the access point resulting from short lifetime estimate due

10

Page 11

to fast signal decay; additionally, the lifetime part tunes the handoff instant ac-

cording to users mobility to benefit from WLAN resources. The MOTWLANis

usually chosen to be a few dB above the wireless interface sensitivity. THOcan be

set to the expected handoff delay between the two access technologies. This delay

includes several signaling delay components such as discovery delay, authentica-

tion delay, and registration delay. These delays vary depending on the adopted

approach for location management, whether it is Mobile-IP [2] or an end-to-end

approach [20,21].

Clearly, the window sizes have significant effect on the lifetime-based algo-

rithm performance. In general, a larger window size results in better estimation

but also larger delay in handoff performance [7]. Using variable window sizes that

adapt to the MT mobility can improve handoff performance. For example, Wav

and WScan be determined as follows.

Wav= max

?

10,

?Dav

V TS

?Ds

V TS

??

,

(8)

WS= 2 ∗ max

?

50,

??

,

(9)

where Davand Dsrepresent the averaging and slope distance windows respec-

tively, ?·? represents the greatest lower integer function, and V is the MT velocity

away from the AP, which can be obtained by many velocity estimators proposed

in the literature, for example [22]. Hence, better estimates due to larger windows

are obtained for slower users; which safely improves the handoff performance and

enables maximizing the benifits of WLANs.

In the MI scenario, several factors need to be considered. The main one is the

WLAN availability, which can be determined by the WLAN RSS. In addition, the

QoS, specified in terms of the available bandwidth, is a key factor in the handoff

decision. Other factors such as security, user preference can be considered. In this

work, we consider a simplified model where the MT performs MI to the WLAN

if RSS[k] > MITWLANand the available bandwidth is greater than the required

bandwidth. The available bandwidth can be estimated based on observing the

Network allocation vector [23] or can be incorporated within the AP beacon or

MIP foreign agent advertisement to decrease the delay between the WLAN dis-

covery and the MI initiation. In our simulation and analysis, we assume that the

WLAN is always in good condition, so that the MT always perform an MI after

an unnecessary MO.

11

Page 12

4 Performance Analysis

Inthissection, weprovideananalyticalframeworkforevaluatingtheperformance

of the cross-layer ALIVE-HO algorithm.

4.1Transition probabilities

The calculation of the transition probabilities is based on recursive computation of

the handoff probabilities similar to [24]. In the integrated heterogeneous network-

ing model, the following probabilities are required for handoff algorithm analysis.

• PW[k]: Pr{MT is associated with the WLAN at instant k}

• PC[k]: Pr{MT is associated with the 3G network at instant k}

• PW|C[k]: Pr{MT associates itself with the WLAN at instant k given that it

is associated with the cellular network at instant k-1}

• PC|W[k]: Pr{MT associates itself with the 3G network at instant k given

that it is associated with the WLAN at instant k-1}

In our model, the MT is assumed to be attached to the WLAN at the beginning;

hence PW[0] = 1 and PC[0] = 0. PW[k] and PC[k] can be calculated recursively

as follows.

PW[k + 1] = PW|C[k + 1]PC[k] +

?

?

1 − PC|W[k + 1]

?

?

PW[k],

(10)

PC[k + 1] = PC|W[k + 1]PW[k] +

The Conditional probabilities PC|W[k + 1] and PW|C[k + 1] depend on the

handoff algorithm initiation strategy. For the proposed cross-layer ALIVE-HO

algorithm, PC|W[k + 1] is determined by

?

where W[k] represents the event that the MT is associated with the WLAN at time

k. In practice, WLANs are designed for low mobility users. The lifetime part of

the MO condition becomes more significant for low mobility users. Hence the

1 − PW|C[k + 1]PC[k].

(11)

PC|W[k+1] = Pr

RSS[k + 1] < MOTWLAN,EL[k + 1] < THO|W[k]

?

(12)

12

Page 13

MO condition can be reduced to EL[k] < THO. Consequently, one can determine

PC|W[k + 1] as follows:

PC|W[k + 1] = Pr{EL[k + 1] < THO|EL[k] > THO}

= Pr

RSS[k + 1] − THOS[k + 1] < γ|RSS[k] − THOS[k] > γ

Let Z[k] = RSS[k] − THOS[k]. Then we have

PC|W[k + 1] = Pr{Z[k + 1] < γ|Z[k] > γ}

=

(13)

??

.

(14)

Pr{Z[k + 1] < γ,Z[k] > γ}

Pr{Z[k] > γ}

.

Clearly, since RSS[k] is a Gaussian process, the processes RSS[k] and S[k] are

Gaussian, and hence Z[k] is Gaussian too. Let its mean be μZ[k] and standard

deviation be σZ[k]. It can be shown that

μZ[k] = μRSS[k] − THOμS[k]

(15)

where

μRSS[k] = μRSS[k] +

1

Wav

Wav−1

?

i=0

10nlog(1 −iV TS

d[k]),

and

μS[k] =E{M1[k]} − E{M2[k]}

WSTS

,

and furthermore

σ2

z[k] = σ2

RSS[k] + T2

HOσ2

S[k] +4THOσ2

RSS

?h=Wav−1

W2

h=0

STSW2

(Wav− |h|)

av

,

where

σ2

RSS[k] =

σ2

Wav,

and

σ2

S[k] =

4σ2

SWav)2

(TSW2

?

WavWS+

Wav−1

?

h=1

(Wav− |h|)(2WS− 6|h|)

?

.

Additionally, Z[k] and Z[k − 1] are jointly Gaussian with correlation coefficient

ρZ[k],Z[k−1]asderivedintheAppendix, whichdefinestheirjointPDFfZ[k]Z[k−1](z1,z2)

[25].

13

Page 14

Then we can compute PC|W[k + 1] by

PC|W[k + 1] =

?γ

−∞

?∞

γfZ[k+1]Z[k](z1,z2)dz1dz2

?

Q

γ−μZ[k][k]

σZ[k][k]

?

,

where Q(x) is the complementary error function. Similarly, PW|C[k + 1] can be

determined by

PW|C[k + 1] = Pr{RSS[k + 1] > MIT|RSS[k] < MIT}

=

Pr{RSS[k] < MIT}

where, similartothe(Z[k+1],Z[k])tuple, the(RSS[k+1],RSS[k])tupleisjointly

Gaussian. These transition probabilities are used to calculate the performance

metrics in the next subsections.

(16)

Pr{RSS[k + 1] > MIT,RSS[k] < MIT}

.

(17)

4.2Handoff Probabilities and the Number of Handoffs

The number of handoffs has major impact on the signaling traffic, which may

overload the network resulting in degradation in the overall performance. The

number of handoffs, denoted NHO, is defined as the sum of MOs and MIs between

WLAN and 3G network as the MT roams across the network boundary. Hence, it

is a random variable that depends on the instantaneous move out/in probabilities,

which can be calculated by

PMO[k + 1] = PC|W[k + 1]PW[k],

PMI[k + 1] = PW|C[k + 1]PC[k].

(18)

(19)

The MT movement between the two networks can be modeled by a two-state

non-homogeneous Markov chain. Each state represents the network with which

theMTisassociated. ThetransitionprobabilitiesarePMO[k]andPMI[k]asshown

in Figure 3. Hence, by using binary impulse rewards for the handoff transition as

shown in [26], we calculate the average accumulated rewards for MO and MI

transitions, which are equivalent to the expected number of MOs, NMO, and the

expected number MIs, NMI, respectively. Hence, the expected number of hand-

offs is

E{NHO} = E{NMO} + E{NMI}

kmax

?

(20)

=

k=1

(PMO[k] + PMI[k]) .

(21)

14

Page 15

MT

using

WLAN

MT

using

3G

MI

MO

P [k]

P [k]

Figure 3: VHO Markov chain model

4.3Available Bandwidth

The available bandwidth to the MT depends on the proportion of time that the MT

stays in the WLAN and the 3G network, as well as the WLAN state when the MT

is connected to the WLAN. To the MT, the WLAN is in one of two states: WLAN

Up and WLAN Down. The WLAN Up state represents the event that the WLAN

signal received at the MT is above the sensitivity level α. WLAN Down is the

reverse case. Let p[k] be the probability that the WLAN is in the Up state at time

k. Clearly

p[k] = Pr{RSS[k] > α}

= Q

?α − μ[k]

σ

?

.

In the adopted handoff algorithm, the MO distance varies; consequently the

captured WLAN Up durations does too. For the rest of the analysis, we are inter-

ested in evaluating the system performance during the transition region, which is

defined as the range of distance between the point when the RSS starts to oscillate

around the interface sensitivity and the WLAN edge. The transition region de-

termination is equivalent to a long-standing complex level crossing problem that

is analytically tractable only for a few simple cases and is usually solved numer-

ically for complex cases. Here, we obtained the transition region starting point,

denoted kstart, from rough estimates based on simulation results.

Then, the WLAN efficiency, ζLT, defined as the percentage of the WLAN up

duration over the MT lifetime in the WLAN, can be estimated as

ζLT=

kmax

?

k=kstart

PMO[k]

?k

h=1p[h]

k

,

15

Page 16

where PMOis a scaled version of PMOto represent a valid PDF within interval

[1,kmax], and kmaxrepresents the time index at which the MT reaches the WLAN

edge and is determined by the planed coverage area.

Hence, the MT available bandwidth, BWAv, assuming RW and RC as the

effective data rates in WLAN and cellular networks respectively, can be computed

as

BWAv

=

ζLTRW(kMO− kstart) + RC(kmax− kMO)

(kmax− kstart)

,

(22)

where kMOdenotes the average time to MO.

4.4Packet Delay

In addition to the MT available bandwidth, RSS degradation in the transition re-

gion impacts on the head of line (HoL) packet delay probability. To study this, we

assume a threshold, θDfor packet delay in the current hop as a part of the end-

to-end delay budget for the real-time application packet from the source to the

destination. A packet is considered excessively delayed3if its HoL delay exceeds

θD. Consequently, the average packet delay probability, D, can be estimated as

?kMO

where PD[k] represents the probability that a packet will be excessively delayed,

which is equal to the probability of WLAN Down runs whose duration is equal

to the delay threshold. Here we have performed an approximation by using kMO,

instead of using kMOand then applying conditional expectation. As shown in the

next section, this approximation produces accurate results over a wide range of

system parameters.

D =

k=kstartPD[k]

(kMO− kstart+ 1),

5 Numerical Results and the Optimization of ASST

5.1Simulation Model

In addition to the above analysis, we have simulated the VHO algorithms using

MATLAB. Table 1 shows the simulation parameter values. The WLAN parame-

ters are used as in [14], which are suitable to model outdoor suburban (e.g. with

3Note that this does not necessarily mean that the packet is lost.

16

Page 17

Table 1: Simulation parameter values

Parameter Value

PT

100 mWatt

n 3.3

σ

7 dB

S 28.7 dB

Dav

0.5 m

Ds

5 m

α

-90 dBm

Parameter

TS

MOTWLAN -85 dBm

MITWLAN

Thandoff

RW

RC

Value

0.01 sec

-80 dBm

1 sec

6 Mbps

0.6 Mbps

tree and low buildings along the road side which is similar to the characteristics

of the commercial WLAN services) and indoor locations with wide areas (such

as hotel lobbies and campuses). These parameters result in a WLAN coverage of

100 meters approximately.

The data rates shown in the table are used for performance evaluation only

and have no effect on the handoff decision. Currently, IEEE802.11b WLANs

are widely deployed and support rates vary from 11 Mbps to 1 Mbps depending

on the distance between the MT and the WLAN AP. On the other hand, cellular

service providers are still deploying their first phase of the 3G network that sup-

ports rates up to 144 Kbps for CDMA1x4. In the future, IEEE802.11a [27] and

CDMA20001x-EV [28] are expected to be widely deployed. The former support

rates that vary from 54 Mbps to 6 Mbps, while the latter supports a peak rate of

2.4 Mbps on the forward link with an average throughput of 600 kbps. Hence,

these values show that the service rate of WLANs is generally approximately one

order of magnitude larger than that of the cellular network.

Additionally, asimplemobilitymodelisassumedinwhichaMTmovingaway

from the WLAN access point in a straight line at a constant speed V . As shown in

[29], this model is suitable for evaluating the performance of signal strength based

algorithms with log-normally distributed shadow fading environments as in our

case. Additionally, the proposed algorithm will function with any mobility pattern

since the algorithm dynamically adapts to the MT velocity, and the algorithm time

resolution is sufficient to track mobility pattern variation, especially for low speed

MTs.

4Effective date rates are much lower than this value.

17

Page 18

0.51 1.52 2.5

Velocity m/s

3 3.54 4.55

0

10

20

30

40

50

60

Average Number of HO

ALIVE−HO analysis

ALIVE−HO sim

LT Analysis

LT sim

HY sim

Figure 4: Number of handoffs (γ= -90 dBm).

5.2 Performance Comparison

We compare the performance of ALIVE-HO with traditional hysteresis VHO,

which is used in [2]. In hysteresis based algorithms, there are two different thresh-

olds MITWLANand MOTWLANfor the MI and MO respectively. The MT per-

forms a MI if the RSS[k] is larger than MITWLANand performs a MO if RSS[k]

is smaller than the predefined MOTWLAN. Usually, MITWLANis chosen larger

than MOTWLANto decrease the number of unnecessary handoffs known as ping-

pong effect. We also consider a non-adaptive WLAN lifetime based VHO algo-

rithm, where the lifetime estimation does not adapt to MT mobility or application

demand, and hence a fixed RSS averaging window of ten samples is used.

Figures 4, 5, and 6 illustrate the number of handoffs, available bandwidth, and

the packet delay probability for the VHO handoff algorithms. In all figures, HY

denotes hysteresis VHO, LT denotes non-adaptive lifetime VHO, and ALIVE-HO

denotes the adaptive lifetime VHO. All figures show good match between analysis

and simulation.

Figure 4 shows that the introduction of the adaptive lifetime approach to the

traditional HY VHO algorithm results in significant decrease of the number of

unnecessary handoffs. Figure 5 demonstrates the improvement on the available

bandwidth by using adaptive lifetime estimation. Clearly, from a pure bandwidth

point of view, it is preferable for the MT to perform MO handoff only once at the

18

Page 19

0.51 1.52 2.5

Velocity m/s

3 3.54 4.55

15

15.5

16

16.5

17

17.5

18

18.5

19

19.5

20

Available Bandwidth (Mbps)

ALIVE−HO analysis

ALIVE−HO sim

LT Analysis

LT sim

HY sim

Figure 5: Available bandwidth (γ = -90 dBm).

0.51 1.52 2.5

Velocity m/s

3 3.54 4.55

0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

HoL Packet delay probability

ALIVE−HO analysis

ALIVE−HO sim

LT Analysis

LT sim

HY sim

Figure 6: HoL packet delay rate (γ = -90 dBm, θD=30 ms).

19

Page 20

0.51 1.52 2.5

Velocity m/s

3 3.54 4.55

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

Number of Vertical Handoffs

Analysis ASST=−90

Sim ASST=−90

Analysis ASST=−89

Sim ASST=−89

Analysis ASST=−88

Sim ASST=−88

Analysis ASST=−87

Sim ASST=−87

Figure 7: Number of handoffs vs. ASST.

WLAN edge, even though the RSS can temporarily go below the MT sensitivity

level in the transition region. However, a drawback of increasing the lifetime of

the MT within the WLAN is increasing the packet delay resulting from channel

condition degradation. As shown in Figure 6, the packet delay probability using

the adaptive approach can be much higher than that when the traditional hysteresis

algorithm is used. This may be critical if the MT is running real-time application.

However, by properly tuning the ASST as shown in the next subsection, ALIVE-

HO can adapt to the active real-time application requirements in the MT.

5.3 Application Signal Strength Threshold Adaptation

Figures 7, 8, and 9 illustrate the effect of the ASST on the number of handoffs,

available bandwidth, and packet delay probability. They show that the number of

handoffs decreases when the ASST is reduced, since reducing the ASST allows

the MT to remain in the WLAN for a longer duration. For the same reason, the

available bandwidth to the MT increases when the ASST is reduced. However,

at the same time, the packet delay probability is increased, since signal outage is

more severe near the WLAN edge. Hence, there is a clear trade off among the

handoff signaling load, available bandwidth, and packet delay.

Clearly, the ASST should not depend on the application QoS alone. Rather, it

20

Page 21

0.51 1.52 2.5

Velocity m/s

3 3.54 4.55

15

15.5

16

16.5

17

17.5

18

18.5

19

19.5

20

Available Bandwidth (Mbps)

Analysis ASST=−90

Sim ASST=−90

Analysis ASST=−89

Sim ASST=−89

Analysis ASST=−88

Sim ASST=−88

Analysis ASST=−87

Sim ASST=−87

Figure 8: Available bandwidth vs. ASST.

0.51 1.52 2.5

Velocity m/s

3 3.54 4.55

0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

HoL Packet delay probability

Analysis ASST=−90

Sim ASST=−90

Analysis ASST=−89

Sim ASST=−89

Analysis ASST=−88

Sim ASST=−88

Analysis ASST=−87

Sim ASST=−87

Figure 9: HoL packet delay rate vs. ASST (θD= 30ms).

21

Page 22

−90−89.5−89−88.5−88−87.5−87

5

6

7

8

9

10

11

12

13

ASST (dBm)

Total Cost

θd = 40 msec (Anal)

θd = 40 msec (Sim)

θd = 50 msec (Anal)

θd = 50 msec (Sim)

θd = 60 msec (Anal)

θd = 60 msec (Sim)

Figure 10: Total cost vs. ASST.

can be optimally tuned based on the various conflicting criteria of VHO. Likewise,

the optimal VHO decision can be made adaptive to the RSS variation, network

delay characteristics, and application QoS demands, through a properly chosen

ASST value. The proposed analytical framework provides a means to carry out

this optimization.

As an example, a possible cost function to aggregate the multiple VHO criteria

may be

Ctotal=cHE{NHO} + cDD

BWav

,

where cHrepresents the signaling cost per handoff, cDrepresents the penalty fac-

tor for packet delay, and Ctotalis normalized to cost per Mbps of data bandwidth5.

Figure 10 plots Ctotalover different ASST values, for V = 2, cH= 100, and

cD = 10000, where each curve represents a delay threshold value of 40 ms, 50

ms, and 60 ms, respectively. Clearly, the optimal ASST increases as the delay

threshold decreases. In particular, when θD = 40ms, an ASST of -87.5 dBm

strikes the optimal balance to minimize the total cost, but when θD= 60ms, the

optimal ASST is -89.5 dBm.

5We emphasize here that this is only one of many possible cost functions, whose suitability

depends on practical application goals and system constraints.

22

Page 23

20 304050 6070 80

−90

−89.5

−89

−88.5

−88

−87.5

−87

θd (msec)

Optimal ASST (dBm)

V = 1.0 m/s

V = 1.5 m/s

V = 2.0 m/s

V = 2.5 m/s

Figure 11: Optimal ASST vs. delay budget threshold and velocity (cH = 100,

cD= 100000).

5.4 Optimal Application Signal Strength Threshold Values

To further study how the optimal ASST is affected by the system parameters, in

Figures 11 and 12, we present numerical analysis results obtained for the optimal

ASST values given various system parameters.

Figure 11 shows the optimal ASST over different delay budgets and MT ve-

locities, where cH= 100 and cD= 100000. It is clear that as the delay constraint

is relaxed, the optimal ASST value decreases approximately linearly in dB (ex-

ponentially in linear scale), and consequently, the MT WLAN lifetime increases.

Additionally, as the MT velocity increases, the optimal ASST decreases. For ex-

ample, if two MTs, MTa and MTb moving at 1.5 m/s and 2 m/s respectively, are

running a real-time application with a 40 ms delay budget in the WLAN, MTa

should set its ASST to -87 dBm while MTb should set it to -87.5 dBm. Clearly,

as the MT velocity increases, the signal decay rate will increase. Hence, the de-

crease in the optimal ASST for the faster MT compensates this to make both MTs

handoff at a similar distance, in order to satisfy the required delay constraint.

Figure 12 plots the optimal ASST over different handoff signalling costs and

packet delay penalties, where V = 2 and θD= 50. Clearly, as the signaling cost

increases, the optimal ASST decreases sub-linearly in dB. With high handoff cost,

the MT is pushed to perform handoff nearer the WLAN edge, and hence reducing

23

Page 24

0 100 200 300 400

Handoff Cost (cH)

500 600 700800900 1000

−90

−89.5

−89

−88.5

−88

−87.5

−87

Optimal ASST (dBm)

cD =10000

cD =100000

cD =1000000

Figure 12: Optimal ASST vs. handoff cost and delay penalty (V = 2, θD= 50).

thenumberofunnecessaryhandoffs. Thesamefigurealsoshowsthatasthepacket

delay penalty increases, the optimal ASST increases. Hence, the MT is allowed

to handoff earlier to avoid the deteriorating channel condition as it approaches the

WLAN edge.

Thus, the propose numerical analysis can provide general guidelines for the

optimal operation of lifetime-based VHO, adapting to various system conditions

through the ASST value. To implement this in practice, a lookup table for the

optimal ASST can be built based on the above analysis results.

6 Conclusions

In converged wireless systems, efficient vertical handoff management between

heterogeneous networks is critical to the overall system performance. We have

presented an application-specific signal strength tuning mechanism to a cross-

layer adaptive VHO approach, which takes into account the wireless channel vari-

ation, network layer latency, and application QoS demands. We have proposed

an analytical framework to evaluate the performance of VHO based on multiple

criteria. The adaptive VHO approach has been shown to improve the system re-

source utilization by increasing the reliance of the MT on the WLAN, as well as

conserving the resources of the 3G network for users located outside the WLAN.

24

Page 25

More importantly, the proposed application signal threshold adaptation provides

a means for flexible system design. Given a predefined priority policy, it can be

used to optimize the tradeoff between handoff signalling, available bandwidth,

and packet delay. Since the ASST can be optimally tuned for any access network

based on practical system characteristics and requirements, it may have a signif-

icant role in future generation wireless networks where access technologies with

vastly differing characteristics are expected to seamlessly co-exist and efficiently

inter-operate.

Appendix: Z[k] Statistics

Variance

σ2

z[k]=

=

E{Z2[k]} − E2{Z[k]}

E{RSS2[k]} + T2

(23)

HOE{S2[k]} − 2THOE{RSS[k]S[k]} − E2{Z[k]} (24)

Since

E{RSS[k]S[k]}

=

E{RSS[k]2?i=WS

2−1

i=0

RSS[k − WS+ 1 + i]

W2

STs

RSS[k − WS+ 1 + i]

W2

STs

} −

E{RSS[k]

=2?i=WS

2?i=WS−1

σ2?i=Wav−1

= μRSS[k]μS[k] −2σ2?h=Wav−1

2?i=WS−1

i=WS

2

}

(25)

2−1

i=0

μRSS[k]μRSS[k − WS+ 1 + i]

W2

STs

μRSS[k]μRSS[k − WS+ 1 + i]

W2

STs

(Wav− |h|)

2WSTsW2

av

−

i=WS

2

−

i=0

WS

(26)

h=0

(Wav− |h|)

STsW2

av

W2

,

(27)

We have,

σ2

z[k]=

μ2

RSS[k] + σ2

RSS[k] + T2

HOμ2

S[k] + T2

HOσ2

S[k] − 2THOμRSS[k]μS[k] +

25

Page 26

4THOσ2?h=Wav−1

h=0

W2

(Wav− |h|)

av

S[k] +4THOσ2?i=Wav−1

STsW2

− E2{Z[k]}

(28)

= σ2

RSS[k] + T2

HOσ2

i=0

W2

(Wav− |h|)

av

STsW2

.

(29)

One-Step Autocorrelation Coefficient

By definition,

ρZ[k]Z[k−1]=Cov(Z[k],Z[k − 1])

σZ[k]σZ[k−1]

.

(30)

Since Z[k] = RSS[k] − THO∗ S[k], we have

Cov(Z[k + 1],Z[k])=

=

E{(Z[k + 1] − μZ[k + 1])(Z[k] − μZ[k])}

RRSS[k + 1,k] − THOE

T2

μZ[k + 1]μZ[k] − μZ[k]μRSS[k + 1] + THOμZ[k]μS[k + 1].

?

RSS[k]S[k + 1]

?

− THOE

?

S[k]RSS[k + 1]

?

(31)

(32)

+

HORS[k + 1,k] − μZ[k + 1]μRSS[k] + THOμZ[k + 1]μS[k] +

It can be shown that

RRSS[k + 1,k] = μRSS[k + 1]μRSS[k] +σ2(Wav− 1)

W2

av

,

(33)

and from (4)

E

?

RSS[k]S[k + 1]

?

= μRSS[k + 1]μS[k] −2σ2(2Wav− 1 +?Wav−1

= μRSS[k + 1]μS[k] −2σ2?Wav−1

h=1

STS

(Wav− |h|))

W2

avW2

(34)

E

?

S[k]RSS[k + 1]

?

h=1

W2

(Wav− |h|)

avW2

STS

(35)

and

RS[k + 1,k]=

μS[k + 1]μS[k] +

4σ2

avW4

?

W2

ST2

S

∗

(

Wav(WS− 2) +

Wav−1

h=1

(Wav− |h|)(2WS− 4|h|) − (Wav+

Wav−1

?

h=1

(Wav− |h|)(2|h|))

(36)

26

Page 27

=

μS[k + 1]μS[k] +

4σ2

avW4

Wav−1

?

W2

ST2

S

∗

(Wav(WS− 3) +

h=1

(Wav− |h|)(2WS− 6|h|)).

(37)

By direct substitution from (15), (34), and (37) in (30), we get

Cov(Z[k],Z[k − 1])=

σ2

W2

4 ∗ T2

W4

av

[(Wav− 1) +4THO

?Wav−1

h=1

W2

(Wav− |h|)

S∗ TS

+

HO

ST2

?

S

(Wav(WS− 3) +

Wav−1

h=1

(Wav− |h|)(2WS− 6 ∗ |h|))],

(38)

and consequently the ρZ[k]Z[k−1]can be obtained by direct substitution from (29) and (38)

in (30).

References

[1] R. Berezdivin, R. Breinig, and R. Topp, “Next-generation wireless commu-

nications concepts and technologies,” IEEE Commun. Mag., vol. 40, no. 3,

pp. 108 – 116, March 2002.

[2] M. M. Buddhikot, G. Chandranmenon, S. Han, Y. W. Lee, S. Miller, and

L. Salgarelli, “Integration of 802.11 and third generation wireless data net-

works,” in Proc. of IEEE INFOCOM, San Francisco, US, Apr. 2003, pp. 503

– 512.

[3] A. K. Salkintzis, C. Fors, and R. Pazhyannur, “WLAN-GPRS integration

for next-generation mobile data networks,” IEEE Wireless Commun., vol. 9,

no. 5, pp. 112 – 124, Oct. 2002.

[4] I. F. Akyildiz, J. McNair, J. Ho, H. Uzunalioglu, and W. Wang, “Mobility

management in current and future communications networks,” IEEE Net-

work, vol. 12, no. 4, pp. 39 – 49, Jul/Aug 1998.

27

Page 28

[5] B. Liang and Z. J. Haas, “Predictive distance-based mobility management

for multi-dimensional pcs networks,” IEEE/ACM Transactions on Network-

ing, vol. 11, no. 5, pp. 718–732, October 2003.

[6] J. Makela, M. Ylianttila, and K. Pahlavan, “Handoff decision in multi-

service networks,” in Proc. of 11th IEEE International Symposium on Per-

sonal, Indoor and Mobile Radio Communications (PIMRC’00), vol. 1, Lon-

don, UK, Sep. 2000, pp. 655 – 659.

[7] G. P. Pollini, “Trends in handover design,” IEEE Commun. Mag., vol. 34,

no. 3, pp. 82 – 90, March 1996.

[8] A. Hatami, P. Krishnamurthy, K. Pahlavan, M. Ylianttila, J. Makela, and

R. Pichna, “Analytical framework for handoff in non-homogeneous mobile

data networks,” in Proc. of (PIMRC’99), Osaka, Japan„ Sep. 1999, pp. 760

– 764.

[9] M. Ylianttila, M. Pande, J. Makela, and P. Mahonen, “Optimization scheme

for mobile users performing vertical hand-offs between IEEE 802.11 and

GPRS/EDGE networks,” in Proc. of IEEE Global Telecommunications Con-

ference GLOECOM’01, vol. 6, San Antonio, Texas, USA, Nov. 2001, pp.

3439 – 3443.

[10] M. Ylianttila, J. Makela, and P. Mahonen, “Supporting resource allocation

with vertical handoffs in multiple radio network environment,” in Proc. of

IEEE International Symposium on Personal, Indoor and Mobile Radio Com-

munications (PIMRC’02), Lisbon, Portugal, Sep. 2002, pp. 64 – 68.

[11] H. Park, S. Yoon, T. Kim, J. Park, M. Do, , and J. Lee, “Vertical handoff

procedure and algorithm between IEEE802.11 WLAN and CDMA cellular

network.” Proc. CDMA Int’l Conf., 2002, pp. 103–112.

[12] M. Ylianttila, R. Pichna, J. Vallstram, J. Makela, A. Zahedi, P. Krish-

namurthy, and K. Pahlavan, “Handoff procedure for heterogeneous wire-

less networks,” in Proc. of IEEE Global Telecommunications Conference

(GLOBECOM’99), vol. 5, Dec. 1999, pp. 2783 – 2787.

[13] K. Pahlavan, P. Krishnamurthy, A. Hatami, M. Ylianttila, J. P. Makela,

R. Pichna, and J. Vallstron, “Handoff in hybrid mobile data networks,” IEEE

Commun. Mag., vol. 7, no. 4, pp. 34 – 47, Apr. 2000.

28

Page 29

[14] A. Majlesi and B. H. Khalaj, “An adaptive fuzzy logic based handoff al-

gorithm for hybrid networks,” in Proc. of 6th International Conference on

Signal Processing, vol. 2, Aug. 2002, pp. 1223 – 1228.

[15] H. Wang, R. H. Katz, and J. Giese, “Policy-enabled handoffs across het-

erogeneous wireless networks,” in Proc. of the Second IEEE Workshop on

Mobile Computer Systems and Applications, New Orleans, Louisiana, Feb.

1999, p. 51.

[16] F. Zhu and J. McNair, “Optimizations for vertical handoff decision algo-

rithms,” in in Proc. of IEEE Wireless Communications and Networking Con-

ference (WCNC), vol. 2, March 2004, pp. 867 – 872.

[17] L. Chen, T. Sun, B. Chen, V. Rajendran, and M. Gerla, “A smart decision

model for vertical handoff.” The 4th Int’l Workshop on Wireless Internet

and Reconfigurability (ANWIRE’04), May 2004.

[18] M. M. Buddhikot, G. Chandranmenon, S. Han, Y. Lee, S. Miller, and L. Sal-

garelli, “Design and implementation of a WLAN/CDMA2000 interworking

architecture,” IEEE Commun. Mag., vol. 41, no. 11, pp. 90 – 100, Nov. 2003.

[19] T. S. Rappaport, Wireless Communications: Principles and Practice. Pren-

tice Hall, July 1999.

[20] K. Murakami, O. Haase, J. Shin, and T. LaPorta, “Mobility management

alternatives for migration to mobile internet session-based services,” IEEE

Journal on Selected Areas in Communications, vol. 22, no. 5, pp. 834 – 848,

June 2004.

[21] C. Guo, Z. Guo, Q. Zhang, and W. Zhu, “A seamless and proactive end-to-

end mobility solution for roaming across heterogeneous wireless networks,”

IEEE Journal on Selected Areas in Communications, vol. 22, no. 5, pp. 834

– 848, Jun. 2004.

[22] C. Tepedelenlioglu and G. Giannakis, “On velocity estimation and corre-

lation properties of narrow-band mobile communication channels,” IEEE

Trans. on Vehicular Technology, vol. 50, no. 4, pp. 1039 – 1052, Jul. 2001.

[23] Q. Zhang, C. Guo, Z. Guo, and W. Zhu, “Efficient mobility management

for vertical handoff between WWAN and WLAN,” IEEE Commun. Mag.,

vol. 41, no. 11, pp. 102 – 108, Nov. 2003.

29

Page 30

[24] N. Zhang and J. M. Holtzman, “Analysis of handoff algorithms using both

absolute and relative measurements,” IEEE Transactions on Vehicular Tech-

nology, vol. 45, no. 1, pp. 174 – 179, Feb. 1996.

[25] A. Papoulis and S. Pillai, Probability, Random Variables and Stochastic Pro-

cesses, 4th ed. McGraw-Hill, 2002.

[26] G. Bolch, S. Greiner, H. de Meer, and K. S. Trivedi, Queuing networks and

Markov Chains: Modeling and Performance Evaluation with Computer Sci-

ence Applications, 2nd ed. Wiley, August 1998.

[27] IEEE Standard, “Part 11: Wireless LAN medium access control (MAC) and

physical layer (PHY) specifications,” IEEE, 1999.

[28] 3GPP2 Standard, “CDMA2000 wireless ip network standard: Packet data

mobility and resource management,” 3GPP2 X.S0011-003-C v1.0, Sep.

2003.

[29] N. Tripathi, J. Reed, and H. VanLandingham, “Handoff in cellular systems,”

IEEE Personal Commun., vol. 5, no. 6, pp. 26– 37, Dec. 1998.

30

Page 31

List of Figures

1 Mobile handoff in heterogeneous wireless system . . . . . . . . .3

2 MO handoff algorithm . . . . . . . . . . . . . . . . . . . . . . .10

3VHO Markov chain model . . . . . . . . . . . . . . . . . . . . . 15

4 Number of handoffs (γ= -90 dBm). . . . . . . . . . . . . . . . . .18

5 Available bandwidth (γ = -90 dBm). . . . . . . . . . . . . . . . .19

6 HoL packet delay rate (γ = -90 dBm, θD=30 ms). . . . . . . . . .19

7 Number of handoffs vs. ASST. . . . . . . . . . . . . . . . . . . .20

8 Available bandwidth vs. ASST. . . . . . . . . . . . . . . . . . . .21

9 HoL packet delay rate vs. ASST (θD= 30ms). . . . . . . . . . . . 21

10 Total cost vs. ASST. . . . . . . . . . . . . . . . . . . . . . . . . .22

11 OptimalASSTvs.delaybudgetthresholdandvelocity(cH= 100,

cD= 100000). . . . . . . . . . . . . . . . . . . . . . . . . . . . .23

12 Optimal ASST vs. handoff cost and delay penalty (V = 2, θD= 50). 24

31

Page 32

List of Tables

1 Simulation parameter values . . . . . . . . . . . . . . . . . . . .17

32