IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 24,

NO. 2,

FEBRUARY 2013

225

Analysis of Distance-Based
Location Management in Wireless
Communication Networks
Keqin Li, Senior Member, IEEE
AbstractThe performance of dynamic distance-based location management schemes (DBLMS) in wireless communication
networks is analyzed. A Markov chain is developed as a mobility model to describe the movement of a mobile terminal in 2D cellular
structures. The paging area residence time is characterized for arbitrary cell residence time by using the Markov chain. The expected
number of paging area boundary crossings and the cost of the distance-based location update method are analyzed by using the
classical renewal theory for two different call handling models. For the call plus location update model, two cases are considered. In the
first case, the intercall time has an arbitrary distribution and the cell residence time has an exponential distribution. In the second case,
the intercall time has a hyper-Erlang distribution and the cell residence time has an arbitrary distribution. For the call without location
update model, both intercall time and cell residence time can have arbitrary distributions. Our analysis makes it possible to find the
optimal distance threshold that minimizes the total cost of location management in a DBLMS.
Index TermsCost analysis, distance-based location management scheme, renewal process, wireless communication network

INTRODUCTION

1.1

Motivation
N order to successfully and efficiently deliver incoming
calls to a mobile terminal, a cellular wireless communication network that provides personal communication
service (PCS) needs to constantly keep track of the location
of a mobile terminal. Therefore, location and mobility
management is an important and fundamental issue in
wireless communication. A location management scheme is
a key and critical component of any wireless communication network to effectively deliver network services to
mobile users. There are two essential tasks in location
management, namely, location update (location registration) and terminal paging (call delivery):

Location update is the process for a mobile terminal to

periodically notify its current location to a network
so that the network can revise the mobile terminals
location profile in a location database.
. Terminal paging is the process for a network to search
a mobile terminal by sending polling signals based
on the information of its last reported location so
that an incoming phone call can be routed to the
mobile terminal.
The location database entry of a mobile terminal is updated
when the mobile terminal performs a location update and/
or when a network performs a terminal paging during the
call delivery to the mobile terminal.
.

. The author is with the Department of Computer Science, State University

of New York, New Paltz, New York 12561. E-mail: [email protected].
Manuscript received 29 July 2011; revised 22 Nov. 2011; accepted 24 Feb.
2012; published online 18 Apr. 2012.
Recommended for acceptance by J. Zhang.
For information on obtaining reprints of this article, please send e-mail to:
[email protected], and reference IEEECS Log Number TPDS-2011-07-0502.
Digital Object Identifier no. 10.1109/TPDS.2012.123.
1045-9219/13/$31.00 2013 IEEE

Both location update and terminal paging consume

significant communication bandwidth of a wireless network, battery power of mobile terminals, memory space in
location registers and databases, and computing time at
base stations. Therefore, both location update cost and
terminal paging cost should be minimized. However, there
is a tradeoff between the cost of location update and the cost
of terminal paging. More location updates are likely to
reduce the cost of terminal paging, but excessive location
update wastes system resources while contributes little to
call delivery. On the other hand, insufficient location
update increases cost of terminal paging. A dynamic
location management scheme has the capability to adjust
its location update and terminal paging strategies based on
the mobility pattern and incoming call characteristics of a
mobile terminal, such that the combined cost of location
update and terminal paging is minimized.

1.2 Existing Work

The design and analysis of any dynamic location management scheme depend on a mobility model of mobile
terminals. Various mobility models have been proposed in
the literature, including the shortest distance mobility
model [2], [3], the fluid flow model [7], [12], the big move
and the random walk models [13], the user mobility pattern
scheme [15], the cell coordinates system [32], the isotropic
diffusive motion model [34], 1D Markov chains [4], [11],
[14], [31], [37], and 2D Markov chains [5], [7], [17], [24].
There are three location update methods, namely, the
distance-based method, the movement-based method, and
the time-based method [11]. Accordingly, there are three
types of dynamic location management schemes, namely,
distance-based location management schemes (DBLMS), movement-based location management schemes (MBLMS), and timebased location management schemes (TBLMS). A DBLMS (an
Published by the IEEE Computer Society

226

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

MBLMS and a TBLMS, respectively) employs the distancebased (movement-based and time-based, respectively) location update method. Furthermore, a DBLMS or an MBLMS
or a TBLMS can use various terminal paging methods.
Dynamic location management has been studied extensively by many researchers. The performance of movementbased location management schemes has been investigated
in [5], [11], [19], [27], [28], [29], [30], [33], [47], [50]. The
performance of distance-based location management
schemes has been studied in [3], [11], [14], [24], [31], [32].
The performance of time-based location management
schemes has been considered in [4], [11], [12], [34]. Terminal
paging methods with low cost and time delay have been
studied by several researchers [2], [5], [10], [24], [36], [44],
[46]. Other studied were reported in [7], [13], [15], [17], [18],
[20], [35], [37], [45], [49]. Dynamic location management in a
wireless communication network with a finite number of
cells has been treated as an optimization problem which is
solved by using bioinspired methods such as simulated
annealing, neural networks, and genetic algorithms [8], [38],
[40], [41], [42], [43]. The reader is also referred to the surveys
in [6], [26], [39], [22, Chapter 15], and [23, Chapter 11].
For distance-based location management schemes, the
studies in [11], [14], [31] assume that a wireless communication network has a linear cellular structure. Such
modeling is certainly practically less interesting, since a
wireless communication network has a 2D configuration. A
discrete-time random walk model was developed in [24],
where a wireless communication network has a 2D
hexagonal cell structure. However, the Markov chain model
of [24] implies that intercall times and cell residence times
are geometrically distributed, and the model cannot be
applied to intercall times and cell residence times that are
continuous random variables with arbitrary distributions. A
modified distance-based location update method is considered in [3] for the 2D mesh cell structure, where the
objective is to find an optimal location registration area such
that the location update cost is minimized. The study in [32]
deviates more from other studies, where the mobility of a
mobile terminal is not modeled using cell residence times
but the speed and direction of movement in a cell
coordinates system.
The investigation in [52] is the most related work to our
study, in which, the expected number of location updates
between two consecutive phone calls is derived, and the
result is further used to determine the optimal distance
threshold that minimizes the total location management
cost. The main problem is that the transition probabilities of
the Markov chain are not accurate (see Section 4 for detailed
discussion). Another limitation is that it is assumed that
the intercall time has an exponential distribution, although
the cell residence time can have an arbitrary distribution.
The approach in [52] is extended to find location distribution of a mobile terminal, which is very useful to reduce
paging cost [51]. Analysis of a random walk of a mobile
terminal in a hexagonal cell configuration is also attempted
in [9]. In [48], the authors even considered irregular cell
structures, where an arbitrary cell topology is represented
by a random planar graph.
In summarizing the current research, we find that 1) there
is no accurate description of the movement of a mobile

VOL. 24, NO. 2,

FEBRUARY 2013

terminal in 2D cell configurations; 2) there is no any

characterization of the paging area (PA) residence time
which is critical in studying the performance of a DBLMS;
3) and there is no analytical expression of the cost of the
distance-based location update method. The objective of
this paper is to solve the above problems.

1.3 Our Contributions

The present paper makes significant contributions to cost
analysis and minimization of dynamic distance-based
location management schemes in wireless communication
networks. We solve three problems and the main contributions of the paper are threefold.
The first contribution of the paper is to develop a very
accurate ring level Markov chain as a mobility model to
describe the movement of a mobile terminal (Section 4).
Results of the Markov chain are critical in studying the
paging area residence time, which in turn, is critical in
studying location update cost in a DBLMS. We would like
to mention that there are many other important applications
of the results from our Markov chain and paging area
residence time. The Markov chain can be used to study
location distribution of a mobile terminal in a paging area,
which is critical in designing paging methods with reduced
cost in a DBLMS or an MBLMS. Our results from the
Markov chain and paging area residence time can also be
employed to study reachability of a mobile terminal in a
paging area, which is critical in analyzing the quality of
service and in reducing paging cost in a TBLMS.
The second contribution of the paper is to characterize
paging area residence time for arbitrary cell residence time
by using the Markov chain (Theorems 8 and 9). The main
obstacle in analyzing the performance of a DBLMS is to find
the characterization of the paging area residence time based
on known information of the cell residence time. It is clear
that the paging area residence time is determined by the cell
residence time, the movement pattern of a mobile terminal,
and the size of a paging area. By using a Markov chain that
characterizes the movement pattern of a mobile terminal in
a wireless communication network, we successfully find the
characterization of the paging area residence time using
known information of the cell residence time.
The third contribution of the paper is to solve the main
problem in analyzing the performance of a DBLMS, namely,
to find the number of paging area boundary crossings
within any time interval, which is directly related to the
number of location updates in a DBLMS. By using the
classical renewal theory, we are able to derive the expected
number of paging area boundary crossings (Theorems 1-5).
This leads to analytical results on location update cost in a
DBLMS. The results from the Markov chain and paging
area residence time are critical in the analysis of performance of a DBLMS for arbitrary intercall time and cell
residence time. Our analysis makes it possible to find the
optimal distance threshold that minimizes the total cost of
location management in a DBLMS.
Our analysis of the distance-based location update
method is conducted for two different call handling models.
For the call plus location update (CPLU) model, we consider
two cases. In the first case, the intercall time has an arbitrary
distribution and the cell residence time has an exponential

LI: ANALYSIS OF DISTANCE-BASED LOCATION MANAGEMENT IN WIRELESS COMMUNICATION NETWORKS

227

Fig. 1. The hexagonal cell configuration.

distribution (Theorem 10). In the second case, the intercall

time has a hyper-Erlang distribution and the cell residence
time has an arbitrary distribution (Theorem 11). For the call
without location update (CWLU) model, both intercall time
and cell residence time can have arbitrary distributions
(Theorem 12). It is well known that the distance-based
location update method has lower cost than the movementbased location update method. However, due to lack of
analytical results of the cost of the distance-based location
update method, it is not clear at all to what extent the
distance-based location update method is better than the
movement-based location update method. Our results in
this paper give clear comparison of the two methods.
The rest of the paper is organized as follows. In Section 2,
we describe the wireless communication network model
used in this paper, location update methods, terminal
paging methods, call handling models, cost analysis, and
notations. In Section 3, we present the mathematical tools
and results used in the paper, especially the probability
distribution and expectation of the number of renewals in a
random time interval in ordinary, equilibrium, and modified renewal processes. In Section 4, we develop a Markov
chain as a mobility model to characterize the movement
pattern of a mobile terminal in a wireless communication
network. In Section 5, we derive an expression for the
paging area residence time using an expression for the cell
residence time based on the Markov chain. In Section 6, we
study the number of paging area boundary crossings
between two consecutive phone calls and the cost of
location update in a DBLMS under two call handling
models. In Section 7, we demonstrate simulation results to
show the quality of our analysis and numerical data to
show the impact of various parameters and performance
optimization. We conclude the paper in Section 8.

PRELIMINARIES

2.1 Dynamic Location Management Schemes

A dynamic location management scheme consists of a
location update method, a terminal paging method, and

Fig. 2. The mesh cell configuration.

a call handling model, all applied in a wireless communication network.

2.1.1 Wireless Communication Networks

A wireless communication network has the common
hexagonal cell configuration or mesh cell configuration. In
the hexagonal cell structure (see Fig. 1), cells are hexagons of
identical size and each cell has six neighbors. In the mesh cell
structure (see Fig. 2), cells are squares of identical size and
each cell has eight neighbors. Throughout the paper, we let
q be a constant such that q 3 for the hexagonal cell
configuration and q 4 for the mesh cell configuration. By
using the constant q, the hexagonal cell configuration and
the mesh cell configuration can be treated in a unified way.
For instance, we can say that each cell has 2q neighbors
without mentioning the particular cell structure. The network is homogeneous in the sense that the behavior of a
mobile terminal is statistically the same in all the cells.
Let s be the cell registered by a mobile terminal in the last
location update. The cells in a wireless networks can be
divided into rings, where s is the center of the network and
called ring 0. The 2q neighbors of s constitute ring 1. In
general, the neighbors of all the cells in ring r, except those
neighbors in rings r  1 and r, constitute ring r 1. For all
r  0, the cells in ring r have distance r to s. For all r  1,
the number of cells in ring r is 2qr. Notice that the rings are
defined with respect to s. When a mobile terminal updates
its location to another cell s0 , s0 becomes the center of the
network, and ring r consists of the 2qr cells whose distance
to s0 is r.
When a mobile terminal u moves out of a cell s, it is
normally assumed that u moves into one of ss 2q neighbors
with equal probability [19], [27], although this assumption is
irrelevant in analyzing location update cost of movementbased schemes. However, how u moves into the neighboring
cells is very important in analyzing location update cost of
distance-based schemes and quality of service of time-base

228

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

schemes, and in reducing paging cost, since the way u moves

into the neighboring cells determines how fast or slow u
reaches the boundary of a paging area. When u leaves ring 0
(i.e., s0 ), u definitely enters ring 1. When u leaves a cell in ring
r  1, we are interested in a defined as the probability that u
moves into ring r 1, and b defined as the probability that u
moves into ring r  1. It is clear that the probability that
u remains in ring r is 1  a  b. For instance, if u moves into
one of s s 2q neighboring cells with equal probability, then for
a typical cell, we have a b q  1=2q.

2.1.2 Location Update Methods

A mobile terminal u constantly moves from cell to cell. Such
movement also results in movement from ring to ring. Let
the sequence of cells visited by u before the next phone call
be denoted as s0 , s1 , s2 ; . . . , sd ; . . . , where s0 s is us last
registered cell (not the cell in which u received the previous
phone call) and considered as us current location. There are
three location update methods proposed in the current
literature, namely, the distance-based method, the movement-based method, and the time-based method. In the
distance-based location update method, location update is
performed as soon as u moves into a cell sj in ring d,
where d is a distance threshold, i.e., the distance of u from
the last registered cell s is d, such that sj is registered as us
current location. It is clear that j  d, i.e., it takes at least d
steps for u to reach ring d. In the movement-based location
update method, location update is performed as soon as u has
crossed cell boundaries for d times since the last location
update, where d is a movement threshold. It is clear that the
sequence of registered cells for u is sd ; s2d ; s3d ; . . . In the
time-based location update method, location update is performed every  units of time, where  is a time threshold,
regardless of the current location of u.
2.1.3 Terminal Paging Methods
In all dynamic location management schemes, a current
paging area consists of rings 0; 1; 2; . . . ; d  1, where d is
some value appropriately chosen. We say that such a PA
has radius d. Since the number of cells in ring r is 2qr, for
all r  1, the total number of cells in a PA is qd2  qd 1.
It should be noticed that a PA is defined with respect to
the current location of a mobile terminal, and is changed
whenever a mobile terminal updates its location. The
radius d of a PA can be adjusted in accordance with
various cost and performance considerations. On the
other hand, the location and size of a cell are fixed in a
wireless network.
Two terminal paging methods have been proposed in the
literature. In the simple paging method, the radius of a PA is
fixed at d, where d is the distance threshold used by a
distance-based location update method, or the movement
threshold used by a movement-based location update
method, or appropriately chosen in accordance with the
time threshold used by a time-based location update method.
In the selective paging method, cells in a PA or the entire
wireless communication network are divided into disjoint
areas, such that these areas are paged one after another
successively, until a mobile terminal is found. The advantage
of the simple paging method is that a mobile terminal is
guaranteed to find in one polling step. The disadvantage of

VOL. 24, NO. 2,

FEBRUARY 2013

the simple paging method is the high cost of polling qd2 

qd 1 cells. The advantage of the selective paging method is
the low expected cost of paging. The disadvantage of the
selective paging method is the long time delay in finding a
mobile terminal. Since the main focus of this paper is to study
location update cost in a DBLMS, we will only consider
the simple paging method. We believe that analysis of the
selective paging method as well as its impact on the
performance of a DBLMS is a deep topic and deserves a
separate paper to deal with.

2.1.4 Call Handling Models and Renewal Processes

We will consider two different call handling models [29]. In
the call plus location update model, the location of a mobile
terminal is updated each time a phone call arrives. That is,
in addition to distance-based or movement-based or timebased location updates, the arrival of a phone call also
initiates location update and defines a new PA. This causes
the original location update cycle of a mobile terminal being
interrupted. In the call without location update model, the
arrival of a phone call has nothing to do with location
update, that is, a mobile terminal still keeps its original
location update cycles.
2.2

Performance Analysis

2.2.1 Notations
Throughout the paper, we use P E to denote the
probability of an event E. For a random variable T , we
use E T to represent the expectation of T and
T E T 1 . The probability density function (pdf) of T
is fT t, and the cumulative distribution function (cdf) of T
is FT t. The Laplace transform of fT t and FT t for a
nonnegative random variable T are defined as
Z 1
est fT tdt;
fT s E esT
0

and
FT s

est FT tdt:

There are several important random variables in the

study of dynamic location management. The intercall time Tc
is defined as the length of the time interval between two
consecutive phone calls. The cell residence time Ts is defined
as the time a mobile terminal stays in a cell before it moves
into a neighboring cell. The paging area residence time Tm is
defined as the time a mobile terminal stays in the current
PA before it moves out of the PA. The quantity  Tc =Ts
is the call-to-mobility ratio.
Throughout the paper, we use DBLMS d to represent a
dynamic location management scheme using the distancebased location update method with distance threshold d
and the simple paging method. We use DBLMS d-CPLU
and DBLMS d-CWLU to represent a dynamic location
management scheme using the distance-based location
update method with distance threshold d under the CPLU
model and the CWLU model, respectively.

LI: ANALYSIS OF DISTANCE-BASED LOCATION MANAGEMENT IN WIRELESS COMMUNICATION NETWORKS

2.2.2 Cost Analysis

The cost of dynamic location management contains two
components, i.e., the cost of location update and the cost of
terminal paging. The cost of location update is proportional
to the number of location updates. If there are Xu location
update between two consecutive phone calls, the cost of
location update is u Xu , where u is a constant. Since Xu is
a random variable, the location update cost is actually
calculated as u E Xu . The cost of terminal paging is
proportional to the number of cells paged. If a PA has
radius d, the cost of paging is p qd2  qd 1, where p
is a constant.
In a DBLMS, location update is performed only when a
mobile terminal moves out of the current PA. In an MBLMS,
a mobile terminal may still resides in the current PA when a
location update is performed. This implies that there might
be unnecessary location updates. For the same value d of
distance threshold and movement threshold, an MBLMS
requires more location update cost than a DBLMS. Using a
DBLMS instead of an MBLMS can reduce location update
cost. However, for large d, if the simple paging method is
used, such reduction in location update cost does not
significantly reduce the total location management cost
because the paging cost grows dramatically as d becomes
large. Therefore, reducing both location update cost and
paging cost is the only effective way to design high
performance dynamic location management schemes.
Dynamic location management is per-terminal based. A
mobile terminal is specified by fTc t and fTs t, where fTc t
is the call pattern and fTs t is the mobility pattern. Since a
location update method determines the location update cost
and a terminal paging method determines the terminal
paging cost, for given a mobile terminal, we need to find a
balanced combination of a location update method and a
terminal paging method such that the total location
management cost for the mobile terminal is minimized.

RENEWAL PROCESSES

A renewal process is defined by a sequence of independent

random variables T1 ; T2 ; T3 ; . . . , where T2 ; T3 ; . . . are a
sequence of independent and identically distributed (i.i.d.)
random variables with a common pdf, but T1 may have a
different pdf. (The reader is referred to [1] for a general
introduction to the renewal theory.) A renewal process has
many associated random variables and properties. The
most interesting property related to our study is the number
of renewals in a random period of time. We use Xt to
denote the number of renewals in a time interval of length t.
Let Sj T1 T2    Tj . If Sj  t < Sj1 , we say that the
number of renewals Xt in a time interval of length t is j.
Let X be the number of renewals in a random time
interval of length Tc . Define
Ht

1
X

FSj t:

j1

The following theorem gives the probability distribution and

the expectation of Xfor an arbitrary renewal process. The
theorem will be used in Theorems 3 and 10. The proof is
given in Appendix 1, which can be found on the Computer

229

Society Digital Library at http://doi.ieeecomputersociety.

org/10.1109/TPDS.2012.123.
Theorem 1. For an arbitrary renewal process, we have
Z 1


FSj t  FSj1 t fTc tdt;
P X j
0

for all j  0; furthermore, the expected number of renewals in

a random time interval of length Tc is
Z 1
HtfTc tdt;
E X
0

for arbitrary T1 ; T2 ; T3 ; . . . and fTc t.

3.1 Ordinary Renewal Processes

An ordinary renewal process is defined by a sequence of i.i.d.
random variables T1 , T2 , T3 ; . . . , with a common pdf fT t
[16]. In modeling dynamic location management schemes,
the Ti s can be cell residence times, or paging area residence
times, or location update times. When the Ti s are cell
residence times, each renewal stands for a cell boundary
crossing, and Xt is the number of cell boundary crossings
in a time interval of length t. When the Ti s are paging area
residence times, each renewal stands for a PA boundary
crossing, and Xt is the number of PA boundary crossings
in a time interval of length t. When the Ti s are location
update times, each renewal stands for a location update,
and Xt is the number of location updates in a time
interval of length t. When t is randomized according to
intercall time distribution fTc t, we are essentially considering the number of cell boundary crossings, or the
number of PA boundary crossings, or the number of
location updates between two consecutive phone calls.
Based on FSj s, we can obtain FSj t by performing the
inversion integral
Z i1
1
FSj t
F  sest ds
2i i1 Sj
Z i1 
1
fT sj st

 e ds;
2i i1
s
for t  0 and  > a , where the integration in the complex
plane is taken to be a straight-line integration parallel to
the imaginary axis and lying to the right of a , the abscissa
of absolute convergence for FSj s. The usual means for
performing this integration is to carry out an integration
around a closed contour, namely, a semicircle of infinite
radius, so that the Cauchy residual theorem can be
applied. This is most easily done if FSj s is in rational
form [25, p. 353]. This is indeed the case for a hyper-Erlang
distribution of the Ti s, that is,
!
k
X
i ei t i ti 1
;
wi
fT t
i  1!
i1
where w1 w2    wk 1. Special forms of hyperErlang distributions include hyperexponential distributions
(i 1 for all 1  i  k); exponential distributions (k 1
and 1 1); chi-square distributions (k 1 and 1 1=2);
Erlang distributions (k 1).

230

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

The following theorem will be used in Section 6.1. The

proof is given in Appendix 1, available in the online
supplemental material.
Theorem 2. If the Ti s have a hyper-Erlang distribution, we have
Y
k 
j
j
FSj t
wi i i i
j
;
j
;
.
.
.
;
j
1 2
k i1
j1 j2 jk j
 i ji 1
k
k
Y 1
X
1
@
i j i

@s
j

1!
i i
i1 i
i1
1
"
#
est Y
1
A;
s i0 6i s i0 i0 ji0
X

si

for all j  0.

3.2 Equilibrium Renewal Processes

Consider a mobile terminal u moving through a sequence of
cells s1 ; s2 ; s3 ; . . . , with a sequence of cell residence times
Ts;1 ; Ts;2 ; Ts;3 ; . . . . Assume that u is in s1 when a phone call
arrives, i.e., u has been in s1 for a while. Clearly, Ts;1 is the
residual time of u in s1 and does not have the same pdf as the
other Ts;i s. Similarly, consider a mobile terminal u moving
through a sequence of paging areas PA1 , PA2 , PA3 ; . . . , with
a sequence of paging area residence times Tm;1 ; Tm;2 ;
Tm;3 ; . . . Assume that u is in PA1 when a phone call arrives,
i.e., u has been in PA1 for a while. Clearly, Tm;1 is the
residual time of u in PA1 and does not have the same pdf as
the other Tm;i s.
An ordinary renewal process T1 ; T2 ; T3 ; . . . , where the
pdf of T1 is the residual time of an ordinary Ti , is called an
equilibrium renewal process, which can be regarded as an
ordinary renewal process that has been running for a long
time before it is first observed [16]. Notice that if the Ti s
have an exponential distribution, an equilibrium renewal
process becomes an ordinary renewal process.
The expected number of renewals in a random time
interval of an equilibrium renewal process is surprisingly
easy to obtain. The following theorem is a simple and
strong result in renewal theory [16]. The theorem will be
used in Theorem 12. The proof is given in Appendix 1,
available in the online supplemental material.
Theorem 3. For any probability distributions of Tc and T , the
expected number of renewals in a random time interval of
E T T =Tc for an equililength Tc is E X E Tc =E
brium renewal process.

3.3 Modified Renewal Processes

A modified renewal process has all the properties of an
ordinary renewal process T1 , T2 , T3 ; . . . , except that T1 has
a different pdf fT1 t from that of other Ti s which have the
same pdf fT t [16]. Such a renewal process occurs in a
DBLMS-CPLU, where each arriving phone call in the
midst of cell residence time initiates a location update and
creates a new PA. However, cell residence times cannot be
altered. This results in a sequence of cell residence times
forming an equilibrium renewal process, which results in
a sequence of paging area residence times forming a
modified renewal process.

VOL. 24, NO. 2,

FEBRUARY 2013

The following theorem gives the probability distribution

of X in a modified renewal process when Tc has a hyperErlang distribution with pdf:
!
kc
X
c;i c;i tc;i 1 ec;i t
;
wc;i
fTc t
c;i  1!
i1
where wc;1 wc;2    wc;kc 1. The proof is given in
Appendix 1, available in the online supplemental material.
Theorem 4. For a modified renewal process, if Tc has a hyperErlang distribution, we have
!

kc
X
c;ic;i
wc;i
P X 0 1 
c;i  1!
i1

c;i 1  

fT1 s
@
;

@s
s
sc;i
and
!

c;ic;i
wc;i
P X j
c;i  1!
i1

c;i 1  


fT1 s1  fT s
@
j1

fT s
;

@s
s
sc;i
kc
X

for all j  1.
The following theorem gives E X for a modified
renewal process based on Theorem 4 by straightforward
calculation. The theorem will be used in Theorem 11.
Theorem 5. If Tc has a hyper-Erlang distribution, the expected
number of renewals in a random time interval of length Tc is
!

kc
X
c;ic;i
wc;i
E X
c;i  1!
i1

c;i 1 

fT1 s
@
;

@s
s1  fT s sc;i
for a modified renewal process with arbitrary fT1 t and fT t.

MOBILITY MODELING

4.1 A Cell Level Markov Chain

It is clear that the movement of a mobile terminal can be
modeled by a continuous-time semi-Markov stochastic
process with discrete state space. Such a process describes
the location of a mobile terminal at any time. In such a
process, the state space is a set of cells (discrete state space).
State transition means moving into a neighboring cell. Since
a mobile terminal can stay in a cell for arbitrary amount of
time, state transitions can occur at any time (continuous
time). Furthermore, since the cell residence time Ts can have
an arbitrary fTs t (not necessarily an exponential distribution), such a process is not a Markov process. However,
notice that at the instants of state transitions, the process
behaves just like an ordinary Markov chain. Therefore, at
those instants, we have an embedded Markov chain ([25, p. 23]).
Consider a mobile terminal u in a cell s. After staying in s
for Ts amount of time, u moves out of s and enters into one

LI: ANALYSIS OF DISTANCE-BASED LOCATION MANAGEMENT IN WIRELESS COMMUNICATION NETWORKS

of the 2q neighbors of s. It is easily observed that the

movement of a mobile terminal can be described by a 2D
Markov chain. For each cell, we have a state in the Markov
chain associated with the cell. The transition probability
from a cell to a neighboring cell is 1=2q. While the cell
level Markov chain accurately describes the real cell
structures, the number of states is exactly the same as the
number of cells. As shown below, we will need to know
the behavior of the Markov chain up to ring d, where d is the
radius of a PA, which implies that the number of states in
the Markov chain is as large as qd2 qd 1. Since we will
perform multiplications of matrices of transition probabilities, the computational complexity will be Oqd2 qd
13 Od6 , not mentioning the huge number of repetitions of matrix multiplications to achieve the desired
numerical accuracy.

4.2 A Ring Level Markov Chain

To reduce the number of states, we can construct a ring level
Markov chain which contains states K0 ; K1 ; K2 ; . . . ; Kd ; . . . ,
where state Kr means that a mobile terminal u is in ring r,
r  0. Initially, u is in state K0 . Instead of the probabilities of
moving into neighboring cells, we are interested in the
probabilities of moving into adjacent rings, i.e., the probability ar of moving into ring r 1 and the probability br of
moving into ring r  1. Let pij denote the transition
probability from Ki to Kj , where i; j  0. Then, we have
p01 1, pr;r1 ar , pr;r1 br , pr;r 1  ar  br , for all r  1.
All other pij s not specified above are zeros. Using the ar s
and the br s, the matrix of transition probabilities P pij  is
2
3
0
1
0
0
0 
6 b1 1  a1  b1
a1
0
0 7
6
7
60
1

a

b
a
0 7
b
2
2
2
2
6
7:
60
0
b3
1  a3  b3 a3    7
4
5
.. . .
..
..
..
..
.
.
.
.
.
.
n

We use pij to denote the n-step transition probability from

Ki to Kj , where i; j  0. The following result is well known
([21, p. 383]), which will be used in Theorem 7.
Theorem 6. If the nth power of P is P n gij , we have
n
pij gij , for all i; j  0.
The ar s and the br s can be obtained as follows. First, we
make the assumption that all the cells in the same ring are
visited with equal probability. Let the cells in ring r be s1 ,
s2 ; . . . , s2qr . Assume that cell si has xi neighbors in ring r 1
and yi neighbors in ring r  1. We call such a cell as an
xi ; yi cell. If x is the average value of the xi s and y is the
average value of the yi s, we have ar x=2q and
br y=2q. That is,
ar

x1 x2    x2qr
;
4q2 r

and

we have a1 1=2 and b1 1=6. When r > 1, there are six

3; 1 cells at the six corners of a hexagonal ring, and all
other cells are 2; 2 cells. Thus, we have
ar

3  6 26r  6 2r 1

;
36r
6r

br

1  6 26r  6 2r  1

:
36r
6r

and

It is clear that the above expressions for ar and br are also

valid for a1 and b1 . Next, we examine the mesh cell structure
(see Fig. 2). When r 1, there are four 3; 1 cells and four
5; 1 cells. Thus, we have a1 1=2 and b1 1=8. When
r > 1, there are four 5; 1 cells at the four corners of a
square ring and eight 3; 2 cells that are adjacent to the
corner cells, and all other cells are 3; 3 cells. Thus, we have
ar

5  4 38r  4 3r 1

;
64r
8r

and
br

1  4 2  8 38r  12 3r  2

:
64r
8r

It is clear that the above expressions for ar and br are also

valid for a1 and b1 . The ar s and the br s in the hexagonal and
mesh cell structures can be unified as follows:
ar

y1 y2    y2qr
:
4q2 r

Let us look at the hexagonal cell structure first (see Fig. 1).
When r 1, all the six cells in ring 1 are 3; 1 cells. Thus,

q  1r 1
;
2qr

and
br

q  1r  q 2
;
2qr

for all r  1. The above ar and br for q 3 were obtained in

[5] and adopted in subsequent studies such as [52].
Next, we notice that the assumption that all the cells in
the same ring are visited with equal probability does not
reflect what happens in a real cell structure. It is easily
observed that when r > 1, the six corner cells of a hexagon
ring and the four corner cells in a square ring are visited less
frequently than other cells in the same ring. This means that
for a hexagon ring r, r > 1, the six 3; 1 cells should carry
less weight in calculating the ar s and the br s. Similarly, for a
square ring r, r > 1, the four 5; 1 cells and the eight 3; 2
cells should carry less weight in calculating the ar s and the
br s. The consequence is that ar should be decreased and br
should be increased, i.e.,
ar

q  1r 1  q
;
2qr

and
br

br

231

q  1r  q 2 q
;
2qr

for all r > 1, where q is an appropriately chosen value to

adjust ar and br . In our study, we set 3 0:15 and 4 0:42
empirically, which are very accurate, as verified in Tables 1,
3, and 4.

232

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

TABLE 1
Comparison of Numerical Data and Experimental
Results of Nd

Let ij be the probability that in a random walk starting

from state Ki , the first entry to state Kj occurs at the nth step,
1
2
n
where n  1. In particular, 0d ; 0d ; . . . ; 0d ; ::: is called the
first-passage distribution for Kd . The following method is well
known ([21, p. 388]), which will be used in Theorems 8 and 9.
n

Theorem 7. The 0d s can be calculated by using the following

equation:
n

p0d

n
X

v nv

0d pdd

n  1;

with

1.
1

Theorem 7 implies that 0d ; 0d ; . . . ; 0d ; . . . can be

obtained successively as follows:
1

0d p0d ;
0d p0d 

n1
X

v nv

0d pdd

n  2:

v1
n

The quantity 0d is the probability that starting from ring 0,

a mobile terminal u takes exactly n steps to reach ring d for
the first time and then immediately performs location
1
2
d1
update. We notice that 0d 0d    0d 0, i.e., u
takes at least d steps to reach ring d.
n
The 0d s, where n  1, can be computed based on Pd , the
d 1  d 1 submatrix of P consisting of the first d 1
n
rows and the first d 1 columns, because the 0d s only
depend on the behavior of the Markov chain in states
K0 ; K1 ; K2 ; . . . ; Kd . (In fact, the last row of Pd , i.e., the pdj s,
are immaterial, because this finite Markov chain with d 1
1
2
states has the same first-passage distribution 0d ; 0d ; :::;
n
0d ; ::: for Kd as the original infinite Markov chain, no
matter what the pdj s are.) Such computation involves the
n
matrix powers Pd ; Pd2 ; Pd3 ; . . . ; Pdn ; . . . , to provide p0d and
n
pdd for all n  1.
The expected number of steps for a random walk starting
from K0 to reach Kd is
Nd

1
X
n1

n 0d

1
X

FEBRUARY 2013

TABLE 2
d
The Probabilities p0r

observe that Nd grows very fast as d increases. To be more

specific, we compute N 0 d when
ar br

q1
:
2q

We find that
N 0 d

dqd  1
;
q1

and

v1
0
pdd

VOL. 24, NO. 2,

n 0d :

nd

In Fig. 1 of Appendix 2, available in the online supplemental material, we display numerical values of Nd. We

1
Nd > N 0 d;
2
for all d  1. This implies that Nd Od2 . The quadratic
growth rate of Nd implies significant reduction in location
update cost in DBLMSs compared with MBLMSs (see the
comment after Theorem 12).
In Table 1, we compare the numerical data of Nd
obtained from our Markov chain and the experimental data
of Nd obtained from computer simulations. For each d, we
simulate a random walk starting from K0 until Kd is
reached and record the length of the random walk. The
average length of 500,000 random walks is reported with 99
percent confidence interval 0:25713 percent. We observe
that the numerical data of Nd obtained from our Markov
chain and the experimental results of Nd obtained from
computer simulations are extremely close, with relative
difference no more than 0:02 percent. Although the exact
values of the ar s and the br s are still mysteries to us, our
values of the ar s and the br s are able to provide very reliable
performance data.
d
In Table 2, we show p0r , that is, the probability that a
mobile terminal is in ring r after crossing cell boundaries for
d times, where 1  d  10. It is noticed that when d > 3, the
probability that a mobile terminal is in ring d after crossing
cell boundaries for d times is very small. This implies that
an MBLMS makes more location updates than necessary,
since a mobile terminal is still inside a PA with high
probability after crossing cell boundaries for d times.
In Tables 1 and 2 of Appendix 3, available in the online
supplemental material, we show numerical values of the
1
2
n
first-passage distribution 0d ; 0d ; . . . ; 0d ; ::: for Kd .

LI: ANALYSIS OF DISTANCE-BASED LOCATION MANAGEMENT IN WIRELESS COMMUNICATION NETWORKS

We would like to mention that our approach developed

in this section is also applicable to an irregular cell
structure [45], which can easily be divided into rings, as
long as the transition probabilities, i.e., the ar s and the br s,
are available [52].

E Ts , we get
Since E Tm n nE
E Tm


n
n
0d fTs s ;

fTs t s es t ;
and
fTs s

to represent the paging area residence time under the

condition that it takes n steps for a mobile terminal to
reach ring d, where d is the radius of a PA. Hence, we
obtain

n
fTs;i s fTs s :

i1
d

d1

Since the random variable n has distribution 0d ; 0d

d2
0d ; :::, we get

0d P Tm n  t;

nd

that is,
1
X

0d FTm n t;

nd

and
1
X

0d fTm n t:

nd

By the property that the Laplace transform of w1 f1 t

w1 f2 t is w1 f1 s w1 f2 s, we have
fTm s

1
X
nd

0d fTm n s:

s
;
s s

Tm n has an Erlang distribution with

fTm n t

s es t s tn1
;
n  1!


s
s s

n
;

and

Tm n Ts;1 Ts;2    Ts;n

fTm t

u
t

and Tm has a hyper-Erlang distribution with

!
1
s t
X
s tn1
n s e
;
0d
fTm t
n  1!
nd

Proof. For a fixed n, we use

FTm t

n 0d NdE
E Ts :

fTm n s

for arbitrary cell residence time Ts characterized by fTs s.

P Tm  t

1
X

and

E Tm NdE
E Ts ;

1
X

nd

When Ts has an exponential distribution with

and

n
Y

0d nE
E Ts

The theorem is proven.

nd

fTm n s

1
X

nd

Theorem 8. The paging area residence time Tm for a PA with

radius d is characterized by
1
X

0d E Tm n

nd

PAGING AREA RESIDENCE TIME

fTm s

1
X

E Ts

5.1 Ordinary Paging Area Residence Time

A paging area residence time is ordinary if the last
location update for the current paging area is ordinary. An
ordinary paging area residence time Tm is a sum of random
number n of i.i.d. random variables Ts;1 ; Ts;2 ; . . . ; Ts;n ,
which are cell residence times with fTs;i s fTs s, for
all 1  i  n. By paging area residence time, we mean
ordinary paging area residence time, unless otherwise
indicated (e.g., see Section 6.1).
The following theorem is the most important result of
our study, which will be used in Theorems 10, 11, and 12.

233

fTm s

1
X
nd

0d

s
s s

n
:

We emphasize that even when Ts has an exponential

distribution, Tm has a hyper-Erlang distribution. It is not
reasonable to assume that Tm has a simple probability
distribution such as an exponential distribution or a
hyperexponential distribution.

5.2 Modified Paging Area Residence Time

An ordinary location update is performed at the boundary of a
cell residence time. In a DBLMS-CWLU and an MBLMSCWLU, all location updates are ordinary. A modified location
update is performed in the midst of a cell residence time. In a
DBLMS-CPLU and an MBLMS-CPLU, the arrival of a
phone call initiates a modified location update. However,
all subsequent location updates before the next phone call
are ordinary.
A paging area residence time is modified if the last
location update for the current paging area is modified. A
modified paging area residence time Tm0 is a sum of random
number n of random variables Ts;1 ; Ts;2 ; . . . ; Ts;n , where
Ts;2 ; Ts;3 ; . . . ; Ts;n are cell residence times with fTs;i s
fTs s, for all 2  i  n, but Ts;1 is the residual cell residence
time with fTs;1 s Ts 1  fTs s=s.
Similar to Theorem 8, we have the following result,
which will be used in Theorems 10 and 11.

234

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

Theorem 9. The modified paging area residence time Tm0 for a PA

with radius d is characterized by
!
n1
1


X
n Ts 1  fTs sfTs s

;
fTm0 s
0d
s
nd
and
E Tm0

E Ts2
Nd  1E
E Ts ;
2E
E Ts

for arbitrary cell residence time Ts characterized by fTs s.

Proof. Let Tm0 n Ts;1 Ts;2    Ts;n . It is clear that
fTm0 n s fTs;1 s

n
Y

fTs;i s

i2

Ts 1  fTs sfTs sn1

:
s

Thus, we get
fTm0 s

1
X

1
X

6.1 The CPLU Model

In a DBLMS d under the CPLU model, when the previous
call C1 arrives, location update is performed immediately
and PA1 is treated as a new PA and the paging area
residence time Tm;1 is set to zero. However, when C1
arrives, a mobile terminal is still in the midst of a cell
residence time, which implies that Tm;1 is a modified paging
area residence time. Consequently, the sequence of random
variables Tm;1 ; Tm;2 ; Tm;3 ; . . . is a modified renewal process.
Let Xm denote the number of PA boundary crossings (i.e.,
the number of renewals) between two consecutive phone
calls (i.e., within a time interval of random length Tc ). Then,
the number of location update is Xu Xm 1, where we
include the location update performed right after C1 arrives.
The expected number of PA boundary crossings E Xm
can be obtained by using Theorem 1, where we need to
calculate

0d

nd

Ts 1  fTs sfTs s

s

FSj s
!
n1

Since

E Tm0 n

1
X

FSj s

1
X

Ht

nd

s1  fTm s

FSj t

1
2i

i1

i1

fTm0 sest
s1  fTm s

ds;

where fTm s and fTm0 s are given by Theorems 8 and 9. We

have the following result.
Theorem 10. The total cost of location management in a DBLMS
d under the CPLU model is

E Ts2
Nd  1E
E Ts :
2E
E Ts

This proves the theorem.

1
X
j1

E Ts2
n  1E
E Ts ;
2E
E Ts

0d E Tm0 n

fTm0 s

and

E Ts2
;
2E
E Ts

which gives rise to

E Xm 1 p qd2  qd 1;
DCPLU d u E
u
t

If Ts is an exponential random variable, Tm0 is identical

to Tm .

j1

(see [25, (5.15), p. 173]), we have

E Tm0

fTm0 sfTm sj1

and
:
H  s

E Ts;1

FEBRUARY 2013

same pdf fTm t. The sequence of random variables

Tm;1 ; Tm;2 ; Tm;3 ; . . . is a renewal process.

n
0d fTm0 n s

nd

VOL. 24, NO. 2,

COST OF LOCATION MANAGEMENT

In a distance-based location management scheme DBLMS

d, where d is a distance threshold and also the radius of a
PA, location update is performed as soon as a mobile
terminal moves out of the current PA. Therefore, the
number of location update between two consecutive phone
calls is related to the number of paging area boundary
crossings between two consecutive phone calls.
Let us consider two consecutive phone calls, i.e., the
previous call C1 and the next call C2 . Assume that a mobile
terminal u resides in PA1 when it receives C1 . Before u
receives the next phone call C2 , u moves through PA1 ,
PA2 ; . . . , PAXm 1 , and receives C2 in PAXm 1 , where Xm
denotes the number of PA boundary crossings. Let Tm;i
denote the residence time of u in PAi , where i  1. The Tm;i s
(except possibly Tm;1 ) are i.i.d. random variables with the

where
E Xm

HtfTc tdt;

for any fTc t and fTs t.

If the cell residence time Ts has an exponential distribution, by Theorem 8, the paging area residence time Tm has a
hyper-Erlang distribution. Furthermore, since the residual
cell residence time of an exponential Ts has the same
distribution as Ts , the modified paging area residence time
Tm0 has an identical hyper-Erlang distribution as the ordinary
paging area residence time Tm . Hence, the sequence of
random variables Tm;1 ; Tm;2 ; Tm;3 ; . . . is an ordinary renewal
process, and Theorem 2 can be applied to find FSj t.
If Tc has a hyper-Erlang distribution, the expected
number of PA boundary crossings E Xm can be obtained
using Theorem 5.
Theorem 11. If Tc has a hyper-Erlang distribution, the total cost
of location management in a DBLMS d under the CPLU
model is

LI: ANALYSIS OF DISTANCE-BASED LOCATION MANAGEMENT IN WIRELESS COMMUNICATION NETWORKS

DCPLU d u E
E Xm 1 p qd2  qd 1;

235

where

where
!

c;ic;i
wc;i
E Xm
c;i  1!
i1
#

c;i 1 "
fTm0 s
@

@s
s1  fTm s

u q  1
:

kc
X

sc;i

for any mobile terminal, where fTm s and fTm0 s are given by
Theorems 8 and 9.

6.2 The CWLU Model

In a DBLMS d under the CWLU model, when the previous
call C1 arrives, a mobile terminal u has been in PA1 for a
while. Hence, Tm;1 is the residual time of u in PA1 and does
not have the same pdf as the other Tm;i s. The sequence of
random variables Tm;1 ; Tm;2 ; Tm;3 ; . . . is an equilibrium
renewal process. According to Theorems 3 and 8, we have
ETc
ETc

ETm NdETs
 
;
1
Ts
1

Nd Tc
Nd

EXm

The performance of a DBLMS d-CWLU is determined by

d. We can find d such that the total location update and
terminal paging cost is minimized. To minimize DCWLU d,
let us consider the derivative of DCWLU d,
!
@DCWLU d
2qd  1

p q2d  1:
@d
d2 qd  12
The optimal value of d which minimizes DCWLU d in a
DBLMS d-CWLU is either bdc or dde, where d satisfies the
following equation:
2qd  1 p qd2 2d  1qd  12 :

SIMULATION RESULTS AND NUMERICAL DATA

In this section, we present simulation results and numerical

data. We consider a hyperexponential distribution of Tc
with
fTc t

kc
X

wc;i c;i ec;i t ;

i1

where  Tc =Ts is the call-to-mobility ratio. The number

of location update is Xu Xm and E Xm is the expected
number of location update between two consecutive
phone calls.
Theorem 12. For any probability distributions of Tc and Ts , the
total cost of location management in a DBLMS d under the
CWLU model is

where wc;1 wc;2    wc;kc 1, and a hypergamma distribution of Ts with

fTs t

ks
X

!
s;j es;j t s;j ts;j 1
;
s;j

ws;j

j1

and

u
p qd2  qd 1;
DCWLU d
Nd

fTs s

for any mobile terminal.

ks
X


ws;j

j1

It is proven in [29] that for any probability distributions

of Tc and Ts , the location update cost in an MBLMS d
under the CWLU model is
u
:
d
We have already known from Section 4 that
Nd


dqd  1
;
q1

where
depends on d and q but is in a small range of
0:5 <
< 0:7 for all d  2. Since Nd is significantly greater
than d even for small to moderate d, compared to an
MBLMS d-CWLU, the location update cost in a DBLMS
d-CWLU is significantly reduced.
If we set
to be a constant in 0:5; 0:7, e.g.,
0:6, then
we get a closed-form expression of DCWLU d:

p qd2  qd 1;
DCWLU d
dqd  1

s;j
s s;j

s;j
;

where ws;1 ws;2    ws;ks 1. It is clear that

!1
 !
kc
ks
X
X
wc;i
s;j
:

ws;j


c;i
s;j
i1
j1
By Theorem 11, the total cost of location management is
!!
kc
X
fTm0 c;i
wc;i
DCPLU d u 1
1  fTm c;i
i1
p qd2  qd 1;
for DBLMS d-CPLU, where
fTm c;i

1
X

n
n 
0d fTs c;i ;

nd

and
fTm0 c;i

1
X
nd

n
0d

!
Ts 1  fTs c;i fTs c;i n1
;
c;i

236

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

TABLE 3
Comparison of Analytical Data and Simulation Results
of E Xu (q 3)

and
fTs c;i

ks
X
j1


ws;j

s;j
c;i s;j

s;j
:

By Theorem 12, the total cost of location management is

DCWLU d

u
p qd2  qd 1;
Nd

for DBLMS d-CWLU.

We assume the following parameter settings. For Tc , let
kc 3, wc;1 0:25, wc;2 0:50, wc;3 0:25, c;1 c =2,
c;2 c , c;3 2c , where c is a variable. For Ts , let
ks 3, ws;1 0:25, ws;2 0:50, ws;3 0:25, s;1 s;2
s;3 s , s;1 s =2, s;2 s , s;3 2s , where s and s
are variables. It is easy to verify that E Tc 9=8c , and
E Tc c s =s .
E Ts 9s =8s , which give  E Ts =E
Furthermore, we have VarTs 19s2 72s =642s , and
the coefficient of variation of Ts ,
q 
19s2 72s 9s :
CTs
The total cost of location management is now determined
by seven variables, namely, q, d, c , s , s , p , and u . By
Theorems 11 and 12, the total cost of location management
is determined by five variables, namely, q, d, , p , and u ,
where  is given by c , s , and s . We will fix p and
examine the impact of u , and fix c and see the impact of
s and s .

7.1 Simulation Results

Extensive simulations have been conducted to validate our
analytical results on the expected number E Xu of location
updates between two consecutive phone calls. In Tables 3
and 4, we display and compare our analytical data and
simulations results for the hexagonal cell structure with q
3 and the mesh cell structure with q 4, respectively, where
c 1, s 40, and s 2. The analytical data of E Xu for
the CPLU model are calculated by using Theorem 11, and
the analytical data of E Xu for the CWLU model are
calculated by using Theorem 12. The simulation results of
E Xu are obtained by simulating a mobile terminal with
random intercall time and random cell residence time
specified by the above probability density functions fTc t

VOL. 24, NO. 2,

FEBRUARY 2013

TABLE 4
Comparison of Analytical Data and Simulation Results
of E Xu (q 4)

and fTs t. For every pair of consecutive phone calls, we

record the number of location updates between the two
consecutive phone calls. We then report the average
number of location updates between all consecutive phone
calls. The number of phone calls is as large as 500,000, such
that the maximum 99 percent confidence interval of our
simulation results is about 1:42622 percent. The relative
difference between a simulation result and its corresponding analytical datum is also given. We observe that for the
hexagonal cell structure, the relative differences of all our
simulation results are no more than 0:69 percent. For the
mesh cell structure, the relative differences of all our
simulation results are no more than 1:22 percent, except
for the case d 2. These simulation results demonstrate the
high quality and accuracy of our analytical data.

7.2 Numerical Data

In Appendix 4, available in the online supplemental
material, with Figures 2-15, we present numerical data to
show the impact of various parameters and performance
optimization.

CONCLUDING REMARKS

The paper has made several contributions to cost analysis

and performance optimization of dynamic distance-based
location management schemes in wireless communication
networks. We have developed a Markov chain as a mobility
model to describe the movement of a mobile terminal in 2D
cellular structures. We characterized the paging area
residence time for arbitrary cell residence time by using
the Markov chain. By using the classical renewal theory, we
have analyzed the expected number of paging area
boundary crossings and the cost of the distance-based
location update method for two different call handling
models. Our analytical results enable clear comparison of
the distance-based location update method and the movement-based location update method.

ACKNOWLEDGMENTS
The author wishes to thank the three anonymous reviewers
for their constructive comments.

LI: ANALYSIS OF DISTANCE-BASED LOCATION MANAGEMENT IN WIRELESS COMMUNICATION NETWORKS

REFERENCES
[1]
[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]
[12]

[13]

[14]

[15]

[16]
[17]

[18]

[19]

[20]

[21]
[22]
[23]
[24]

[25]
[26]

[27]

http://en.wikipedia.org/wiki/Renewal_theory, 2012.
A. Abutaleb and V.O.K. Li, Paging Strategy Optimization in
Personal Communication Systems, Wireless Networks, vol. 3,
no. 3, pp. 195-204, 1997.
A. Abutaleb and V.O.K. Li, Location Update Optimization in
Personal Communication Systems, Wireless Networks, vol. 3,
no. 3, pp. 205-216, 1997.
I.F. Akyildiz and J.S.M. Ho, Dynamic Mobile User Location
Update for Wireless Pcs Networks, Wireless Networks, vol. 1,
pp. 187-196, 1995.
I.F. Akyildiz, J.S.M. Ho, and Y.-B. Lin, Movement-Based Location
Update and Selective Paging for PCS Networks, IEEE/ACM
Trans. Networking, vol. 4, no. 4, pp. 629-638, Aug. 1996.
I.F. Akyildiz, J. McNair, J.S.M. Ho, H. Uzunalio
glu, and W. Wang,
Mobility Management in Next-Generation Wireless Systems,
Proc. IEEE, vol. 87, no. 8, pp. 1347-1384, Aug. 1999.
I.F. Akyildiz and W. Wang, A Dynamic Location Management
Scheme for Next-Generation Multitier PCS Systems, IEEE Trans.
Wireless Comm., vol. 1, no. 1, pp. 178-189, Jan. 2002.
E. Alba, J. Garca-Nieto, J. Taheri, and A. Zomaya, New Research
in Nature Inspired Algorithms for Mobility Management in GSM
Networks, Proc. Conf. Applications of Evolutionary Computing,
pp. 1-10, 2008.
J.H. Baek and B.H. Ryu, Modeling and Analysis of DistanceBased Registration with Implicit Registration, ETRI J., vol. 25,
no. 6, pp. 527-530, 2003.
A. Bar-Noy, Y. Feng, and M.J. Golin, Paging Mobile Users
Efficiently and Optimally, Proc. 26th IEEE Intl Conf. Computer
Comm., pp. 1910-1918, 2007.
A. Bar-Noy, I. Kessler, and M. Sidi, Mobile Users: To Update or
Not Update? Wireless Networks, vol. 1, pp. 175-185, 1995.
F.V. Baumann and I.G. Niemegeers, An Evaluation of Location
Management Procedures, Proc. IEEE Intl Conf. Universal Personal
Comm., pp. 359-364, 1994.
T.X. Brown and S. Mohan, Mobility Management for Personal
Communications Systems, IEEE Trans. Vehicular Technology,
vol. 46, no. 2, pp. 269-278, May 1997.
V. Casares-Giner and J. Mataix-Oltra, Global versus DistanceBased Local Mobility Tracking Strategies: A Unified Approach,
IEEE Trans. Vehicular Technology, vol. 51, no. 3, pp. 472-485, May
2002.
E. Cayirci and I.F. Akyildiz, User Mobility Pattern Scheme for
Location Update and Paging in Wireless Systems, IEEE Trans.
Mobile Computing, vol. 1, no. 3, pp. 236-247, July-Sept. 2002.
D.R. Cox, Renewal Theory. Methuen & Co., Ltd., 1962.
P.G. Escalle, V.C. Giner, and J.M. Oltra, Reducing Location
Update and Paging Costs in a PCS Network, IEEE Trans. Wireless
Comm., vol. 1, no. 1, pp. 200-209, Jan. 2002.
Y. Fang, General Modeling and Performance Analysis for
Location Management in Wireless Mobile Networks, IEEE Trans.
Computers, vol. 51, no. 10, pp. 1169-1181, Oct. 2002.
Y. Fang, Movement-Based Mobility Management and Trade off
Analysis for Wireless Mobile Networks, IEEE Trans. Computers,
vol. 52, no. 6, pp. 791-803, June 2003.
Y. Fang, I. Chlamtac, and Y.-B. Lin, Portable Movement
Modeling for PCS Networks, IEEE Trans. Vehicular Technology,
vol. 49, no. 4, pp. 1356-1363, July 2000.
W. Feller, An Introduction to Probability Theory and Its Applications,
third ed., vol. 1. John Wiley & Sons, 1968.
Wireless Internet Handbook - Technologies, Standards, and Applications, B. Furht, M. Ilyas, eds. CRC Press, 2003.
S.G. Glisic, Advanced Wireless Networks - 4G Technologies. John
Wiley & Sons, 2006.
J.S.M. Ho and I.F. Akyildiz, Mobile User Location Update and
Paging under Delay Constraints, Wireless Networks, vol. 1,
pp. 413-425, 1995.
L. Kleinrock, Queueing Systems: Theory, vol. 1. John Wiley & Sons,
1975.
N.E. Kruijt, D. Sparreboom, F.C. Schoute, and R. Prasad, Location
Management Strategies for Cellular Mobile Networks, IEE
Electronics & Comm. Eng. J., vol. 10, no. 2, pp. 64-72, Apr. 1998.
J. Li, H. Kameda, and K. Li, Optimal Dynamic Mobility
Management for PCS Networks, IEEE/ACM Trans. Networking,
vol. 8, no. 3, pp. 319-327, June 2000.

237

[28] J. Li, Y. Pan, and X. Jia, Analysis of Dynamic Location

Management for PCS Networks, IEEE Trans. Vehicular Technology, vol. 51, no. 5, pp. 1109-1119, Sept. 2002.
[29] K. Li, Cost Analysis and Minimization of Movement-Based
Location Management Schemes in Wireless Communication
Networks: A Renewal Process Approach, Wireless Networks,
vol. 17, no. 4, pp. 1031-1053, 2011.
[30] Y.-B. Lin, Reducing Location Update Cost in a PCS Network,
IEEE/ACM Trans. Networking, vol. 5, no. 1, pp. 25-33, Feb. 1997.
[31] U. Madhow, M.L. Honig, and K. Steiglitz, Optimization of
Wireless Resources for Personal Communications Mobility Tracking, IEEE/ACM Trans. Networking, vol. 3, no. 12, pp. 698-707, Dec.
1995.
[32] C.K. Ng and H.W. Chan, Enhanced Distance-Based Location
Management of Mobile Communication Systems Using a Cell
Coordinates Approach, IEEE Trans. Mobile Computing, vol. 4,
no. 1, pp. 41-55, Jan./Feb. 2005.
[33] R.M. Rodrguez-Dagnino and H. Takagi, Movement-Based
Location Management for General Cell Residence Times in
Wireless Networks, IEEE Trans. Vehicular Technology, vol. 56,
no. 5, pp. 2713-2722, Sept. 2007.
[34] C. Rose, Minimizing the Average Cost of Paging and Registration: A Timer-Based Method, Wireless Networks, vol. 2, no. 2,
pp. 109-116, 1996.
[35] C. Rose, State-Based Paging/Registration: A Greedy Technique,
IEEE Trans. Vehicular Technology, vol. 48, no. 1, pp. 166-173, Jan.
1999.
[36] C. Rose and R. Yates, Minimizing the Average Cost of Paging
under Delay Constraints, Wireless Networks, vol. 1, pp. 211-219,
1995.
[37] C.U. Saraydar, O.E. Kelly, and C. Rose, One-Dimensional
Location Area Design, IEEE Trans. Vehicular Technology, vol. 49,
no. 5, pp. 1626-1632, Sept. 2000.
[38] R. Subrata1 and A.Y. Zomaya, Dynamic Location Management
for Mobile Computing, Telecomm. Systems, vol. 22, nos. 1-4,
pp. 169-187, 2003.
[39] S. Tabbane, Location Management Methods for Third-Generation Mobile Systems, IEEE Comm. Magazine, vol. 35, no. 8, pp. 7284, Aug. 1997.
[40] J. Taheri and A.Y. Zomaya, A Simulated Annealing Approach for
Mobile Location Management, Computer Comm., vol. 30, no. 4,
pp. 714-730, 2007.
[41] J. Taheri and A.Y. Zomaya, A Combined Genetic-Neural
Algorithm for Mobility Management, J. Math. Modelling and
Algorithms, vol. 6, no. 3, pp. 481-507, 2007.
[42] J. Taheri and A.Y. Zomaya, Clustering Techniques for Dynamic
Location Management in Mobile Computing, J. Parallel and
Distributed Computing, vol. 67, no. 4, pp. 430-447, 2007.
[43] J. Taheri and A.Y. Zomaya, Bio-Inspired Algorithms for Mobility
Management, Proc. Intl Symp. Parallel Architectures, Algorithms,
and Networks, pp. 216-223, 2008.
[44] M. Verkama, Optimal Paging - A Search Theory Approach,
Proc. IEEE Intl Conf. Universal Personal Comm., pp. 956-960, 1996.
[45] G. Wan and E. Lin, Cost Reduction in Location Management
Using Semi-Realtime Movement Information, Wireless Networks,
vol. 5, pp. 245-256, 1999.
[46] W. Wang, I.F. Akyildiz, G.L. Stuber, and B.-Y. Chung,
Effective Paging Schemes with Delay Bounds as QoS
Constraints in Wireless Systems, Wireless Networks, vol. 7,
pp. 455-466, 2001.
[47] X. Wang, P. Fan, J. Li, and Y. Pan, Modeling and Cost Analysis of
Movement-Based Location Management for PCS Networks with
HLR/VLR Architecture, General Location Area and Cell Residence time Distributions, IEEE Trans. Vehicular Technology,
vol. 57, no. 6, pp. 3815-3831, Nov. 2008.
[48] V.W.S. Wong and V.C.M. Leung, An Adaptive Distance-Based
Location Update Algorithm for Next-Generation PCS Networks,
IEEE J. Selected Areas in Comm., vol. 19, no. 10, pp. 1942-1952, Oct.
2001.
[49] A. Yener and C. Rose, Highly Mobile Users and Paging: Optimal
Polling Strategies, IEEE Trans. Vehicular Technology, vol. 47, no. 4,
pp. 1251-1257, Nov. 1998.
[50] Y. Zhang, J. Zheng, L. Zhang, Y. Chen, and M. Ma, Modeling
Location Management in Wireless Networks with Generally
Distributed Parameters, Computer Comm., vol. 29, no. 12,
pp. 2386-2395, 2006.

238

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

[51] Y.-h. Zhu and V.C.M. Leung, Derivation of Moving Distance

Distribution to Enhance Sequential Paging in Distance-Based
Mobility Management for PCS Networks, IEEE Trans. Wireless
Comm., vol. 5, no. 11, pp. 3029-3033, Nov. 2006.
[52] Y.-h. Zhu and V.C.M. Leung, Optimization of Distance-Based
Location Management for PCS Networks, IEEE Trans. Wireless
Comm., vol. 7, no. 9, pp. 3507-3516, Sept. 2008.

VOL. 24, NO. 2,

FEBRUARY 2013

Keqin Li is a SUNY distinguished professor of

computer science in the State University of New
York at New Paltz. He is also an Intellectual
Ventures endowed visiting chair professor at the
National Laboratory for Information Science and
Technology, Tsinghua University, Beijing, China.
His research interests are mainly in design and
analysis of algorithms, parallel and distributed
computing, and computer networking. He has
contributed extensively to processor allocation
and resource management; design and analysis of sequential/parallel,
deterministic/probabilistic, and approximation algorithms; parallel and
distributed computing systems performance analysis, prediction, and
evaluation; job scheduling, task dispatching, and load balancing in
heterogeneous distributed systems; dynamic tree embedding and
randomized load distribution in static networks; parallel computing using
optical interconnections; dynamic location management in wireless
communication networks; routing and wavelength assignment in optical
networks; energy-efficient computing and communication. His current
research interests include lifetime maximization in sensor networks, file
sharing in peer-to-peer systems, power management and performance
optimization, and cloud computing. He has published more than 240
journal articles, book chapters, and research papers in refereed
international conference proceedings. He has received several Best
Paper Awards for his highest quality work. He has served in various
capacities for numerous international conferences as general chair,
program chair, workshop chair, track chair, and steering/advisory/award/
program committee member. He is currently on the editorial board of
IEEE Transactions on Parallel and Distributed Systems, IEEE Transactions on Computers, Journal of Parallel and Distributed Computing,
International Journal of Parallel, Emergent and Distributed Systems,
International Journal of High Performance Computing and Networking,
and Optimization Letters. He is a senior member of the IEEE.

. For more information on this or any other computing topic,

please visit our Digital Library at www.computer.org/publications/dlib.