INT J COMPUT COMMUN, ISSN 1841-9836
Vol.7 (2012), No. 2 (June), pp. 231-240

The Effect of Heterogeneous Traffic Distributions on Load
Balancing in Mobile Communications: An Analytical Model

K.-C. Chu, C.-S. Wang, W.-W. Jiang, N.-C. Hsieh

Kuo-Chung Chu, Wey-Wen Jiang, Nan-Chen Hsieh
Department of Information Management,
National Taipei University of Nursing and Health Sciences
No.365, Mingde Rd., Beitou Dist.,
Taipei City 11219, Taiwan (R.O.C.)
{kcchu, jiang6, nchsie})@ntunhs.edu.tw

Chun-Sheng Wang
Department of Information Management,
Jinwen University of Science and Technology
No.99, Anzhong Rd., Xindian Dist.,
New Taipei City 23154, Taiwan (R.O.C.)
seanwang@just.edu.tw

Abstract: This paper investigates the load balancing problem in an environ-
ment of heterogeneous traffic distributions. An analytical model is proposed
to determine the effect of heterogeneous traffic distributions on load balancing,
in which a generic measure of load balancing level (LBL) that is a function
of traffic type coefficient (TYC) and call blocking probability of cells is to an-
alyze the expected level of the load balance. We consider both voice traffic
and data traffic to determine which kind of traffic has the greater effect. The
performance of cellular systems with sectorization is evaluated; they are nor-
mal case (N) of homogeneous distribution and linear case (L) of heterogeneous
distribution. The analysis results indicate that the TYC has a significant ef-
fect on the accommodation capacity, in which voice calls outperform data calls
because the LBL can easily distinguish between normal and linear distribu-
tions. Load balancing can be achieved more easily for voice only traffic than
for data only traffic. Sectorization is more effective in achieving load balancing
in the scenario of the heavier loads than in the lighter loads. The paper results
are useful for network planning to optimize the channel allocation for different
traffic type’s distribution.
Keywords: analytical model, heterogeneous distributions, load balancing,
mobile communications, QoS.

1 Introduction

Several studies have evaluated the capacity of mobile cellular systems, but most of them (e.g.,
[1,2]) assume a homogeneous spatial traffic distribution, as it best fits the system’s characteristics
to have all signals share all the spectral resources However, homogeneous traffic distribution
among base station (BS)/sectors/cells (equal cell loads) is very uncommon in practice. Even
though sufficient capacity is planned in a cellular system, heterogeneous traffic distribution may
occur in other cells, creating a “hot spot" that exceeds the pre-determined capacity and introduces
a large blocking probability, as the quality of service (QoS) of such cells may be degraded,
especially below a pre-defined threshold. This is why load balancing is the most important issue
to be discussed before network planning in terms of optimal resource allocation.

Copyright c⃝ 2006-2012 by CCC Publications



232 K.-C. Chu, C.-S. Wang, W.-W. Jiang, N.-C. Hsieh

In [3], Ning et al. discuss load balancing by using a hybrid scheme of channel borrowing
scheme and load transfer, it allows borrowing channels from light load cells, and ongoing calls
can be transferred from heavy load cells into the overlapping cells they are light load. To
improve global resource utilization and reduce regional congestion given heterogeneous arrivals,
[4] requires load balancing among multiple cells. However, their works cannot be applied to
general system because a lot of issues are different from other systems (WCDMA, CDMA2000,
HSPA), e.g. channel definition, interferences, soft handoff. In general, soft handoff enforced
by power control has been proposed as a possible solution to local traffic imbalances among
cells [5]. Actually, power control is one of the most important processes, as interference is the
predominant factor that affects the capacity and signal-to-interference ratio (SIR). To maximize
system capacity, power control can be used efficiently to adapt cell sizes for load balancing;
however, the trade-off between coverage and capacity should be carefully considered [6, 7]. An
adaptive load-shedding scheme combines the power control and the soft handoff function to force
some mobile stations (MSs) farthest from the cell to enter forced soft handoff, and transfer
their traffic load to neighboring cells that are lightly loaded. In this way, heavily loaded cells
dynamically down-size their coverage area in order to handle traffic, while adjacent cells that
are less heavily loaded increase their coverage to accommodate the extra traffic. However, in
a hot-spot sector, powering up all MSs in the sector results in excessive interference with the
MSs in neighboring cells, so they cannot maintain sufficient SIR levels at their sector sites.
Previous studies also attempt to achieve constant received mean power from each MS within a
sector [8–10].

This paper investigates the load balancing problem in an environment of heterogeneous traffic
distributions. An analytical model is proposed to determine the effect of heterogeneous traffic
distributions on load balancing, in which a generic measure of load balancing level (LBL) that is
a function of traffic type coefficient (TYC) and call blocking probability of cells is to analyze the
expected level of the load balance. We consider both voice traffic and data traffic to determine
which kind of traffic has the greater effect. The performance of cellular systems with sectoriza-
tion is evaluated; they are normal case (N) of homogeneous distribution and linear case (L) of
heterogeneous distribution.

The remainder of this paper is organized as follows. In Section 2, we discuss the mobile
cellular system background. In Section 3, we present an analytical model of load balancing,
and define SIR. Section 4 details the numerical results, and Section 5 contains some concluding
remarks.

2 Background of Mobile Cellular Systems

2.1 Sectorization

Generally speaking, a BS configuration is uniformly sectorized in one sector (with omni-
directional antenna, 360◦ per sector), in three sectors (120◦ per sector), and in six sectors (60◦

per sector). The capacity of each sector is calculated subject to the system’s SIR requirements.
A sector that is lightly loaded usually experiences more interference than a heavily loaded sector,
which leads to a higher blocking probability in the lightly loaded sector. Denote B as a set of
BSs, and S as a set of sectors configured in the BS. We further denote the sector s in BS j as
sectorjs (∀s ∈ S, j ∈ B) and K as the set of sector configurations. In this paper, the following
two probable configurations are given for a BS (|K| = 2): a single sector configuration with an
omni-directional antenna, (360◦ per sector), and a three-sector configuration (120◦ per sector); k
is assigned as the identification (ID) for each configuration. The sector ID i identifies the sector
in the configuration k in an anti-clockwise direction. Table 1 summarizes the sector candidates



The Effect of Heterogeneous Traffic Distributions on Load Balancing in Mobile
Communications: An Analytical Model 233

Table 1: Sector candidates for the BS
The value of S Candidate sk,i Configuration I.D. (k) Sector I.D. (i)

S = 1 S(1,1) 1 1
S = 2 S(2,1) 2 1
S = 3 S(2,2) 2 2
S = 4 S(2,3) 2 3

Table 2: Coverage of candidate sectors
Candidate Sk,i Sector I.D.i Coverage of Sk,i

1 1 (ϕ,ϕ + 360◦)
2 1 (ϕ,ϕ + 120◦)
3 2 (ϕ + 120◦,ϕ + 240◦)
4 3 (ϕ + 240◦,ϕ + 360◦)

for each BS for a combination of k and i. Let S be the set of sectors; then, each sector sk,i
(∀sk,i ∈ S) is defined by the sector configuration (k) and the sector ID (i). Table 2 details the
coverage (in degrees) of each sector, where ϕ is the degree of the baseline. In general, it can be
assigned arbitrarily, but in this paper we given −30◦ in our cellular structure example, as shown
in Figure 1.

baseline

axis-x (0 )

30f = -

R

Figure 1: The baseline deployed in a cellular structure

2.2 Interference between sectors

To calculate the interference between sectors, the sector configuration information in Table
1 and the sector coverage information in Table 2 must be given. Without loss of generality,
sector sk,i(sk′,i′) is replaced by s(s′); and sector s in BS j is denoted by sectorjs, as shown in
Figure 2. Because of the baseline degree deployed in all cells is the same using ϕ = −30◦, no
matter what the BS is configured, the mutual interference between sectors can be well-known.
If we define the interference indicator functions ΩULjsj′s′ and Ω

DL
jsj′s′ for the respective uplink (UL)

and downlink (DL) connections between sectorjs and sectorj′s′, they can be pre-calculated. To
pre-calculate the indicator functions, the sector candidates to be configured in the BS must be
defined. Assuming (xjs,yjs) and (xj′s′,yj′s′) are the respective locations of BS j and BS j′, the
vectors

−→
A1(

−→
A′1) and

−→
A2(

−→
A′2) covering sectorjs(sectorj′s′) are defined as follows:

−→
A1 =

[
x1js − xjs,y

1
js − yjs

]
,
−→
A2 =

[
x2js − xjs,y

2
js − yjs

]
−→
A′1 =

[
x1j′s′ − xj′s′,y

1
j′s′ − yj′s′

]
,
−→
A′2 =

[
x2j′s′ − xj′s′,y

2
j′s′ − yj′s′

]



234 K.-C. Chu, C.-S. Wang, W.-W. Jiang, N.-C. Hsieh

1, ' 'j s
q

2, ' 'j s
q

1, js
q

2, js
q

s e c t o r  b o u n d a r y

a u x i li a r y  i n t e r f e r e n c e

a x i s -x  ( 0 ))

js
q

( ),tMS tx ty

( )' ', 'tMS tx ty

sector
js

' '
sector

j s

' 'j s
q

( )' ' ' ',j s j sx y

( ),js jsx y

( )1 1' ' ' ',j s j sx y

( )2 2' ' ' ',j s j sx y

( )1 1,js jsx y

( )2 2,js jsx y

Figure 2: Mutual interference between sectors

where

x1js = Rj cos (θ1,js) + xjs,y
1
js = Rj sin (θ1,js) + yjs,

x2js = Rj cos (θ2,js) + xjs,y
2
js = Rj sin (θ2,js) + yjs

x1j′s′ = Rj′ cos (θ1,j′s′) + xj′s′,y
1
j′s′ = Rj′ sin (θ1,j′s′) + yj′s′,

x2j′s′ = Rj′ cos (θ2,j′s′) + xj′s′,y
2
j′s′ = Rj′ sin (θ2,j′s′) + yj′s′

Then, the DL interference ΩDLjsj′s′ between sectorjs and sectorj′s′ can be analyzed by an auxil-
iary vector

−→
A3 = [tx′ − xjs,ty′ − yjs], where (tx′, ty′) is the arbitrary position of MS t′ (∀t′ ∈ T

and T is the set of MSs) serviced by sectorj′s′. Furthermore, (θ1,js,θ2,js) and (θ1,j′s′,θ2,j′s′) can
also be calculated easily. According to

−→
A3, θjs is calculated by θjs = tan−1

ty′−yjs
tx′−xjs

,0 ≤ θjs < 360◦.
After calculating θjs, the Algorithm Cal_InterF2 is applied to calculate ΩDLjsj′s′ Meanwhile,

the UL interference ΩULjsj′s′ between sectorjs and sectorj′s′ can be analyzed by an auxiliary vector−→
A′3 = [tx − xj′s′,ty − yj′s′], where (tx,ty) is the arbitrary position of MS t(∀t ∈ T) serviced by
sectorjs; θj′s′ is calculated by θj′s′ = tan−1

ty−yj′s′
tx−xj′s′

,0 ≤ θj′s′ < 360◦. Again, applying Algorithm
Cal_InterF to the calculation of UL interference ΩULjsj′s′.

3 Analytical Model of Load Balancing

3.1 SIR definition

Denote zjst as a decision variable, which is 1 if MS t is admitted by sectorjs subject to the
SIR requirements and 0 otherwise. Assuming the power of both the UL and DL are perfectly
controlled, the received power in sectorjs from MS t with constant value PULc(t) will be in the same
traffic class−c(t) in the UL, and the received power at MS t from sectorjs with constant value
PDL
c(t)

will be in same traffic class−c(t) in the DL. If Djt is the distance from MS t to sectorjs,
and given an attenuation factor τ = 4 which is the degree to which a beam of radiation has
been attenuated, the intra-sector interference on the UL and the DL is given by (1) and (2)
respectively, where both αUL

c(t)
and αDL

c(t)
are activity factors of traffic class−c(t). The inter-sector

2Detailed algorithm procedure is omitted due to the length limitation of the paper. A complete version of the
procedure is available upon request.



The Effect of Heterogeneous Traffic Distributions on Load Balancing in Mobile
Communications: An Analytical Model 235

interference on the UL and the DL is expressed by (3) and (4) respectively, where both ΩULj′s′js
and ΩDLj′s′js indicate the interference between sectors.

IULjst,intra =
∑

t′∈T
t′≠t

αULc(t′)P
UL
c(t′)zjst′ (1)

IDLjst,intra =
∑

t′∈T
t′̸=t

αDLc(t′)P
DL
c(t′)

(
Djt′
Djt

)τ
zjst′ (2)

IULjst,inter =
∑
j′∈B
j′̸=j

∑
s′∈S
s′≠s

∑
t′∈T
t′̸=t

ΩULj′s′jsα
UL
c(t′)P

UL
c(t′)

(
Dj′t′
Djt′

)τ
zj′s′t′ (3)

IDLjst,inter =
∑
j′∈B
j′̸=j

∑
s′∈S
s′≠s

∑
t′∈T
t′̸=t

ΩDLj′s′jsα
DL
c(t′)P

DL
c(t′)

(
Dj′t′
Dj′t

)τ
zj′s′t′ (4)

SIRULjs,c(t) =
WUL

dUL
c(t)

·
PUL
c(t)

+ (1 − zjst)V
(1 − ρUL)IULjst,intra + I

UL
jst,inter

(5)

SIRDLjs,c(t) =
WDL

dDL
c(t)

·
PDL
c(t)

+ (1 − zjst)V
(1 − ρDL)IDLjst,intra + I

DL
jst,inter

(6)

Let WUL(WDL) be the spectrum allocated to the UL (DL), and dUL
c(t)

(dDL
c(t)

) be the information
rate in the UL (DL). The SIR values SIRUL

js,c(t)
and SIRDL

js,c(t)
in the UL and the DL are defined

in (5) and (6) respectively, where ρUL (ρDL) is the UL (DL) orthogonality factor. Equations
(5) and (6) give a very large artificial constant value V in the numerator in order to satisfy the
SIR constraints. This is because the SIR value must be larger than a pre-defined threshold,
say the bit energy to noise ratio (BENR), if MS t is to be admitted by sectorjs(zjst = 1); in
other words, the constraint BENR 5 SIR must be satisfied. For example, in the UL in Equation
(5), if MS t is to be admitted by sectorjs(zjst = 1), the SIR value SIRULjs,c(t) is calculated by
(WUL/dUL

c(t)
) · PUL

c(t)
/((1 − ρUL)IULjst,intra + I

UL
jst,inter) to determine whether the SIR constraint can

be satisfied. In contrast, if MS t (zjst = 0) is rejected, the SIR value is always larger than
BENR (BENR ≪ SIR) because the value V is dominant PUL

c(t)
; thus, SIRUL

js,c(t)
is calculated as a

very large value. This implies that the constraint BENR 5 SIR can be ignored, as it is always
satisfied.

3.2 The Analytical Model

In this paper, we consider traffic with multiple classes, and use the Kaufman model [12] as a
performance measure to analyze the blocking probability of each traffic class effectively. Assume
that M channels are shared by all traffic requirements. Then, for each traffic class−c(∀c ∈ C)
with distinct channel requirements, the traffic arrival is a stationary Poisson process with mean
rate λ; and the channel requirement b is an arbitrary discrete random variable (Prob{b = bc} =
qc,∀c ∈ C). A call request with channel requirement bc has a mean holding time of 1/µc. Thus,
traffic with channel requirement bc is generated in the Poisson arrival process with mean rate
λc = λqc and the class−c offered load ac = λc/µc. The blocking probability of traffic class−c is
defined in (7) [12], where the distribution of q(·), which is the probability of the total number of
channels occupied by the complete sharing policy, satisfies Equation (8) [11], and q(x) = 0 for
x < 0 and

∑M
j=0 q(x) = 1.

Bc (a,b) =
∑bc−1

i=0
q (|M| − i) ∀c ∈ C (7)



236 K.-C. Chu, C.-S. Wang, W.-W. Jiang, N.-C. Hsieh

∑
c∈C

acbcq (j − bc) = jq(j) j = 0,1, . . . ,M (8)

To deal with variations in the traffic load, we can seek load balancing with an average value
of resource utilization [12]. In order to evaluate the experiment results, we define a diversity
function for system load balancing, from which the standard deviation (SD) of the call blocking
probability among sectors can be derived. The smaller the SD, the better the balancing results
will be. Let gcjs =

∑
t∈T zjst/µc(t) be the traffic intensity of class−c. Then, gjs =

∑
c∈C g

c
js is the

aggregate traffic (in Erlangs) in sectorjs, where gjs is equivalent to the traffic load a in (7). We
also define mjs =

∑
t∈T zjstm

c(t) as the total number of channels allocated in sectorjs, where mjs
is equivalent to the required channels b in (7), and mc(t) is the number of channels required for
traffic class−c(t). The performance measure Bcjs (the call blocking probability of traffic class−c
in sectorjs) is expressed by (7), where the sub-script js in Bjs indicates Bc in sectorjs. If we
define LLB as the LBL, the load balancing model can be formulated as (9), where SD

(
Bcjs

)
is

the SD function of Bcjs. The model is calculated subject to SIR Constraints (5) and (6).

LLB =
∑

j∈B

∑
s∈S

∑
c∈C

KcSD
(
Bcjs

)
∀j ∈ B,s ∈ S (9)

To assess the impact of different traffic types on load balancing, we denote Kc as a ratio of
traffic class−c, where

∑
c∈C K

c = 1. Kc is a traffic type coefficient (TYC) used to analyze the
expected level of the load balance. Given two classes of call requests, e.g., voice and data traffic,
if Kv = 1 and Kd = 0, we only investigate the effect of voice traffic on load balancing; however,
if Kv = 0 and Kd = 1, we only investigate the effect of data traffic on load balancing.

4 Numerical Results

4.1 Parameters

Two heterogeneous traffic distributions are considered in the structure of a 5 × 5 two-
dimensional array with hexagonal cells, and their impact on the system’s load balance is compared
with that of a homogeneous distribution between sectors, as shown in Figure 3. In the figure,
each dark cell has an heterogeneous load that is either heavier or lighter than the load of the nor-
mal cells (the light cells). It is assumed that the user density in each cell is homogeneous. Each
cell is configured with 3 sectors (|S| = 3), and assigned a radius Rjs = 5.0km. The required
BENR for voice (v) and data (d) traffic is given by (Eb/NTOTAL)

UL
v = (Eb/NTOTAL)

DL
v =

7dB and (Eb/NTOTAL)
DL
v = (Eb/NTOTAL)

DL
d = 10dB respectively [13]. The information

rates dULv = d
DL
v = 9.6bps, d

UL
d = 19.2bps, d

DL
d = 38.4bps [14–16], and the activity factors

αULv = α
DL
v = α

UL
d = α

DL
d = 0.5 [13, 16, 17] are also given. The number of channels required

is mv = 1,md = 4, and the orthogonality factor is ρUL = 0.9,ρDL = 0.7 [13]; and the power is
perfectly controlled by PULv = 10dB, P

DL
v = 15dB, P

UL
d = 15dB, P

DL
d = 20dB. The assigned

service rate is Φvjs = Φ
d
js = 0.1 [17, 18].

4.2 Traffic Models

For each sector, call requests for both voice and data calls are generated in the Poisson arrival
process with λv and λd respectively. The mean call holding time is given as 1/µv = 180(sec),
1/µd = 600 (sec) [18]. Denote (Eb/NTOTAL)

UL
c(t) ≤ SIR

UL
js,c(t)

and (Eb/NTOTAL)
DL
c(t) ≤ SIR

DL
js,c(t)

as the QoS requirements of the UL and DL respectively. All traffic calculated in gjs must satisfy
the QoS requirements and the condition zjstDjt ≤ Rjsδjst, where δjst is the indicator function
if MS t is in the coverage of sectorjs. Power is perfectly controlled in both the UL and the



The Effect of Heterogeneous Traffic Distributions on Load Balancing in Mobile
Communications: An Analytical Model 237

22 23 2524

20

14 1513

9 10

191816 17

11 12

21

3 42

87

51

(a) Heterogeneous linear model

    

22 23 2524

20

14 1513

9 10

191816 17

11 12

21

3 42

87

51

(b) Heterogeneous hot spot model

22 23 2524

20

14 1513

9 10

191816 17

11 12

21

3 42

87

51

(c) Uniform model

Figure 3: Traffic distribution scenarios

DL, and soft handoff is not taken into account. The traffic distributions considered in this work
are uniform (U), hot spot (H), and linear (L), as shown in Figure 3. Recall that the cells with
heterogeneous loads in Figure 3 (a) and Figure 3 (b) have either heavier or lighter loads than
normal cells in a homogeneous distribution. To evaluate the heterogeneous scenario, we introduce
two traffic models. We denote heterogeneous cells with heavier loads as M1 and heterogeneous
cells with lighter loads as M2. If normal cells are given traffic arrivals λc for traffic class−c,
arrivals in heterogeneous cells are assigned 200% of λc in M1, and 50% of λc in M2. Thus, the
level of load balance for traffic with multiple classes can be evaluated effectively in a near-realistic
environment.

4.3 Analysis

Without loss of generality, the level of load balance is represented in logarithmic form
log (FLB). If a smaller value of log (FLB) is calculated, a better level of load balance will be
achieved. In Figure 4 (a), no matter what the distribution (linear, hot spot, or uniform) and the
offered voice arrivals λv are, log (FLB) is a decreasing function of Kv. This implies that load
balancing can be achieved more easily for voice only traffic than for data only traffic. If only
data traffic is considered, given Kv = 0 for all distributions, log (FLB) is nearly −1.75, whereas
log (FLB) is nearly −2.4 if only voice traffic is considered (Kv = 1). With regard to the effect
of traffic intensity, it is easier to achieve load balancing with more offered voice traffic than less
offered traffic. For example, in Figure 4 (a), given Kv = 0.5 with λd = 6, log (FLB) calculates
(−1.2,−1.6,−1.9) for arrivals (λv =12, 30, 48). Again, given λv = 12 in Figure 4 (a), log (FLB)
is in the range −1.15 to −1.4 in Figure 4 (a), but it is in the range −1.7 to −1.75 in Figure 4(b)
with λd = 24.

For the traffic models (M1 vs. M2), there is no significant difference in the load balance with
λv = 12 and λd = 6 in both Figure 4 (a) and Figure 5 (a). However, given λd = 6 in Figure 5
(a), the load balance level varies in heavily loaded voice traffic (λv = 30,48). In another case,
given λd = 24 in Figure 5 (b), log (FLB) calculates the same results for variations in the load
balance. From the analysis, we conclude that the level of load balancing is more stable in M1
than in M2. A better scheme is needed to handle load balancing in cases of heterogeneous cells
with light traffic loads.

5 Conclusion

In this paper, we propose a load balancing model to deal with the ever-increasing number
of heterogeneous distributions in mobile wireless communication systems. We have studied the



238 K.-C. Chu, C.-S. Wang, W.-W. Jiang, N.-C. Hsieh

lo
g

(F
L
B
)

   

     

   

Linear

Hot Spot

Uniform

O

´

D

v
K

12
v
l =

30
v
l =

48
v
l =

(a) λd = 6

lo
g

(F
L
B
)

   

     

   

Linear

Hot Spot

Uniform

O

´

D

v
K

1212
v
ll =

30
v
l =

48
v
l =

12

(b) λd = 24

Figure 4: BLF as a function of BLC Kv with respect to λv, given traffic model M1 and |S| = 3

lo
g

(F
L
B
)

   

     

   

Linear

Hot Spot

Uniform

O

´

D

v
K

12
v
l =

30
v
l =

48
v
l =

(a) λd = 6

lo
g

(F
L
B
)

   

     

   

Linear

Hot Spot

Uniform

O

´

D

v
K

12
v
l =

30
v
l =

48
v
l =

 

(b) λd = 24

Figure 5: BLF as a function of BLC Kv with respect to λv, given traffic model M2 and |S| = 3

effect of heterogeneous traffic distributions on load balancing, as well as the effect of sectoriza-
tion. The numerical results indicate that the level of load balancing is affected by spatial traffic
distributions, especially by lighter loads in heterogeneous cells (the M2 model). In the scenario
of heterogeneous cells with heavier loads (the M1 model), the level of load balancing has similar
values of log (FLB) in three distributions, i.e., the linear, hot spot, and uniform models. Sector-
ization is more effective in achieving load balancing in the scenario of the heavier loads than in
the lighter loads. To achieve load balancing as well as capacity maximization in a system with
heterogeneous distributions, a hybrid FMDA/CDMA scheme can be utilized. Usually, available
wideband spectrum can be divided into a number of subspectra with smaller bandwidths; each
of them is further deployed by CDMA technique. Each subspectrum employs direct sequence
spectrum spreading with reduced processing gain, which is transmitted in one and only one
subspectrum. The scheme moderately mitigates interference by allocating an appropriate sub-
spectrum in each cell. The results of this work are useful for network planning to optimize the
channel allocation for different traffic type’s distribution.



The Effect of Heterogeneous Traffic Distributions on Load Balancing in Mobile
Communications: An Analytical Model 239

Bibliography

[1] J.-S. Wu and J.-K. Chung, Analysis of uplink and downlink capacities for two-tier cellular
system, IEEE Proc.-Communications., 144(6):405-411, 1997

[2] J.C. Liberti, T. S. Rappaport, Analytical Results for Capacity Improvements in CDMA,
IEEE Transactions on Vehicular Technology, vol. 43, pp. 680-690, Sep. 1994

[3] G. Ning, G. Zhu, L. Peng, and X. Lu, Research on hybrid dynamic load balancing algorithm
in heterogeneous hierarchical wireless networks, Journal on Communication, 28(1):75-81+86,
2007

[4] A. Sang, X. Wang, M. Madihian, and R. D. Gitlin, Coordinated load balancing, handoff/cell-
site selection, and scheduling in multi-cell packet data systems, Wireless Networks, 14(1):103-
120, 2008

[5] X.H. Chen, Adaptive traffic-load shedding and its capacity gain in CDMA cellular systems,
in Proc. IEE Communications, pp. 186-192, 1995

[6] V.V. Veeravalli and A. Sendonaris, The coverage-capacity tradeoff in cellular CDMA systems,
IEEE Transactions on Vehicular Technology, 48(5):1443-1450, 1999

[7] G. Hampel, K.L. Clarkson, J.D. Hobby, and P.A. Polakos, The tradeoff between coverage
and capacity in dynamic optimization of 3G cellular networks, in Proc. IEEE VTC-Fall, vol.
2, pp. 927-932, 2003

[8] W.-M. Tam and F.C.M. Lau, Analysis of power control and its imperfections in CDMA
cellular systems, IEEE Transactions on Vehicular Technology, 48(5):1706-1717, 1999

[9] X.H. Chen and K.L. Lee, A novel adaptive traffic load shedding scheme for CDMA cellular
mobile systems, in Proc. IEEE ICCS, vol. 2, pp. 566-570, 1994

[10] X.H. Chen, Adaptive traffic-load shedding and its capacity gain in CDMA cellular mobile
systems, IEE Proceedings-Communications, 142(3):186-192, 1995

[11] J.S. Kaufman, Blocking in a shared resource environment, IEEE Transactions on Commu-
nications, vol. 29, pp. 1474-1481, 1981

[12] A. He, Performance comparison of load balancing methods in multiple carrier CDMA sys-
tems, in Proc. IEEE PIMRC, pp. 113-118, 2000

[13] S.W. Kim, D.G. Jeong, W.S. Jeon, and C.-H. Choi, Forward link performance of combined
soft and hard handoff in multimedia CDMA systems, IEICE Transactions on Communica-
tions, vol. E85-B, no.7, pp. 1276-1282, July 2002

[14] W. Choi and J.Y. Kim, Forward-link capacity of a DS-CDMA system with mixed multirate
sources, IEEE Transactions on Vehicular Technology, 50(3):737-749, 2001

[15] D. Kim and D.G. Jeong, Capacity unbalance between uplink and downlink in spectrally
overlaid narrow-band and wide-band CDMA mobile systems, IEEE Transactions on Vehic-
ular Technology, 49(4):1086-1093, 2000

[16] W.S. Jeon and D.G. Jeong, Call admission control for mobile multimedia communications
with traffic asymmetry between uplink and downlink, IEEE Transactions on Vehicular Tech-
nology, 50(1):59-66, 2001



240 K.-C. Chu, C.-S. Wang, W.-W. Jiang, N.-C. Hsieh

[17] W.S. Jeon and D.G. Jeong, Call admission control for CDMA mobile communications
systems supporting multimedia services, IEEE Transactions on Wireless Communications,
1(4):649-659, 2002

[18] K. Kim and Y. Han, A call admission control scheme for multi-rate traffic based on total
received power, IEICE Transactions on Communications, vol. E84-B, no. 3, pp. 457-463,
March 2001