This action might not be possible to undo. Are you sure you want to continue?

# UMTS Capacity and Throughput Maximization

**for Different Spreading Factors
**

Robert Akl and Son Nguyen

Dept. of Computer Science and Engineering, University of North Texas

Denton, Texas, 76207

Email: {rakl, stn}@cse.unt.edu

Abstract—An analytical model for calculating capacity

in multi-cell UMTS networks is presented. Capacity is

maximized for different spreading factors and for perfect

and imperfect power control. We also design and implement

a local call admission control (CAC) algorithm which allows

for the simulation of network throughput for different

spreading factors and various mobility scenarios. The design

of the CAC algorithm uses global information; it incorpo-

rates the call arrival rates and the user mobilities across

the network and guarantees the users’ quality of service

as well as pre-speciﬁed blocking probabilities. On the other

hand, its implementation in each cell uses local information;

it only requires the number of calls currently active in that

cell. The capacity and network throughput were determined

for signal-to-interference threshold from 5 dB to 10 dB and

spreading factor values of 256, 64, 16, and 4.

Index Terms—WCDMA, Call Admission Control, Mobil-

ity, Network Throughput, Optimization.

I. INTRODUCTION

3G cellular systems are identiﬁed as International

Mobile Telecommunications-2000 under International

Telecommunication Union and as Universal Mobile

Telecommunications Systems (UMTS) by European

Telecommunications Standards Institute. Besides voice

capability in 2G, the new 3G systems are required to

have additional support on a variety of data-rate services

using multiple access techniques. Code Division Multiple

Access (CDMA) is the fastest-growing digital wireless

technology since its ﬁrst commercialization in 1994. The

major markets for CDMA are North America, Latin

America, and Asia (particularly Japan and Korea). In total,

CDMA has been adopted by more than 100 operators

across 76 countries around the globe [1].

Since the ﬁrst comparisons of multiple access schemes

for UMTS [2], which found that Wideband CDMA

(WCDMA) was well suited for supporting variable bit

rate services, several research on WCDMA capacity has

been considered. In [3], the authors present a method

to calculate the WCDMA reverse link Erlang capacity

based on the Lost Call Held (LCH) model as described in

[4]. This algorithm calculates the occupancy distribution

and capacity of UMTS/WCDMA systems based on a

Based on Capacity Allocation in Multi-cell UMTS Networks for

Different Spreading Factors with Perfect and Imperfect Power Control,

by R. Akl and S. Nguyen, which appeared in the Proceedings of IEEE

CCNC 2006: Consumer Communications and Networking Conference,

vol. 2, pp. 928-932, January 2006.

system outage condition. The authors derive a closed form

expression of Erlang capacity for a single type of trafﬁc

loading and compare analytical results with simulations

results.

The same LCH model was also used in [5] to calculate

the forward link capacity of UMTS/WCDMA systems

based on the system outage condition. In the forward

link, because many users share the base station (BS)

transmission power, the capacity is calculated at the BS.

The transmission power from the BS is provided to each

user based on each user’s relative need. The access in

the calculation of forward link capacity is one-to-many

rather than many-to-one as in the reverse link. The au-

thors provide capacity calculation results and performance

evaluation through simulation.

An alternate approach, where mobile stations (MSs)

are synchronized on the uplink, i.e., signals transmitted

from different MSs are time aligned at the BS, has

been considered. Synchronous WCDMA looks at time

synchronization for signal transmission between the BS

and MS to improve network capacity. The performance

of an uplink-synchronous WCDMA is analyzed in [6].

Scrambling codes are unique for each cell. MSs in the

same cell share the same scrambling code, while different

orthogonal channelization codes are derived from the set

of Walsh codes. In [6], the potential capacity gain is

about 35.8% in a multi-cell scenario with inﬁnite number

of channelization codes per cell and no soft handoff

capability between MSs and BSs. However, the capacity

gain in a more realistic scenario is reduced to 9.6%

where soft handoff is enabled. The goal of this uplink-

synchronous method in WCDMA is to reduce intra-cell

interference. But the implementation is fairly complex

while the potential capacity gain is not very high.

Our contributions are two-fold. First, we calculate

the maximum reverse link capacity in UMTS/WCDMA

systems for both perfect and imperfect power control

with a given set of quality of service requirements and

for different spreading factors. Second, we also design,

analyze, and simulate a local CAC algorithm for UMTS

networks by formulating an optimization problem that

maximizes the network throughput for different spreading

factors using signal-to-interference constraints as lower

bounds. The solution to this problem is the maximum

number of calls that can be admitted in each cell. The

design is optimized for the entire network, and the im-

40 JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006

© 2006 ACADEMY PUBLISHER

plementation is simple and considers only a single cell

for admitting a call. Numerical results are presented for

signal-to-interference thresholds from 5 dB to 10 dB and

spreading factor values of 256, 64, 16, and 4.

The remainder of this paper is organized as follows.

The user and interference models are presented in Section

II. In Section III, we analyze capacity for perfect and

imperfect power control. In Section IV, we describe our

call admission control algorithm. Network throughput is

determined in Section V. Spreading factors are discussed

in Section VI. Numerical results are presented in Section

VII, and ﬁnally Section VIII concludes the paper.

II. USER AND INTERFERENCE MODEL

This study assumes that each user is always com-

municating and is power controlled by the BS that has

the highest received power at the user. Let r

i

(x, y) and

r

j

(x, y) be the distance from a user to BS i and BS j,

respectively. This user is power controlled by BS j in the

cell or region C

j

with area A

j

, which BS j services. This

study assumes that both large scale path loss and shadow

fading are compensated by the perfect power control

mechanism. Let I

ji,g

be the average inter-cell interference

that all users n

j,g

using services g with activity factor v

g

and received signal S

g

at BS j impose on BS i. Modifying

the average inter-cell interference given by [7], it becomes

I

(g)

ji

= S

g

v

g

n

j,g

e

(βσ

s

)

2

A

j

_ _

C

j

r

m

j

(x, y)

r

m

i

(x, y)

w(x, y) dA(x, y),

(1)

where β = ln(10)/10, σ

s

is the standard deviation of

the attenuation for the shadow fading, m is the path loss

exponent, and w(x, y) is the user distribution density at

(x, y). Let κ

ji,g

be the per-user (with service g) relative

inter-cell interference factor from cell j to BS i,

κ

ji,g

=

e

(βσ

s

)

2

A

j

_ _

C

j

r

m

j

(x, y)

r

m

i

(x, y)

w(x, y) dA(x, y). (2)

The inter-cell interference density I

inter

ji

from cell j to

BS i from all services G becomes

I

inter

ji

=

1

W

G

g=1

I

(g)

ji

, (3)

where W is the bandwidth of the system. Eq. (3) can be

rewritten as

I

inter

ji

=

1

W

G

g=1

S

g

v

g

n

j,g

κ

ji,g

. (4)

Thus, the total inter-cell interference density I

inter

i

from

all other cells to BS i is

I

inter

i

=

1

W

M

j=1,j=i

G

g=1

S

g

v

g

n

j,g

κ

ji,g

, (5)

where M is the total number of cells in the network.

If the user distribution density can be approximated,

then, κ

ji,g

needs to be calculated only once. The user

distribution is modeled with a 2-dimensional Gaussian

function as follows [8]

w(x, y) =

η

2πσ

1

σ

2

e

−

1

2

(

x−µ

1

σ

1

)

2

e

−

1

2

(

y−µ

2

σ

2

)

2

, (6)

where η is a user density normalizing parameter.

By specifying the means µ

1

and µ

2

and the standard

deviations σ

1

and σ

2

of the distribution for every cell,

an approximation can be found for a wide range of user

distributions ranging from uniform to hot-spot clusters.

These results are compared with simulations to determine

the value of η experimentally.

III. UMTS CAPACITY

A. Capacity with Perfect Power Control

In WCDMA, with perfect power control (PPC) between

BSs and MSs, the energy per bit to total interference

density at BS i for a service g is given by [9]

_

E

b

I

0

_

i,g

=

S

g

R

g

N

0

+ I

inter

i

+ I

own

i

− S

g

v

g

, (7)

where N

0

is the thermal noise density, and R

g

is the

bit rate for service g. I

inter

i

was calculated in section II.

I

own

i

is the total intra-cell interference density caused by

all users in cell i. Thus I

own

i

is given by

I

own

i

=

1

W

G

g=1

S

g

v

g

n

i,g

. (8)

Let τ

g

be the minimum signal-to-noise ratio, which

must received at a BS to decode the signal of a user with

service g, and S

∗

g

be the maximum signal power, which

the user can transmit. Substituting (5) and (8) into (7), we

have for every cell i in the UMTS network, the number

of users n

i,g

in BS i for a given service g needs to meet

the following inequality constraints

τ

g

≤

S

∗

g

R

g

N

0

+

S

∗

g

W

_

G

g=1

n

i,g

v

g

+

M

j=1,j=i

G

g=1

n

j,g

v

g

κ

ji,g

−v

g

_

,

for i=1,...,M. (9)

After rearranging terms, (9) can be rewritten as

G

g=1

n

i,g

v

g

+

M

j=1,j=i

G

g=1

n

j,g

v

g

κ

ji,g

−v

g

≤ c

(g)

eff

,

for i=1,...,M, (10)

where

c

(g)

eff

=

W

R

g

_

1

τ

g

−

R

g

S

∗

g

/N

0

_

. (11)

The maximized capacity in a UMTS network is de-

ﬁned as the maximum number of simultaneous users

(n

1,g

, n

2,g

, ..., n

M,g

) for all services g = 1, ..., G that

satisfy (10).

JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006 41

© 2006 ACADEMY PUBLISHER

B. Capacity with Imperfect Power Control

The calculation of UMTS network capacity, which

was formulated in section III-A, assumes perfect power

control between the BSs and MSs. However, transmitted

signals between BSs and MSs are subject to multi-path

propagation conditions, which make the received

_

E

b

I

o

_

i,g

signals vary according to a log-normal distribution with

a standard deviation on the order of 1.5 to 2.5 dB [4].

Thus, in the imperfect power control (IPC) case, the

constant value of (E

b

)

i,g

in each cell i for every user

with service g needs to be replaced by the variable

(E

b

)

i,g

=

i,g

(E

b

)

o,g

, which is log-normally distributed.

We deﬁne

x

i,g

= 10log

10

_

i,g

(E

b

)

o,g

I

0

_

, (12)

to be a normally distributed random variable with mean

m

c

and standard deviation σ

c

.

According to [4], by evaluating the nth moment of

i,g

using the fact that x

i,g

is Gaussian with mean m

c

and

standard deviation σ

c

, then taking the expected value, we

have

E

_

(E

b

)

o,g

I

0

i,g

_

=

(E

b

)

i,g

I

0

e

(βσ

c

)

2

2

. (13)

As a result of (13), c

(g)

eff IPC

becomes c

(g)

eff

/ e

(βσ

c

)

2

2

.

IV. UMTS CALL ADMISSION CONTROL

A. Feasible States

Recall from section III-A that the number of calls in

every cell must satisfy (10). A set of calls n satisfying

(10) is said to be in feasible call conﬁguration or a feasible

state, which meet the

E

b

I

0

constraint.

Denote by Ω the set of feasible states. Deﬁne the set

of blocking states for service g in cell i as

B

i,g

=

_

_

_

n ∈ Ω :

_

_

n

1,1

... n

1,G

... ... ...

n

M,1

... n

M,G

_

_

∈ Ω

_

_

_

. (14)

If a new connection or a handoff connection with the

service g arrives to cell i, it is blocked when the current

state of the network, n, is in B

i,g

.

B. Mobility Model

There are several mobility models that have been

discussed in the literature [10]–[12]. These models have

ranged from general dwell times for calls to ones that have

hyper-exponential and sub-exponential distributions. For

the CAC problem that we are investigating here, however,

such assumptions makes the problem mathematically in-

tractable. The mobility model that we use is presented in

[13] where a call stops occupying a cell either because

user mobility has forced the call to be handed off to

another cell, or because the call is completed.

The call arrival process with service g to cell i is

assumed to be a Poisson process with rate λ

i,g

inde-

pendent of other call arrival processes. The call dwell

time is a random variable with exponential distribution

having mean 1/µ, and it is independent of earlier arrival

times, call durations and elapsed times of other users.

At the end of a dwell time a call may stay in the same

cell, attempt a handoff to an adjacent cell, or leave the

network. Deﬁne q

ii,g

as the probability that a call with

service g in progress in cell i remains in cell i after

completing its dwell time. In this case, a new dwell time

that is independent of the previous dwell time begins

immediately. Let q

ij,g

be the probability that a call with

service g in progress in cell i after completing its dwell

time goes to cell j. If cells i and j are not adjacent,

then q

ij,g

= 0. We denote by q

i,g

the probability that a

call with service g in progress in cell i departs from the

network.

This mobility model is attractive because we can easily

deﬁne different mobility scenarios by varying the values

of these probability parameters [13]. For example, if q

i,g

is constant for all i and g, then the average dwell time of

a call of the same service in the network will be constant

regardless of where the call originates and what the values

of q

ii,g

and q

ij,g

are. Thus, by varying q

ii,g

’s and q

ij,g

’s

for a service g, we can obtain low and high mobility

scenarios and compare the effect of mobility on network

attributes (e.g., throughput).

We assume that the occupancy of the cells evolves

according to a birth-death process, where the total arrival

rate or offered trafﬁc for service g to cell i is ρ

i,g

, and

the departure rate from cell i when the network is in

state n is n

i,g

µ

i,g

= n

i,g

µ(1 −q

ii,g

). Let ρ be the matrix

of offered trafﬁc of service g to the cells, µ the matrix

of departure rates, and let p(ρ, µ, n) be the stationary

probability that the network is in state n. The new call

blocking probability for service g in cell i, B

i,g

, is given

by

B

i,g

=

n∈B

i,g

p(ρ, µ, n). (15)

This is also the blocking probability of handoff calls due

to the fact that handoff calls and new calls are treated in

the same way by the network.

Let A

i

be the set of cells adjacent to cell i. Let ν

ji,g

be

the handoff rate out of cell j offered to cell i for service g.

ν

ji,g

is the sum of the proportion of new calls of service

g accepted in cell j that go to cell i and the proportion of

handoff calls with service g accepted from cells adjacent

to cell j that go to cell i. Thus

ν

ji,g

= λ

j,g

(1 −B

j,g

)q

ji,g

+ (1 −B

j,g

)q

ji,g

x∈A

j

ν

xj,g

.

(16)

Equation (16) can be rewritten as

ν

ji,g

= ν(B

j,g

, ρ

j,g

, q

ji,g

) = (1 −B

j,g

)q

ji,g

ρ

j,g

, (17)

where ρ

j,g

, the total offered trafﬁc to cell j for service g,

is given by

ρ

j,g

= ρ(v, λ

j,g

, A

j

) = λ

j,g

+

x∈A

j

ν

xj,g

, (18)

42 JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006

© 2006 ACADEMY PUBLISHER

and where v denotes the matrix whose components are

the handoff rates ν

ij

for i, j = 1, ...M.

The total offered trafﬁc can be obtained from a ﬁxed

point model [14], which describes the offered trafﬁc as a

function of the handoff rates and new call arrival rates, the

handoff rates as a function of the blocking probabilities

and the offered trafﬁc, and the blocking probabilities as

a function of the offered trafﬁc. For a given set of arrival

rates, we use an iterative method to solve the ﬁxed point

equations. We deﬁne an initial value for the handoff rates.

We calculate the offered trafﬁc by adding the given values

of the arrival rates to the handoff rates. The blocking

probabilities are now calculated using the offered trafﬁc.

We then calculate the new values of the handoff rates

and repeat. This approach has been extensively utilized in

the literature to obtain solutions of ﬁxed point problems

[15]–[20]. The questions of existence and uniqueness of

the solution and whether the iterative approach in fact

converges to the solution (if a unique solution exists)

are generally difﬁcult to answer due to the complexity

of the equations involved. Kelly has shown that for ﬁxed

alternate routing the solution to the ﬁxed point problem

is in fact not unique [21]; in all the numerical examples

we solved, the iterative approach converged to a unique

solution.

C. Admissible States

A CAC algorithm can be constructed as follows. A

call arriving to cell i with service g is accepted if and

only if the new state is a feasible state. Clearly this CAC

algorithm requires global state, i.e., the number of calls

in progress in all the cells of the network. Furthermore, to

compute the blocking probabilities, the probability of each

state in the feasible region needs to be calculated. Since

the cardinality of Ω is O(c

eff

MG

), the calculation of

the blocking probabilities has a computational complexity

that is exponential in the number of cells combined with

number of available services.

In order to simplify the CAC algorithm, we consider

only those CAC algorithms which utilize local state, i.e.,

the number of calls in progress in the current cell. To this

end we deﬁne a state n to be admissible if

n

i,g

≤ N

i,g

for i = 1, ..., M and g = 1, ..., G, (19)

where N

i,g

is a parameter which denotes the maximum

number of calls with service g allowed to be admitted

in cell i. Clearly the set of admissible states denoted Ω

**is a subset of the set of feasible states Ω. The blocking
**

probability for cell i with service g is then given by

B

i,g

= B(A

i,g

, N

i,g

) =

A

N

i,g

i,g

/N

i,g

!

N

i,g

k=0

A

k

i,g

/k!

, (20)

where A

i

= ρ

i,g

/µ

i,g

= ρ

i,g

/µ(1 − q

ii,g

) is the Erlang

trafﬁc in cell i with service g. We note that the complexity

to calculate the blocking probabilities in (20) is O(MG),

and the bit error rate requirement is guaranteed since Ω

⊂

Ω.

Once the maximum number of calls with different

service that are allowed to be admitted in each cell, N,

is calculated (this is done ofﬂine and described in the

next section), the CAC algorithm for cell i for service

g will simply compare the number of calls with service

g currently active in cell i to N

i,g

in order to accept

or reject a new arriving call. Thus our CAC algorithm

is implemented with a computational complexity that is

O(1).

V. NETWORK THROUGHPUT

The throughput of cell i consists of two components:

the new calls that are accepted in cell i minus the forced

termination due to handoff failure of the handoff calls into

cell i for all services g. Hence the total throughput, T, of

the network is

T(B, ρ, λ) =

M

i=1

G

g=1

{λ

i,g

−B

i,g

ρ

i,g

} , (21)

where B is the vector of blocking probabilities and λ is

the matrix of call arrival rates.

A. Calculation of N

We formulate a constrained optimization problem in

order to maximize the throughput subject to upper bounds

on the blocking probabilities and a lower bound on the

signal-to-interference constraints in (10). The goal is to

optimize the utilization of network resources and provide

consistent GoS while at the same time maintaining the

QoS, β

g

, for all the users for different services g. In this

optimization problem the arrival rates are given and the

maximum number of calls that can be admitted in all the

cells are the independent variables. This is given in the

following

max

N

T(B, ρ, λ),

subject to B(A

i,g

, N

i,g

) ≤ β

g

,

G

g=1

N

i,g

v

g

+

M

j=1,j=i

G

g=1

N

j,g

v

g

κ

ji,g

−v

g

≤ c

(g)

eff

,

for i = 1, ..., M. (22)

The optimization problem in (22) is solved ofﬂine to

obtain the values of N.

B. Maximization of Throughput

A second optimization problem can be formulated in

which the arrival rates and the maximum number of calls

that can be admitted in all the cells are the independent

JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006 43

© 2006 ACADEMY PUBLISHER

Fig. 1. Generation of OVSF codes for different Spreading Factors.

variables and the objective function is the throughput.

This is given in the following

max

λ, N

T(B, ρ, λ),

subject to B(A

i,g

, N

i,g

) ≤ β

g

,

G

g=1

N

i,g

v

g

+

M

j=1,j=i

G

g=1

N

j,g

v

g

κ

ji,g

−v

g

≤ c

(g)

eff

,

for i = 1, ..., M. (23)

The optimized objective function of (23) provides an

upper bound on the total throughput that the network can

carry. This is the network capacity for the given GoS and

QoS.

VI. SPREADING FACTOR

Communication from a single source is separated by

channelization codes, i.e., the dedicated physical channel

in the uplink and the downlink connections within one

sector from one MS. The Orthogonal Variable Spreading

Factor (OVSF) codes, which were originally introduced

in [22], were used to be channelization codes for UMTS.

The use of OVSF codes allows the orthogonality and

spreading factor (SF) to be changed between different

spreading codes of different lengths. Fig. 1 depicts the

generation of different OVSF codes for different SF

values.

The data signal after spreading is then scrambled with

a scrambling codes to separate MSs and BSs from each

other. Scrambling is used on top of spreading, thus it only

makes the signals from different sources distinguishable

from each other. Fig. 2 depicts the relationship between

the spreading and scrambling process. Table I describes

the different functionality of the channelization and the

scrambling codes.

The typical required data rate or Dedicated Trafﬁc

Channel (DTCH) for a voice user is 12.2 Kbps. However,

the Dedicated Physical Data Channel (DPDCH), which

is the actual transmitted data rate, is dramatically in-

creased due to the incorporated Dedicated Control Chan-

nel (DCCH) information, and the processes of Channel

Fig. 2. Relationship between spreading and scrambling.

TABLE I

FUNCTIONALITY OF THE CHANNELIZATION AND SCRAMBLING

CODES.

Channelization code Scrambling code

Usage Uplink: Separation of

physical data (DPDCH)

and control channels

(DPCCH) from same MS

Downlink: Separation of

downlink connections to

different MSs within one

cell.

Uplink: Separation of

MSs

Downlink: Separation of

sectors (cells)

Length Uplink: 4-256 chips same

as SF

Downlink 4-512 chips

same as SF

Uplink: 10 ms = 38400

chips

Downlink: 10 ms =

38400 chips

Number of codes Number of codes under

one scrambling code =

spreading factor

Uplink: Several millions

Downlink: 512

Code family Orthogonal Variable

Spreading Factor

Long 10 ms code: Gold

Code

Short code: Extended

S(2) code family

Spreading Yes, increases transmis-

sion bandwidth

No, does not affect trans-

mission bandwidth

Coding, Rate Matching, and Radio Frame Alignment. Fig.

3 depicts the process of creating the actual transmitted

signal for a voice user. Fig. 4 shows the DPDCH data

rate requirement for 64 Kbps data user. Table II shows

the approximation of the maximum user data rate with

1

2

rate coding for different values of DPDCH.

VII. NUMERICAL RESULTS

The results shown are for a twenty-seven cell network

topology used in [23], [24]. The COST-231 propagation

model with a carrier frequency of 1800 MHz, average

base station height of 30 meters and average mobile height

Fig. 3. 12.2 Kbps Uplink Reference channel.

44 JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006

© 2006 ACADEMY PUBLISHER

Fig. 4. 64 Kbps Uplink Reference channel.

TABLE II

UPLINK DPDCH DATA RATES.

DPDCH Spreading Factor DPDCH bit rate User data rate

1

2

rate coding

256 15 Kbps 7.5 Kbps

128 30 Kbps 15 Kbps

64 60 Kbps 30 Kbps

32 120 Kbps 60 Kbps

16 240 Kbps 120 Kbps

8 480 Kbps 240 Kbps

4 960 Kbps 480 Kbps

4, with 6 parallel codes 5740 Kbps 2.8 Mbps

of 1.5 meters, is used to determine the coverage region.

The path loss coefﬁcient m is 4. The shadow fading

standard deviation σ

s

is 6 dB. The processing gain

W

R

g

is

6.02 dB, 12.04 dB, 18.06 dB, and 24.08 dB for Spreading

Factor equal to 4, 16, 64, and 256, respectively. The

activity factor, v, is 0.375. Fig. 5 shows the 2-D Gaussian

approximation of users uniformly distributed in the cells

with σ

1

= σ

2

= 12000.

The UMTS network with 27 omnidirectional antenna

cells (1 sector per cell) was analyzed for evaluation of

capacity using user modeling with the 2-D Gaussian

function and traditional methods of modeling uniform

user distribution. The network with different values for

E

b

I

0

was analyzed for different SF values of 4, 16, 64, and

256.

Fig. 5. 2-D Gaussian approximation of users uniformly distributed in

the cells. σ

1

= σ

2

= 12000, µ

1

= µ

2

= 0.

Fig. 6. Average number of slot per sector for perfect and imperfect

power control analysis with a Spreading Factor of 256.

Fig. 7. Average number of slot per sector for perfect and imperfect

power control analysis with a Spreading Factor of 64.

A. Capacity Allocation with SF of 256

First, we set SF to 256, which is used to carry data for

the control channel. Fig. 6 shows the maximized average

number of slots per sector for the 27 cells UMTS network

as the

E

b

I

0

is increased from 5 dB to 10 dB and the standard

deviation of imperfect power control is increased from

0 to 2.5 dB. Because of IPC, to get the same average

number of slots per sector as PPC, we have to decrease the

SIR threshold by 0.5 dB to 1.5 dB. Fig. 6 also shows that

the traditional uniform user distribution modeling matches

well with the 2-D Gaussian model.

B. Capacity Allocation with SF of 64

As a result of lowering the SF to 64, the number of slots

per sector decreases by almost a factor of 4 compared to

SF equal 256 (from 60.58 to 15.56 slots when

E

b

I

o

= 7.5

dB in PPC) as shown in Fig. 7.

JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006 45

© 2006 ACADEMY PUBLISHER

Fig. 8. Average number of slot per sector for perfect and imperfect

power control analysis with a Spreading Factor of 16.

Fig. 9. Average number of slot per sector for perfect and imperfect

power control analysis with a Spreading Factor of 4.

C. Capacity Allocation with SF of 16

As a result of lowering the SF to 16, the number of slots

per sector decreases by almost a factor of 4 compared to

SF equal 64 (from 15.56 to 4.30 slots when

E

b

I

o

= 7.5 dB

in PPC) as shown in Fig. 8.

D. Capacity Allocation with SF of 4

Next, we set SF to 4, which is used for 256 kbps

data communication between BSs and MSs. As a result

of lowering the SF to 4, the number of slots per sector

decreases signiﬁcantly to 1.49 while keeping

E

b

I

o

= 7.5 dB

in PPC as shown in Fig. 9.

The following set of results are for the calculation

of throughput. Three mobility scenarios: no mobility,

low mobility, and high mobility of users are considered.

We assume that the mobility characteristics for a given

service g stays the same throughout different cells in the

network. The following parameters are used for the no

mobility case: q

ij,g

= 0, q

ii,g

= 0.3 and q

i,g

= 0.7 for

all cells i and j. Tables III and IV show respectively the

TABLE III

THE LOW MOBILITY CHARACTERISTICS AND PARAMETERS.

A

i

q

ij,g

q

ii,g

q

i,g

3 0.020 0.240 0.700

4 0.015 0.240 0.700

5 0.012 0.240 0.700

6 0.010 0.240 0.700

TABLE IV

THE HIGH MOBILITY CHARACTERISTICS AND PARAMETERS.

A

i

q

ij,g

q

ii,g

q

i,g

3 0.1 0 0.700

4 0.075 0 0.700

5 0.060 0 0.700

6 0.050 0 0.700

• A

i

is the number of cells, which are adjacent to cell i.

• q

ii,g

is the probability that a call with service g in progress in cell i

remains in cell i after completing its dwell time.

• q

ij,g

is the probability that a call with service g in progress in cell i after

completing its dwell time goes to cell j.

• q

i,g

is the probability that a call with service g in progress in cell i departs

from the network.

mobility characteristics and parameters for the low and

high mobility cases. In all three mobility scenarios, the

probability that a call leaves the network after completing

its dwell time is 0.7. Thus, regardless of where the call

originates and mobility scenario used, the average dwell

time of a call in the network is constant. In the numerical

results below, for each SF value, we analyze the average

throughput per cell by dividing the results from (23) by

the total number of cells in the network and multiplying

by the maximum data rate in Table II.

E. Throughput Optimization with SF of 256

First, we set SF equal to 256, which is used to carry

data for the control channel. Table V shows the optimized

values of N for each cell for all three mobility models with

perfect power control and 2% blocking probability. Fig.

10 shows the optimized throughput per cell for a blocking

probability from 1% to 10%. The results for the average

throughput for no mobility and high mobility cases are

almost identical while the throughput for low mobility

is higher for each blocking probability. The low mobility

case has an equalizing effect on trafﬁc resulting in slightly

higher throughput.

F. Throughput Optimization with SF of 64

Next, we set SF equal to 64, which is used for voice

communication as shown in Fig. 3. As a result of low-

ering the SF to 64, the number of possible concurrent

connections within one cell is also decreased. Because

the throughput is calculated based on the number of

simultaneous connections between MSs and BSs, the

lower trunking efﬁciency leads to lower throughput as

shown in Fig. 11. Table VI shows the optimized values

of N for each cell for all three mobility cases and SF

equal to 64.

46 JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006

© 2006 ACADEMY PUBLISHER

Fig. 10. Average throughput in each cell for SF = 256.

TABLE V

CALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF =

256 AND BLOCKING PROBABILITY = 0.02.

No Mobility Low Mobility High Mobility

Cell ID N

i

N

i

N

i

Cell

1

52.86 52.86 52.86

Cell

2

53.95 53.95 53.95

Cell

3

51.84 51.84 51.84

Cell

4

51.84 51.84 51.84

Cell

5

53.95 53.95 53.95

Cell

6

51.85 51.85 51.85

Cell

7

51.85 51.85 51.85

Cell

8

53.00 53.00 53.00

Cell

9

50.73 50.73 50.73

Cell

10

62.74 62.74 62.74

Cell

11

63.29 63.29 63.29

Cell

12

62.73 62.73 62.73

Cell

13

50.73 50.73 50.73

Cell

14

53.01 53.01 53.01

Cell

15

50.73 50.73 50.73

Cell

16

62.71 62.71 62.71

Cell

17

63.27 63.27 63.27

Cell

18

62.71 62.71 62.71

Cell

19

50.74 50.74 50.74

Cell

20

73.40 73.40 73.40

Cell

21

71.84 71.84 71.84

Cell

22

71.86 71.86 71.86

Cell

23

73.43 73.43 73.43

Cell

24

73.43 73.43 73.43

Cell

25

71.83 71.83 71.83

Cell

26

71.82 71.82 71.82

Cell

27

73.40 73.40 73.40

G. Throughput Optimization with SF of 16

Next, we set SF equal to 16, which is used for 64 Kbps

data communication as shown in Fig. 4. As a result of

lowering the SF to 16, the average number of slots within

one cell decreases to 4.30. The resulting throughput, as

shown in Fig. 12, is much lower compared to the case

with SF equal to 64 or 256. Table VII shows the optimized

values of N for each cell for all three mobility cases with

SF equal to 16.

H. Throughput Optimization with SF of 4

Next, we set SF equal to 4, which is normally used for

256 Kbps data communication between BSs and MSs.

As a result of lowering the SF to 4, the average slots per

Fig. 11. Average throughput in each cell for SF = 64.

TABLE VI

CALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF =

64 AND BLOCKING PROBABILITY = 0.02.

No Mobility Low Mobility High Mobility

Cell ID N

i

N

i

N

i

Cell

1

13.58 13.58 13.58

Cell

2

13.86 13.86 13.86

Cell

3

13.32 13.32 13.32

Cell

4

13.32 13.32 13.32

Cell

5

13.86 13.86 13.86

Cell

6

13.32 13.32 13.32

Cell

7

13.32 13.32 13.32

Cell

8

13.61 13.61 13.61

Cell

9

13.03 13.03 13.03

Cell

10

16.11 16.11 16.11

Cell

11

16.26 16.26 16.26

Cell

12

16.11 16.11 16.11

Cell

13

13.03 13.03 13.03

Cell

14

13.62 13.62 13.62

Cell

15

13.03 13.03 13.03

Cell

16

16.11 16.11 16.11

Cell

17

16.25 16.25 16.25

Cell

18

16.11 16.11 16.11

Cell

19

13.03 13.03 13.03

Cell

20

18.85 18.85 18.85

Cell

21

18.45 18.45 18.45

Cell

22

18.46 18.46 18.46

Cell

23

18.86 18.86 18.86

Cell

24

18.86 18.86 18.86

Cell

25

18.45 18.45 18.45

Cell

26

18.45 18.45 18.45

Cell

27

18.85 18.85 18.85

sector decreases signiﬁcantly to 1.49 with perfect power

control and

E

b

I

0

= 7.5 dB as shown in Fig. 9. Table VII

shows the optimized values of N for each cell for all

three mobility models with SF equal to 4. The average

throughput for all three mobility cases are almost identical

as shown in Fig. 13.

VIII. CONCLUSIONS

An analytical model has been presented for calculating

capacity in multi-cell UMTS networks. Numerical results

show that the SIR threshold for the received signals is

decreased by 0.5 to 1.5 dB due to the imperfect power

control. As expected, we can have many low rate voice

users or fewer data users as the data rate increases. The

JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006 47

© 2006 ACADEMY PUBLISHER

Fig. 12. Average throughput in each cell for SF = 16.

TABLE VII

CALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF =

16 AND BLOCKING PROBABILITY = 0.02.

No Mobility Low Mobility High Mobility

Cell ID N

i

N

i

N

i

Cell

1

3.75 3.75 3.75

Cell

2

3.83 3.83 3.83

Cell

3

3.68 3.68 3.68

Cell

4

3.68 3.68 3.68

Cell

5

3.83 3.83 3.83

Cell

6

3.68 3.68 3.68

Cell

7

3.68 3.68 3.68

Cell

8

3.76 3.76 3.76

Cell

9

3.60 3.60 3.60

Cell

10

4.46 4.46 4.46

Cell

11

4.50 4.50 4.50

Cell

12

4.46 4.46 4.46

Cell

13

3.60 3.60 3.60

Cell

14

3.77 3.77 3.77

Cell

15

3.60 3.60 3.60

Cell

16

4.45 4.45 4.45

Cell

17

4.49 4.49 4.49

Cell

18

4.45 4.45 4.45

Cell

19

3.60 3.60 3.60

Cell

20

5.21 5.21 5.21

Cell

21

5.10 5.10 5.10

Cell

22

5.10 5.10 5.10

Cell

23

5.22 5.22 5.22

Cell

24

5.22 5.22 5.22

Cell

25

5.10 5.10 5.10

Cell

26

5.10 5.10 5.10

Cell

27

5.21 5.21 5.21

results also show that the determined parameters of the

2-dimensional Gaussian model matches well with tradi-

tional methods for modeling uniform user distribution.

An analytical model was also presented for CAC algo-

rithm for optimizing the throughput in multi-cell UMTS

networks. Numerical results show that as the spreading

factor increases, the optimized throughput is better, due

to the trunking efﬁciency for all three mobility models

(no, low, and high mobility). Our methods for maximizing

capacity and implementing the CAC algorithm are fast,

accurate, and can be implemented for large multi-cell

UMTS networks.

REFERENCES

[1] CDMA Development Group, “CDG : Worldwide : CDMA World-

Fig. 13. Average throughput in each cell for SF = 4.

TABLE VIII

CALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF =

4 AND BLOCKING PROBABILITY = 0.02.

No Mobility Low Mobility High Mobility

Cell ID N

i

N

i

N

i

Cell

1

1.24 1.16 1.32

Cell

2

1.48 1.42 1.24

Cell

3

1.30 1.32 1.28

Cell

4

1.30 1.32 1.28

Cell

5

1.48 1.42 1.23

Cell

6

1.30 1.32 1.28

Cell

7

1.30 1.32 1.28

Cell

8

0.94 1.37 1.23

Cell

9

0.93 0.94 1.27

Cell

10

1.59 1.58 1.54

Cell

11

1.53 1.53 1.55

Cell

12

1.59 1.58 1.54

Cell

13

0.93 0.94 1.27

Cell

14

0.94 1.37 1.23

Cell

15

0.93 0.93 1.27

Cell

16

1.59 1.58 1.54

Cell

17

1.53 1.53 1.55

Cell

18

1.59 1.58 1.54

Cell

19

0.93 0.93 1.27

Cell

20

1.92 1.83 1.82

Cell

21

1.79 1.80 1.76

Cell

22

1.79 1.80 1.76

Cell

23

1.92 1.83 1.82

Cell

24

1.92 1.83 1.82

Cell

25

1.79 1.80 1.76

Cell

26

1.79 1.80 1.76

Cell

27

1.92 1.83 1.82

wide,” http://www.cdg.org/worldwide/index.asp?h area=0.

[2] T. Ojanpera, J. Skold, J. Castro, L. Girard, and A. Klein, “Compar-

ison of multiple access schemes for UMTS,” IEEE Veh. Technol.

Conf., vol. 2, pp. 490–494, May 1997.

[3] Q. Zhang and O. Yue, “UMTS air interface voice/data capacity-

part 1: reverse link analysis,” IEEE Veh. Technol. Conf., vol. 4, pp.

2725 – 2729, May 2001.

[4] A. Viterbi, CDMA Principles of Spread Spectrum Communication.

Addison-Wesley, 1995.

[5] Q. Zhang, “UMTS air interface voice/data capacity-part 2: forward

link analysis,” IEEE Veh. Technol. Conf., vol. 4, pp. 2730 – 2734,

May 2001.

[6] J. Carnero, K. Pedersen, and P. Mogensen, “Capacity gain of an

uplink-synchronous WCDMA system under channelization code

constraints,” IEEE Veh. Technol. Conf., vol. 53, pp. 982 – 991,

July 2004.

[7] R. Akl, M. Hegde, M. Naraghi-Pour, and P. Min, “Multi-cell

CDMA network design,” IEEE Trans. Veh. Technol., vol. 50, no. 3,

pp. 711–722, May 2001.

48 JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006

© 2006 ACADEMY PUBLISHER

[8] S. Nguyen and R. Akl, “Approximating user distributions in

WCDMA networks using 2-D Gaussian,” Proceedings of Inter-

national Conf. on Comput., Commun., and Control Technol., July

2005.

[9] D. Staehle, K. Leibnitz, K. Heck, B. Schroder, A. Weller, and

P. Tran-Gia, “Approximating the othercell interference distribution

in inhomogenous UMTS networks,” IEEE Veh. Technol. Conf.,

vol. 4, pp. 1640–1644, May 2002.

[10] T. Tugcu and C. Ersoy, “Application of a realistic mobility model

to call admissions in DS-CDMA cellular systems,” IEEE Veh.

Technol. Conf., vol. 2, pp. 1047–1051, Spring 2001.

[11] P. Orlik and S. Rappaport, “On the handoff arrival process in

cellular communications,” Wireless Networks, vol. 7, pp. 147–157,

2001.

[12] Y. Fang, I. Chlamtac, and Y.-B. Lin, “Modeling PCS networks un-

der general call holding time and cell residence time distributions,”

IEEE/ACM Trans. on Networking, vol. 5, pp. 893–906, 1997.

[13] C. Vargas, M. Hegde, and M. Naraghi-Pour, “Implied costs for

multi-rate wireless networks,” J.Wireless Networks, vol. 10, pp.

323–337, May 2004.

[14] V. Istratescu, Fixed Point Theory : An Introduction. D. Reidel,

1981.

[15] R. Akl, M. Hegde, and P. Min, “Effects of call arrival rate

and mobility on network throughput in multi-cell CDMA,” IEEE

International Conf. on Commun., vol. 3, pp. 1763–1767, June

1999.

[16] F. Kelly, “Routing in circuit-switched network: Optimization,

shadow prices and decentralization,” Advances in Applied Prob-

ability, vol. 20, pp. 112–144, 1988.

[17] D. Mitra, J. Morrison, and K. Ramakrishnan, “ATM network

design and optimization: a multirate loss network framework,”

IEEE/ACM Trans. on Networking, vol. 4, no. 4, pp. 531–543,

August 1996.

[18] C. Vargas, M. Hegde, and M. Naraghi-Pour, “Blocking effects of

mobility and reservations in wireless networks,” IEEE Interna-

tional Conf. on Commun., vol. 3, pp. 1612–1616, June 1998.

[19] ——, “Implied costs in wireless networks,” IEEE Veh. Technol.

Conf., vol. 2, pp. 904–908, May 1998.

[20] C. Vargas-Rosales, “Communication network design and evalu-

ation using shadow prices,” Ph.D. dissertation, Louisiana State

University, 1996.

[21] F. Kelly, “Blocking probabilities in large circuit switched net-

works,” Advances in Applied Probability, vol. 18, pp. 473–505,

1986.

[22] A. F., S. M., and O. K., “Tree-structured generation of orthogonal

spreading codes with different lengths for forward link of DS-

CDMA mobile radio,” Electr. Lett., vol. 33, pp. 27–28, January

1997.

[23] R. Akl and A. Parvez, “Impact of interference model on capacity in

CDMA cellular networks,” Proceedings of SCI 04: Communication

and Network Systems, Technologies and Applications, vol. 3, pp.

404–408, July 2004.

[24] R. Akl, M. Hegde, and M. Naraghi-Pour, “Mobility-based CAC

algorithm for arbitrary trafﬁc distribution in CDMA cellular sys-

tems,” IEEE Trans. Veh. Technol., vol. 54, pp. 639–651, March

2005.

Robert Akl received the B.S. degree in computer science from

Washington University in St. Louis, in 1994, and the B.S., M.S.

and D.Sc. degrees in electrical engineering in 1994, 1996, and

2000, respectively. He also received the Dual Degree Engineer-

ing Outstanding Senior Award from Washington University in

1993. He is a senior member of IEEE.

Dr. Akl is currently an Assistant Professor at the University of

North Texas, Department of Computer Science and Engineering.

In 2002, he was an Assistant Professor at the University of New

Orleans, Department of Electrical and Computer Engineering.

From October 2000 to December 2001, he was a senior systems

engineer at Comspace Corporation, Coppell, TX. His research

interests include wireless communication and network design

and optimization.

Son Nguyen received the B.S. degree in computer science

and the M.S. degree in computer science from The University

of North Texas in 2001 and 2005, respectively. His research

interests include 3G wireless network design and optimization.

JOURNAL OF NETWORKS, VOL. 1, NO. 3, JULY 2006 49

© 2006 ACADEMY PUBLISHER

This study assumes that both large scale path loss and shadow fading are compensated by the perfect power control mechanism. 16. κji. y) Let τg be the minimum signal-to-noise ratio.M . Let Iji.j=i g=1 where M is the total number of cells in the network. κji.. After rearranging terms. respectively. Substituting (5) and (8) into (7). the total inter-cell interference density all other cells to BS i is inter Ii 1 = W M G W Rg 1 Rg . y) dA(x. an approximation can be found for a wide range of user distributions ranging from uniform to hot-spot clusters.g κji.. NO. we describe our call admission control algorithm. If the user distribution density can be approximated. Let κji. In Section III. − ∗ τg Sg /N0 (11) j=1..g ) for all services g = 1.g −vg ni. (2) m ri (x. 3.. and ﬁnally Section VIII concludes the paper. (9) can be rewritten as G M G (9) where W is the bandwidth of the system. it becomes (g) Iji distribution is modeled with a 2-dimensional Gaussian function as follows [8] w(x. g=1 (8) Cj m rj (x.g using services g with activity factor vg and received signal Sg at BS j impose on BS i.g vg + g=1 j=1. y). nj.M . which the user can transmit. own Ii is the total intra-cell interference density caused by own all users in cell i. The user and interference models are presented in Section II. then. ..g . .g e(βσs ) Aj 2 Cj m rj (x. Ii was calculated in section II.g − vg ≤ cef f . n2. nM.g be the per-user (with service g) relative inter-cell interference factor from cell j to BS i. y) w(x. the energy per bit to total interference density at BS i for a service g is given by [9] Eb I0 = i.g vg + g=1 j=1. Numerical results are presented in Section VII. y) m (x. 1.g in BS i for a given service g needs to meet the following inequality constraints τg ≤ ∗ Sg Rg M inter The inter-cell interference density Iji from cell j to BS i from all services G becomes inter Iji = 1 W G Iji . (10) (g) for i=1. σs is the standard deviation of the attenuation for the shadow fading. By specifying the means µ1 and µ2 and the standard deviations σ1 and σ2 of the distribution for every cell.. y) be the distance from a user to BS i and BS j. The user © 2006 ACADEMY PUBLISHER The maximized capacity in a UMTS network is deﬁned as the maximum number of simultaneous users (n1.g Sg Rg inter + I own − S v N0 + Ii g g i . . y).g be the average inter-cell interference that all users nj.. This user is power controlled by BS j in the cell or region Cj with area Aj . Let ri (x. y) dA(x.. Eq.. Numerical results are presented for signal-to-interference thresholds from 5 dB to 10 dB and spreading factor values of 256. y). 64.. Modifying the average inter-cell interference given by [7].g . y) and rj (x.g κji. These results are compared with simulations to determine the value of η experimentally. G that satisfy (10).. with perfect power control (PPC) between BSs and MSs. Capacity with Perfect Power Control In WCDMA. (7) = Sg vg nj.JOURNAL OF NETWORKS. g=1 (g) g N0 + W S∗ G G .g vg κji. JULY 2006 41 plementation is simple and considers only a single cell for admitting a call. g=1 inter Ii (4) from ni.g needs to be calculated only once.g . y) w(x. and 4. U SER AND I NTERFERENCE M ODEL This study assumes that each user is always communicating and is power controlled by the BS that has the highest received power at the user. which BS j services. The remainder of this paper is organized as follows. In Section IV. we analyze capacity for perfect and imperfect power control. we have for every cell i in the UMTS network.g = e(βσs ) Aj 2 1 W G Sg vg ni. and Sg be the maximum signal power.j=i g=1 nj. ri where N0 is the thermal noise density. UMTS C APACITY A.g . the number of users ni. and Rg is the inter bit rate for service g. m is the path loss exponent. Spreading factors are discussed in Section VI. (5) (g) Thus.g . 2πσ1 σ2 (6) where η is a user density normalizing parameter.j=i g=1 (3) for i=1.g vg κji. y) is the user distribution density at (x. VOL. which must received at a BS to decode the signal of a user with ∗ service g. III.. where cef f = Sg vg nj. II. Network throughput is determined in Section V.. Thus Ii is given by own Ii = (1) where β = ln(10)/10. (3) can be rewritten as inter Iji 1 = W G Sg vg nj. and w(x. y) = 1 x−µ1 2 1 y−µ2 2 η e− 2 ( σ1 ) e− 2 ( σ2 ) ..

Mobility Model There are several mobility models that have been discussed in the literature [10]–[12]. assumes perfect power control between the BSs and MSs..g (Eb )o.g ρj.g = n ∈ Ω : .g .g ).g .g = 0. or because the call is completed.g are. Let Ai be the set of cells adjacent to cell i. In this case.5 dB [4]. throughput). UMTS C ALL A DMISSION C ONTROL A.g )qji.. We assume that the occupancy of the cells evolves according to a birth-death process.g (Eb )o. Aj ) = λj. attempt a handoff to an adjacent cell. The call arrival process with service g to cell i is assumed to be a Poisson process with rate λi.g = i.g be the handoff rate out of cell j offered to cell i for service g. however. then the average dwell time of a call of the same service in the network will be constant regardless of where the call originates and what the values of qii. (18) .. which was formulated in section III-A. 1.g = ni. The call dwell © 2006 ACADEMY PUBLISHER time is a random variable with exponential distribution having mean 1/µ.g independent of other call arrival processes. nM. For example.g ’s for a service g. Bi. νji.g . cef f (g) IP C becomes cef f / e (g) (βσc )2 2 . then qij. (16) Equation (16) can be rewritten as νji.g is Gaussian with mean mc and standard deviation σc . and the departure rate from cell i when the network is in state n is ni. At the end of a dwell time a call may stay in the same cell. For the CAC problem that we are investigating here. (13) i. Let νji.g µi. µ the matrix of departure rates. (15) Bi.g and qij. IV..g (1 − Bj.g using the fact that xi. is in Bi.g .. Feasible States Recall from section III-A that the number of calls in every cell must satisfy (10)..g . then taking the expected value. is given by p(ρ. n1. ρj. the constant value of (Eb )i.g. n. (12) xi.g ’s and qij. which meet the E0 constraint. n) be the stationary probability that the network is in state n. According to [4].g as the probability that a call with service g in progress in cell i remains in cell i after completing its dwell time.1 .g . Let qij. by evaluating the nth moment of i. The mobility model that we use is presented in [13] where a call stops occupying a cell either because user mobility has forced the call to be handed off to another cell.g = 10log10 I0 to be a normally distributed random variable with mean mc and standard deviation σc .42 JOURNAL OF NETWORKS. we have (Eb )o. (14) Bi. which make the received Eo I i.g ) = (1 − Bj. VOL.g = λj.g . Deﬁne qii.g is constant for all i and g. Thus.g )qji. µ..G . 3. if qi.G If a new connection or a handoff connection with the service g arrives to cell i. nM.g . JULY 2006 B.g in each cell i for every user with service g needs to be replaced by the variable (Eb )i. n).g signals vary according to a log-normal distribution with a standard deviation on the order of 1. and let p(ρ.g . (17) where ρj. NO. call durations and elapsed times of other users. by varying qii. Deﬁne the set of blocking states for service g in cell i as n1. qji..g .g )qji. λj. or leave the network. Capacity with Imperfect Power Control The calculation of UMTS network capacity. it is blocked when the current state of the network. These models have ranged from general dwell times for calls to ones that have hyper-exponential and sub-exponential distributions. We deﬁne i.g (Eb )i. we can obtain low and high mobility scenarios and compare the effect of mobility on network attributes (e. This mobility model is attractive because we can easily deﬁne different mobility scenarios by varying the values of these probability parameters [13]. Thus νji.g = ρ(v. A set of calls n satisfying (10) is said to be in feasible call conﬁguration or a feasible b state.g µ(1 − qii. µ. transmitted signals between BSs and MSs are subject to multi-path b propagation conditions. B.g + x∈Aj νxj.g be the probability that a call with service g in progress in cell i after completing its dwell time goes to cell j. The new call blocking probability for service g in cell i.g .g x∈Aj νxj.g = n∈Bi. the total offered trafﬁc to cell j for service g. If cells i and j are not adjacent. . However.g . and it is independent of earlier arrival times.. We denote by qi. where the total arrival rate or offered trafﬁc for service g to cell i is ρi. Thus. ∈ Ω . is given by ρj.g + (1 − Bj.g the probability that a call with service g in progress in cell i departs from the network.g is the sum of the proportion of new calls of service g accepted in cell j that go to cell i and the proportion of handoff calls with service g accepted from cells adjacent to cell j that go to cell i. a new dwell time that is independent of the previous dwell time begins immediately. I Denote by Ω the set of feasible states.g This is also the blocking probability of handoff calls due to the fact that handoff calls and new calls are treated in the same way by the network. Let ρ be the matrix of offered trafﬁc of service g to the cells. such assumptions makes the problem mathematically intractable. which is log-normally distributed..5 to 2. in the imperfect power control (IPC) case.g (βσc )2 E = e 2 ..g = ν(Bj.1 .g I0 I0 As a result of (13).

Clearly the set of admissible states denoted Ω is a subset of the set of feasible states Ω.g −vg ≤ cef f . G M G subject to where Ni. G. . 1. the calculation of the blocking probabilities has a computational complexity that is exponential in the number of cells combined with number of available services. Since the cardinality of Ω is O(cef f M G ). which describes the offered trafﬁc as a function of the handoff rates and new call arrival rates. VOL. We deﬁne an initial value for the handoff rates. . λ). Maximization of Throughput A second optimization problem can be formulated in which the arrival rates and the maximum number of calls that can be admitted in all the cells are the independent Ni... We then calculate the new values of the handoff rates and repeat. C.g is a parameter which denotes the maximum number of calls with service g allowed to be admitted in cell i. This approach has been extensively utilized in the literature to obtain solutions of ﬁxed point problems [15]–[20]. the handoff rates as a function of the blocking probabilities and the offered trafﬁc. In this optimization problem the arrival rates are given and the maximum number of calls that can be admitted in all the cells are the independent variables. λ) = i=1 g=1 {λi. B(Ai..g for i = 1. N ETWORK T HROUGHPUT The throughput of cell i consists of two components: the new calls that are accepted in cell i minus the forced termination due to handoff failure of the handoff calls into cell i for all services g.g ≤ Ni. Calculation of N We formulate a constrained optimization problem in order to maximize the throughput subject to upper bounds on the blocking probabilities and a lower bound on the signal-to-interference constraints in (10). ρ. .g vg κji.g vg + g=1 j=1. i. Ni. ρ. is calculated (this is done ofﬂine and described in the next section).g ! Ni.. M and g = 1.. N. the CAC algorithm for cell i for service g will simply compare the number of calls with service g currently active in cell i to Ni. (20) B.g Ak /k! i. Furthermore. and the blocking probabilities as a function of the offered trafﬁc..g ) is the Erlang trafﬁc in cell i with service g.e. We note that the complexity to calculate the blocking probabilities in (20) is O(M G). In order to simplify the CAC algorithm. j = 1. Admissible States A CAC algorithm can be constructed as follows.g .g k=0 where Ai = ρi.g } . NO.M .g = B(Ai. the iterative approach converged to a unique solution. the number of calls in progress in the current cell.. Kelly has shown that for ﬁxed alternate routing the solution to the ﬁxed point problem is in fact not unique [21].j=i g=1 (g) Nj. Once the maximum number of calls with different service that are allowed to be admitted in each cell... The blocking probabilities are now calculated using the offered trafﬁc. (19) and the bit error rate requirement is guaranteed since Ω ⊂ Ω. (21) where B is the vector of blocking probabilities and λ is the matrix of call arrival rates.g − Bi. Ni. The total offered trafﬁc can be obtained from a ﬁxed point model [14]. This is given in the following max N T (B. we use an iterative method to solve the ﬁxed point equations. We calculate the offered trafﬁc by adding the given values of the arrival rates to the handoff rates. The questions of existence and uniqueness of the solution and whether the iterative approach in fact converges to the solution (if a unique solution exists) are generally difﬁcult to answer due to the complexity of the equations involved.g /µi. in all the numerical examples we solved. for i = 1. of the network is M G T (B.g = ρi. A call arriving to cell i with service g is accepted if and only if the new state is a feasible state. © 2006 ACADEMY PUBLISHER . To this end we deﬁne a state n to be admissible if ni. 3. to compute the blocking probabilities. βg .g .. i. The blocking probability for cell i with service g is then given by Bi. (22) N The optimization problem in (22) is solved ofﬂine to obtain the values of N. V. JULY 2006 43 and where v denotes the matrix whose components are the handoff rates νij for i.g ρi. Hence the total throughput. . we consider only those CAC algorithms which utilize local state.g in order to accept or reject a new arriving call.g /Ni. The goal is to optimize the utilization of network resources and provide consistent GoS while at the same time maintaining the QoS. the number of calls in progress in all the cells of the network.. Clearly this CAC algorithm requires global state.e. A.g ) ≤ βg . For a given set of arrival rates.JOURNAL OF NETWORKS. T . for all the users for different services g. the probability of each state in the feasible region needs to be calculated.g ) = i. Thus our CAC algorithm is implemented with a computational complexity that is O(1).g Ai.g /µ(1 − qii.. M.. .

This is the network capacity for the given GoS and QoS. The data signal after spreading is then scrambled with a scrambling codes to separate MSs and BSs from each other. 1 depicts the generation of different OVSF codes for different SF values. NO. Fig. is dramatically increased due to the incorporated Dedicated Control Channel (DCCH) information.. Fig. B(Ai.e.g vg κji. Table II shows the approximation of the maximum user data rate with 1 2 rate coding for different values of DPDCH. 3. (23) Spreading Yes. . S PREADING FACTOR Communication from a single source is separated by channelization codes.g Code family Channelization code Uplink: Separation of physical data (DPDCH) and control channels (DPCCH) from same MS Downlink: Separation of downlink connections to different MSs within one cell. 3 depicts the process of creating the actual transmitted signal for a voice user.44 JOURNAL OF NETWORKS. Usage variables and the objective function is the throughput. which is the actual transmitted data rate. The typical required data rate or Dedicated Trafﬁc Channel (DTCH) for a voice user is 12. and Radio Frame Alignment. The use of OVSF codes allows the orthogonality and spreading factor (SF) to be changed between different spreading codes of different lengths. JULY 2006 Fig. the Dedicated Physical Data Channel (DPDCH). 12. i. However. 1. Fig. Fig. Table I describes the different functionality of the channelization and the scrambling codes.g ) ≤ βg . λ). VOL. This is given in the following max λ. 2. VII. G M G Length subject to Number of codes Ni.j=i g=1 (g) Nj. The Orthogonal Variable Spreading Factor (OVSF) codes. Generation of OVSF codes for different Spreading Factors. TABLE I F UNCTIONALITY OF THE CHANNELIZATION AND SCRAMBLING CODES . 1. Rate Matching. the dedicated physical channel in the uplink and the downlink connections within one sector from one MS.g . does not affect transmission bandwidth −vg ≤ cef f . were used to be channelization codes for UMTS.. increases transmission bandwidth The optimized objective function of (23) provides an upper bound on the total throughput that the network can carry. Fig.2 Kbps Uplink Reference channel. thus it only makes the signals from different sources distinguishable from each other. ρ. N T (B. M. N UMERICAL R ESULTS The results shown are for a twenty-seven cell network topology used in [23]. for i = 1. 4 shows the DPDCH data rate requirement for 64 Kbps data user. Scrambling is used on top of spreading. .2 Kbps. VI.g vg + g=1 j=1. Relationship between spreading and scrambling. [24]. The COST-231 propagation model with a carrier frequency of 1800 MHz.. average base station height of 30 meters and average mobile height Fig. which were originally introduced in [22]. Ni.. Uplink: 4-256 chips same as SF Downlink 4-512 chips same as SF Number of codes under one scrambling code = spreading factor Orthogonal Variable Spreading Factor Scrambling code Uplink: Separation of MSs Downlink: Separation of sectors (cells) Uplink: 10 ms = 38400 chips Downlink: 10 ms = 38400 chips Uplink: Several millions Downlink: 512 Long 10 ms code: Gold Code Short code: Extended S(2) code family No. 2 depicts the relationship between the spreading and scrambling process. 3. and the processes of Channel © 2006 ACADEMY PUBLISHER Coding.

to get the same average number of slots per sector as PPC. with 6 parallel codes DPDCH bit rate 15 Kbps 30 Kbps 60 Kbps 120 Kbps 240 Kbps 480 Kbps 960 Kbps 5740 Kbps 1 User data rate 2 rate coding 7. 7. 6. NO. The processing gain Rg is 6. which is used to carry data for the control channel. µ1 = µ2 = 0. 4. 2-D Gaussian approximation of users uniformly distributed in the cells. Fig. 6 also shows that the traditional uniform user distribution modeling matches well with the 2-D Gaussian model.02 dB.5 I dB in PPC) as shown in Fig.5 dB. 64 Kbps Uplink Reference channel. v.375. is 0. Fig. 1. we set SF to 256. The network with different values for Eb I0 was analyzed for different SF values of 4.5 meters.JOURNAL OF NETWORKS. of 1. Average number of slot per sector for perfect and imperfect power control analysis with a Spreading Factor of 64. Fig. Capacity Allocation with SF of 256 First. A. Capacity Allocation with SF of 64 As a result of lowering the SF to 64. Fig.04 dB. 3. 5 shows the 2-D Gaussian approximation of users uniformly distributed in the cells with σ1 = σ2 = 12000. The UMTS network with 27 omnidirectional antenna cells (1 sector per cell) was analyzed for evaluation of capacity using user modeling with the 2-D Gaussian function and traditional methods of modeling uniform user distribution. DPDCH Spreading Factor 256 128 64 32 16 8 4 4. 7. is used to determine the coverage region. 16.5 dB to 1. 6 shows the maximized average number of slots per sector for the 27 cells UMTS network b as the E0 is increased from 5 dB to 10 dB and the standard I deviation of imperfect power control is increased from 0 to 2. © 2006 ACADEMY PUBLISHER . Average number of slot per sector for perfect and imperfect power control analysis with a Spreading Factor of 256. and 24. TABLE II U PLINK DPDCH DATA RATES . The activity factor. 5. B. and 256.06 dB. VOL. 64. and 256. The path loss coefﬁcient m is 4. Fig.08 dB for Spreading Factor equal to 4. 64.5 Kbps 15 Kbps 30 Kbps 60 Kbps 120 Kbps 240 Kbps 480 Kbps 2.58 to 15.56 slots when Eo = 7. Because of IPC. the number of slots per sector decreases by almost a factor of 4 compared to b SF equal 256 (from 60. respectively. σ1 = σ2 = 12000. 18. The shadow fading W standard deviation σs is 6 dB. JULY 2006 45 Fig. 12.8 Mbps Fig.5 dB. we have to decrease the SIR threshold by 0. 16.

g 0. the average dwell time of a call in the network is constant. Tables III and IV show respectively the © 2006 ACADEMY PUBLISHER First. Throughput Optimization with SF of 256 Fig. Because the throughput is calculated based on the number of simultaneous connections between MSs and BSs.700 0. Ai 3 4 5 6 qij. . qii. qij.240 qi. • • Ai is the number of cells.240 0. and high mobility of users are considered. which is used to carry data for the control channel.g 0. we analyze the average throughput per cell by dividing the results from (23) by the total number of cells in the network and multiplying by the maximum data rate in Table II. we set SF to 4. Three mobility scenarios: no mobility.050 qii. the number of slots per sector decreases by almost a factor of 4 compared to b SF equal 64 (from 15. The results for the average throughput for no mobility and high mobility cases are almost identical while the throughput for low mobility is higher for each blocking probability. Table V shows the optimized values of N for each cell for all three mobility models with perfect power control and 2% blocking probability.1 0.g is the probability that a call with service g in progress in cell i departs from the network. Fig.g 0.g = 0. 3. E.g 0. the lower trunking efﬁciency leads to lower throughput as shown in Fig.010 qii. Ai 3 4 5 6 • • qij.5 dB I in PPC) as shown in Fig. the number of slots per sector b decreases signiﬁcantly to 1.56 to 4. qii. we set SF equal to 256.700 0. which are adjacent to cell i.012 0. The following set of results are for the calculation of throughput. JULY 2006 TABLE III T HE LOW MOBILITY CHARACTERISTICS AND PARAMETERS .075 0. 8.700 0.49 while keeping Eo = 7. In all three mobility scenarios. 9. We assume that the mobility characteristics for a given service g stays the same throughout different cells in the network. Table VI shows the optimized values of N for each cell for all three mobility cases and SF equal to 64.30 slots when Eo = 7. 3. Average number of slot per sector for perfect and imperfect power control analysis with a Spreading Factor of 16.240 0. 10 shows the optimized throughput per cell for a blocking probability from 1% to 10%. qi. Capacity Allocation with SF of 4 Next. NO.015 0. In the numerical results below. VOL. Thus.g is the probability that a call with service g in progress in cell i after completing its dwell time goes to cell j. As a result of lowering the SF to 64. low mobility. 8. The following parameters are used for the no mobility case: qij. for each SF value. regardless of where the call originates and mobility scenario used.g is the probability that a call with service g in progress in cell i remains in cell i after completing its dwell time. D. which is used for voice communication as shown in Fig.700 TABLE IV T HE HIGH MOBILITY CHARACTERISTICS AND PARAMETERS .240 0. we set SF equal to 64. Throughput Optimization with SF of 64 Next. 11. Average number of slot per sector for perfect and imperfect power control analysis with a Spreading Factor of 4.g 0. 1. Capacity Allocation with SF of 16 As a result of lowering the SF to 16.3 and qi. C.g = 0. the probability that a call leaves the network after completing its dwell time is 0.700 0. which is used for 256 kbps data communication between BSs and MSs.g 0 0 0 0 qi. the number of possible concurrent connections within one cell is also decreased.060 0.7. F.7 for all cells i and j. The low mobility case has an equalizing effect on trafﬁc resulting in slightly higher throughput.46 JOURNAL OF NETWORKS.700 Fig. 9.700 0.g = 0. mobility characteristics and parameters for the low and high mobility cases.020 0.5 dB I in PPC as shown in Fig.700 0. As a result of lowering the SF to 4.

74 73.71 50.27 62.32 13.86 18. No Mobility Ni 52.29 62.40 71.30.43 71.40 Low Mobility Ni 52.40 High Mobility Ni 52.86 13.03 13.83 71.5 dB due to the imperfect power control.61 13.82 73.46 18.03 18.82 73.86 13.11 13.84 53.83 71.73 50.11 16.84 53. Throughput Optimization with SF of 16 Next.84 71.32 13.85 53.85 18. VIII.86 73. the average slots per © 2006 ACADEMY PUBLISHER sector decreases signiﬁcantly to 1.84 51.61 13.86 53.74 73.45 18. C ALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF = 256 AND BLOCKING PROBABILITY = 0.71 63.40 TABLE VI C ALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF = 64 AND BLOCKING PROBABILITY = 0. the average number of slots within one cell decreases to 4.45 18.73 53.32 13.32 13.73 62.83 71.5 dB as shown in Fig.5 to 1.46 18. Table VII I shows the optimized values of N for each cell for all three mobility models with SF equal to 4.01 50. The average throughput for all three mobility cases are almost identical as shown in Fig.00 50.62 13. TABLE V Fig.74 63.43 73.73 50. is much lower compared to the case with SF equal to 64 or 256.11 13.45 18.JOURNAL OF NETWORKS.95 51.26 16.71 50.27 62.71 63. we set SF equal to 4. 10. JULY 2006 47 Fig. which is normally used for 256 Kbps data communication between BSs and MSs.71 50.73 53. H.85 18.85 Low Mobility Ni 13.03 16.11 13.95 51.86 13.85 51.86 73.03 16. 11.73 62. No Mobility Ni 13.32 13.02. 13.85 53.43 71.03 13. Numerical results show that the SIR threshold for the received signals is decreased by 0.73 62.84 51. 12.86 13.85 51.03 16. The resulting throughput.86 18. Average throughput in each cell for SF = 64.32 13.73 53. as shown in Fig.32 13. we set SF equal to 16.03 16.45 18.58 13.40 71.11 13.85 51. 3. Throughput Optimization with SF of 4 Next.11 16.01 50.03 18.45 18.86 53.58 13.29 62.95 51.11 16.27 62.43 73. As a result of lowering the SF to 4.61 13.02.86 18.25 16.03 18.03 16.84 71.86 53.11 13.26 16.43 71.86 13.32 13. Average throughput in each cell for SF = 256.00 50.26 16.46 18. As expected.40 71.45 18.58 13. VOL.11 16.11 16.32 13. which is used for 64 Kbps data communication as shown in Fig. Table VII shows the optimized values of N for each cell for all three mobility cases with SF equal to 16. 1.03 13.29 62.32 13.00 50.74 63.32 13.85 Cell ID Cell1 Cell2 Cell3 Cell4 Cell5 Cell6 Cell7 Cell8 Cell9 Cell10 Cell11 Cell12 Cell13 Cell14 Cell15 Cell16 Cell17 Cell18 Cell19 Cell20 Cell21 Cell22 Cell23 Cell24 Cell25 Cell26 Cell27 Cell ID Cell1 Cell2 Cell3 Cell4 Cell5 Cell6 Cell7 Cell8 Cell9 Cell10 Cell11 Cell12 Cell13 Cell14 Cell15 Cell16 Cell17 Cell18 Cell19 Cell20 Cell21 Cell22 Cell23 Cell24 Cell25 Cell26 Cell27 G.32 13.49 with perfect power b control and E0 = 7.45 18.62 13.82 73.43 73.95 51.73 62. The .73 62.25 16.71 63.25 16.11 13.11 16.74 73.03 16.45 18.84 71.86 13. 4.01 50.86 73.84 51.85 High Mobility Ni 13. C ONCLUSIONS An analytical model has been presented for calculating capacity in multi-cell UMTS networks.73 62.86 18. NO.73 50.95 51.62 13. As a result of lowering the SF to 16.86 18.85 18.95 51.85 53.84 53.74 63.45 18.86 18. we can have many low rate voice users or fewer data users as the data rate increases. 9.

92 1.30 1. Hegde. 1.30 1.22 5.10 5. TABLE VII Fig.82 Cell ID Cell1 Cell2 Cell3 Cell4 Cell5 Cell6 Cell7 Cell8 Cell9 Cell10 Cell11 Cell12 Cell13 Cell14 Cell15 Cell16 Cell17 Cell18 Cell19 Cell20 Cell21 Cell22 Cell23 Cell24 Cell25 Cell26 Cell27 Cell ID Cell1 Cell2 Cell3 Cell4 Cell5 Cell6 Cell7 Cell8 Cell9 Cell10 Cell11 Cell12 Cell13 Cell14 Cell15 Cell16 Cell17 Cell18 Cell19 Cell20 Cell21 Cell22 Cell23 Cell24 Cell25 Cell26 Cell27 results also show that the determined parameters of the 2-dimensional Gaussian model matches well with traditional methods for modeling uniform user distribution.59 0.21 5.83 3. No Mobility Ni 1.68 3. Min. J. Castro. [7] R.50 4.” IEEE Veh.10 5.10 5.10 5.16 1.76 3.22 5.02. No Mobility Ni 3. “CDG : Worldwide : CDMA World- wide.80 1.27 1.68 3. 3. “UMTS air interface voice/data capacitypart 1: reverse link analysis.82 1. © 2006 ACADEMY PUBLISHER .59 1. Pedersen.45 3. “UMTS air interface voice/data capacity-part 2: forward link analysis. Mogensen. R EFERENCES [1] CDMA Development Group.45 3.28 1. pp. Viterbi. Conf.23 1.” IEEE Veh. 711–722. “Comparison of multiple access schemes for UMTS.. [2] T.83 3.68 3.68 3.10 5.68 3.60 4.10 5..53 1.54 1.93 1.83 1.53 1. vol. Conf..53 1.28 1.59 1. [5] Q.83 3. “Multi-cell CDMA network design.54 1.” IEEE Trans. [3] Q.10 5.23 1. and A.48 1.79 1. TABLE VIII C ALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF = 16 AND BLOCKING PROBABILITY = 0.22 5.22 5. 12. May 2001. Numerical results show that as the spreading factor increases.82 1. Addison-Wesley.45 3.80 1. May 1997. 53.79 1. 1995.” IEEE Veh.60 4.10 5.48 JOURNAL OF NETWORKS. accurate.10 5. and can be implemented for large multi-cell UMTS networks.79 1.32 1. NO. and P.53 1.28 1.79 1. pp.58 1.22 5.76 1.45 4.58 0.94 1. July 2004.46 3.46 4. 2725 – 2729.80 1. Average throughput in each cell for SF = 16. Ojanpera. Carnero.92 1. no. “Capacity gain of an uplink-synchronous WCDMA system under channelization code constraints.92 1.94 1. Technol.54 1. An analytical model was also presented for CAC algorithm for optimizing the throughput in multi-cell UMTS networks.10 5.27 1. M.68 3.46 4. Yue. due to the trunking efﬁciency for all three mobility models (no.93 1. 490–494. K.48 1.77 3. Technol. vol. the optimized throughput is better.50 4.49 4. Naraghi-Pour. and high mobility).asp?h area=0. [6] J.60 4. vol.10 5. CDMA Principles of Spread Spectrum Communication.21 High Mobility Ni 3.93 1.83 1. vol.59 0. Technol. 4.76 1.77 3. JULY 2006 Fig.68 3.27 1.27 1.50 4. Akl.76 1. Conf. 50.21 Low Mobility Ni 3. M.10 5. 4. 13.55 1. pp..68 3.93 1.77 3.68 3. vol.60 4.68 3.org/worldwide/index.93 0. Conf. Skold. pp.83 High Mobility Ni 1.80 1.49 4.46 3.32 1.21 C ALCULATION OF N FOR UNIFORM USER DISTRIBUTION WITH SF = 4 AND BLOCKING PROBABILITY = 0..76 3.49 4.83 3.30 1.58 0. Veh.76 1.30 0.60 4.” IEEE Veh.83 1.75 3.02.75 3.28 1.42 1.60 5.76 3.94 0.68 3.23 1.82 1. L.22 5.68 3.60 5.32 1. VOL. 2730 – 2734.37 0. Average throughput in each cell for SF = 4. 2. J. May 2001.60 3.24 1. and P.45 4.83 3. pp. Zhang and O.45 4. 982 – 991.32 1. Technol.60 4.42 1.cdg. 3.54 1.58 1.83 3. Klein. Technol.93 1. Zhang.55 1. Girard.60 3. [4] A.” http://www.24 1.21 5.37 0.46 4.92 Low Mobility Ni 1.46 3.21 5.75 3.60 3. Our methods for maximizing capacity and implementing the CAC algorithm are fast. May 2001.94 0. low.60 5.32 1.

Min. Heck. M. “On the handoff arrival process in cellular communications. [13] C. he was a senior systems engineer at Comspace Corporation. 1. 1763–1767. Conf.S. and Control Technol. © 2006 ACADEMY PUBLISHER . Technol. 18. March 2005.. [12] Y. degree in computer science and the M. Technol. and P.” Electr. Akl is currently an Assistant Professor at the University of North Texas. pp. June 1999. Conf. [19] ——.S. Fixed Point Theory : An Introduction.. August 1996. F.” IEEE Veh. K. pp. Tugcu and C. 1988. Hegde. 147–157. and 2000. pp. Department of Computer Science and Engineering. vol. vol. Spring 2001. Technologies and Applications. vol. 323–337. M.Sc. and M.D. degrees in electrical engineering in 1994. vol. K. January 1997. Vargas.. B. Weller. “Routing in circuit-switched network: Optimization. pp. July 2005. and K.. Robert Akl received the B. VOL. JULY 2006 49 [8] S. Naraghi-Pour. pp.” Advances in Applied Probability. pp. pp. 33. Chlamtac. and M. Ersoy. vol. Tran-Gia. [22] A. “ATM network design and optimization: a multirate loss network framework. Mitra. Conf. Rappaport. Akl. Lett. 1996. July 2004. “Approximating the othercell interference distribution in inhomogenous UMTS networks. May 2004. pp. respectively. pp. Technol.-B.S. and the B. dissertation. 1986. Hegde..” Ph.. M.. 531–543.” J. “Blocking probabilities in large circuit switched networks. 473–505. 3.JOURNAL OF NETWORKS. 20. 7. [24] R. [21] F. 54. Fang. vol. 639–651. K. 1997. 4. Louisiana State University. [9] D.” Advances in Applied Probability. 404–408.. His research interests include 3G wireless network design and optimization. “Blocking effects of mobility and reservations in wireless networks..” Proceedings of International Conf. on Networking. May 1998.” IEEE/ACM Trans. [14] V. A.S. Vargas-Rosales.S. 904–908. 1981. 3. “Communication network design and evaluation using shadow prices. Schroder. 5.” IEEE Veh. “Approximating user distributions in WCDMA networks using 2-D Gaussian. pp. [20] C. Orlik and S. Kelly. Leibnitz. on Networking. 112–144. “Effects of call arrival rate and mobility on network throughput in multi-cell CDMA. Dr. vol. “Modeling PCS networks under general call holding time and cell residence time distributions. 893–906. Morrison. he was an Assistant Professor at the University of New Orleans. vol. vol. Istratescu. D. I. Technol. 2001. In 2002. Nguyen and R.. [17] D. vol. [16] F.” Wireless Networks. vol.” IEEE/ACM Trans. and Y. Parvez. Akl. Akl and A.. 27–28. “Mobility-based CAC algorithm for arbitrary trafﬁc distribution in CDMA cellular systems. vol. Coppell. Commun. 1612–1616. TX. “Impact of interference model on capacity in CDMA cellular networks. on Comput. 1640–1644. pp. From October 2000 to December 2001. 1996. “Implied costs in wireless networks. 3. Kelly. in 1994. Naraghi-Pour. Son Nguyen received the B. S. He also received the Dual Degree Engineering Outstanding Senior Award from Washington University in 1993. and O. Hegde. 3.” IEEE International Conf. May 2002. 2. “Application of a realistic mobility model to call admissions in DS-CDMA cellular systems. He is a senior member of IEEE. [11] P. vol. Staehle. Akl. Ramakrishnan. Vargas. degree in computer science from Washington University in St.” IEEE International Conf. M. June 1998. Reidel. M.” IEEE Trans. 10. Lin.. no. [23] R. pp. Naraghi-Pour. [15] R. [18] C. 1047–1051. pp. His research interests include wireless communication and network design and optimization. vol. 4. M. Veh.. Department of Electrical and Computer Engineering.Wireless Networks. 2. shadow prices and decentralization. “Implied costs for multi-rate wireless networks. [10] T. respectively. NO. and P. Hegde. degree in computer science from The University of North Texas in 2001 and 2005. J. and M.” Proceedings of SCI 04: Communication and Network Systems.” IEEE Veh. 4. “Tree-structured generation of orthogonal spreading codes with different lengths for forward link of DSCDMA mobile radio. pp.. on Commun. and D. on Commun. Louis.