You are on page 1of 5

Calculation of Erlang Capacity for Cellular CDMA Uplink Systems

Ling Ding and James S. Lehnert


School of Electrical and Computer Engineering
Purdue University, West Lafayette, IN 47907-1285, U.S.A.
E-mail: ding@ecn.purdue.edu, lehnert@purdue.edu

Abstract Three methods of determining the


Erlang capacity of a cellular CDMA uplink system are presented. The first method decouples
the analysis of blocking and outage performance,
and avoids iterative search. Based on two approximationsregarding the mobile traffic and the
interference, the second method jointly examines
blocking and outage performance. The third,
more detailed, method considers the mobile traffic characteristics and the interference with mobile traffic fluctuations. The Erlang capacity of
a cellular CDMA uplink system using a power
control scheme called truncated channel inversion is calculated with each of the methods. The
methods are also used to compare the Erlang capacities for different power control schemes.
I. INTRODUCTION
The economic value of cellular wireless systems can
be effectively measured by Erlang capacity, defined as
the maximum load that can be supported with a given
blocking probability. In channelized multiple-access systems, such as TDMA/FDMA, each cell is typically assigned a fixed number of channels. Therefore, Erlang
capacity of these systems can be easily obtained by
the well-known Erlang-B formula. However, in cellular CDMA systems, where users all share a common
spectral frequency allocation over time, the notion of
channels per cell is soft, in the sense that a new user can
be admitted as long as the signal-to-interference ratio is
adequate for receiver processing [I]. Consequently, the
blocking performance is coupled with the outage performance, and it is difficult to calculate Erlang capacity in
cellular CDMA systems.
Here, we present three methods of determining the
Erlang capacity. The first method decouples the analysis of blocking and outage performance to avoid iterative
search. Based on two approximations on the mobile traffic and the interference, the second method attempts to
jointly analyze blocking and outage performance. The
third method exactly captures the mobile traffic characteristic and the interference under mobile traffic fluctuation, thereby providing the most accurate results.
The research described in the paper is supported by the U.S.
Government DARPA Glomo Project A 0 No. F383, AFRL contract number F30602-97-C-0314.

0-7803-6596-8/00/$10.00 0 2000 IEEE

To compare these methods of calculating Erlang capacity, we consider a power-controlled cellular CDMA
system. The power control scheme is based on truncated channel inversion, which was recently proposed in
[2] for data traffic and shown to outperform substantially the traditional channel inversion scheme in terms
of system throughput and power consumption. The Erlang capacities of the truncated and traditional power
control schemes are also compared using the presented
methods.

11. SYSTEM DESCRIPTION


OF THE POWER
CONTROLLED
CDMA UPLINK
Consider a cellular CDMA system where basestations
are located at centers of the hexagons and serve mobiles
in the system. In this paper, we concentrate on the
uplink, which is from a mobile to its basestation.
A data mobile arrives at a cell according to a Poisson
process with rate A,. The basestation decides whether
to accept the mobile based on some admission control
policy. In this paper, we consider a simple admission
control in which a mobile is admitted provided that the
total number of mobiles in the cell does not exceed a
fixed threshold K . Once the mobile has been admitted, it stays in the cell for an exponentially distributed
duration with mean l/pu. The arrival and departure
processes of mobiles are all independent. In this paper,
we do not consider handoffs.
While in the cell, a mobile generates and transmits
data packets. Data traffic is described by an on-off
model. Specifically, a data source assumes alternate
Lonand LLofflstates. Durations of being in on and
off states are exponentially distributed with means
b-l and U-, respectively. At on state, data packets
are generated according to a Poisson process with rate
A. No packets are generated at off state. The size of
each packet is independently exponentially distributed
with mean L,.
In order to satisfy the bit error rate requirement, it is
generally required that
exceed a certain threshold,
where Eb is the received signal energy per bit and No is
the power density of interference. Let Pt and P, denote
the transmitted and received powers, respectively. It
=
=
where R is the transfollows that

338

q,

mission bit rate and C is the channel gain between the


mobile and its basestation. In a wireless environment,
the channel gain generally consists of three components,
namely, path loss, slow log-normal shadowing, and fast
Rayleigh fading. For the purpose of studying power control, it is commonly assumed [3] that Rayleigh fading is
averaged, and thus it will not be taken into account in
this paper. Hence, the channel gain at time t is modeled by C = r(t)-~lOO.'c(t),
where r(t) is the distance
between the mobile and the basestation, q is an order of
power path loss, and t(t2is a zeremean normal random
variable with variance U representing the shadowing effect.
As the channel gain C changes from time to time,
power control is used to adjust Pt to meet the Eb/No
requirement. Note that in a CDMA system, the signal
from a mobile is the interference to all other mobiles.
The idea of power control is to allocate an appropriate
transmission power level for each mobile in the CDMA
system to maintain its requirement without generating
unnecessary interference to others.
The power control scheme employed in the current
voice CDMA system is based on channel inversion [3].
The requirement of constant bit rate for voice mobiles
indicates that P,. has to be controlled at a constant level,
normalized to be 1. Therefore, the power control scheme
is to make Pt proportional to the inverse of the channel
gain. The problem of the channel inversion scheme is
that when the channel condition is bad, mobiles have
to increase their transmission power dramatically, thus
causing excessive interference to neighboring cells and
adversely impacting system capacity.
The requirement of constant received power is unnecessary for data traffic that does not have a strict delay
constraint. In order to achieve higher system capacity,
the power control for data traffic can be done in an a d a p
tive manner with respect to the channel condition. Now,
let the received power P,. be g, where g represents an
adaptive strategy to be used. In [2], a truncated channel inversion power control scheme is examined, where
g is given by

Here, P is the power control threshold. Equation (1)


indicates that depending on whether C 2 p, the channel condition is categorized into good and bad states.
The channel service time of transmitting data packets at
good state is exponentially distributed with mean p - I , '
where p = z. Mobile transmission is suspended when
LP
the channel is in a bad state, thereby reducing interference to neighboring cells.
Blocking and outage are two important performance
measures in the above power-controlled CDMA system.

Blocking occurs when an incoming mobile cannot be admitted. Outage occurs when a mobile admitted in the
cell cannot maintain the Eb/No requirement. Given the
requirements of the blocking probability (Pb)O and the
outage probability (P,,t)o,Erlang capacity is defined as
the maximum load, in terms of Au/pu, that the system
can support. In order to determine the Erlang capacity, in the next section we develop an analysis model to
calculate the blocking and outage probabilities.

111. ANALYSIS
MODEL
Given the admission control threshold K , the calllevel performance can be model by a Markov chain. The
probability that there are i mobiles in the cell is given
by

for i = 0,1,2, ...,K . The blocking probability is therefore equal to p ~ .


Note that in the admission control, K is treated
as the effective system capacity. Recall that in a
TDMA/FDMA system, the system resource is completely channelized, and K is thus simply equal to the
number of channels. In a CDMA system the concept
of system resource, i.e. capacity, is soft. The determination of K is based on the outage probability that
mobiles can tolerate. In the remainder of this section,
we examine the outage probability.
Recall from Equation ( 1 ) that the wireless channel
switches between good and bad states as r-~(t)lOO.lc(t)
crosses level P. It has been shown in [2] that with a
certain wireless channel propagation model, the dwell
time of the wireless channel staying in either a good
or bad condition can be approximately modeled as an
exponentially distributed random variable. Thus, the
wireless channel can be modeled as a two-state Markov
chain, where the channel switches between a good state
and a bad state according to an alternating renewal process. Recall that the data arrival process is a Markov
modulated Poisson process, which is independent of the
service process of the wireless channel. Based on the
memoryless property of the two processes, the arrival
and service process models can be integrated together
to develop a two-dimensional continuous-time Markov
chain for the entire queueing system of a data mobile
transmitting packets. The Markov chain model is used
to investigate the power activity of data traffic, defined
as a binary random variable Z ( r ) , where Z(T) = 1 if
the mobile is transmitting, and Z(T) = 0, otherwise.
From the above discussion of the wireless channel model,
it follows that Z ( r ) = X ( r ) Y ( r ) ,where Y ( r ) is a binary random variable representing whether the channel
is good, and X ( T ) is a binary random variable representing whether the mobile is transmitting given that

339

the channel is good. We define conditional power activity factor C(r) as the expected value of X ( r ) . C(r) can
be determined from the stochastic models discussed previously. C(r)is determined by solving the Markov chain.
The details can be found in [2], and are omitted here.
We next examine the interference. We focus on an
arbitrary cell 0, The interference received at the basestation 0 consists of intra-cell interference I, caused by
mobiles belonging to cell 0, and out-cell interference I,
caused by mobiles belonging to all other cells.
Suppose mobiles are densely and uniformly distributed in the system. Denote by 7 the user density.
With the radius of each cell normalized to unity, the average number of mobiles per cell is given by K = 9 7 0 .
It follows that
K

I,

=C

Z(j),
j=1

2 / 2

where Q ( g ) = s g w k d x and f(E) is the Gaussian


density function with zero mean and variance cr2.
We are now ready to examine the outage probability.
The outage probability is defined as

where Z ( j ) is the power activity variable of mobile j. It


can be shown that

E[Ia] = K G , and
m r [ ~ a I = K[Ina - (Cna121
with

Cna

(3)

2 ),

where (%)o is a constant depending on the physi-

(4) cal layer communication requirement. Clearly, received

2 3,

a+bpL

I, consists of interference from all neighboring cells.


It follows that

. +(to

;(

Pout
= Prob - < (-)o

&/No is given by
=
where W denotes total
bandwidth. With the assumption that la+Ieis Gaussian, the outage probability at cell 0 is given by

- tz,~o/ri)dA,

where @(Eo - &,ro/ri) is the indicator function of the


mobile belonging to cell i, i.e., if (ro/ri)QIO(c~-cO)/10
>
CAPACITY
IV. ERLANG
1, then @(SO - Ji,ro/ri) = 1. Here, r0,ri denote the
distance from the mobile to basestations 0 and i , reGiven the above analysis model of the blocking and
spectively, and &, & denote the associated shadowing
outage probabilities, we examine three methods to d e
variables. The cell index i is determined with the smalltermine the Erlang capacity in this section.
est distance principle [3]

ri

It can be shown that

= minrl.
If0

(5)

A . Method I
This first method is very close to the analysis approach commonly used in a TDMA/FDMA system.
Specifically, the approach is to first determine the effective system capacity K given the outage probability
requirement, and then determine the Erlang capacity
given K and the blocking probability requirement.
When calculating K , we assume that each cell is
equally loaded with K mobiles, that is, operates at its
capacity. Hence, for a given (PqUt)o
requirement, we
determine the maximum K such that PoutI (Pout)o.
From K and a given ( p b ) O requirement, we determine
the maximum Xu/pu, such that p~ 5 ( p b ) O , where p~

340

is given by Equation (2). This method decouples the


blocking and outage performance, and thus does not require any iterative search.

The first assumption follows from the distance symmetry


of all first ring cells and all second ring cells. The second
assumption is based on Equations (3), (4), (6), and (7).

B. Method I1

Let us denote the out-cell interferences from the first


ring and the second ring as Iel and l e 2 , respectively.
Then, according to the above observations, we have

The second method follows an approach as in [l].In


[l],no explicit admission control is applied, and incoming mobiles are always admitted. Therefore, there is no
blocking. The only performance measure is the outage
probability. When calculating the outage probability,
two important approximations are made. First, though
random, the number of mobiles is the same in each cell.
Second, the out-cell interference, in terms of both mean
and variance, is equivalent to a fraction of the intra-cell
interference.
For the sake of comparison, admission control is still
employed in the second method. We use iterative search
to determine the Erlang capacity. Specifically, for a
given Au/pu, we determine the minimum K such that
p~ 5 (Pa),-,, and given K, further calculate the distribution of the number of mobiles in a cell, according to
Equation (2). Then, the outage probability is obtained
using the method described in [I]. We adjust Au/pu and
repeat the above procedure until the outage probability
requirement is satisfied.

C. Method III
The two methods presented above are not always very
accurate. Specifically, as mobiles arrive at and depart
from cells randomly, the loads in individual cells fluctuate and are not always equal to If. The first method
assumes all cells operate at the capacity level, thus overestimating the interference and leading to pessimistic
results of the Erlang capacity. The second method attempts to capture the fluctuation of mobile numbers.
However, the method is based on the two approximations, which may not accurately reflect mobile traffic
and interference.

We next present an exact calculation method that


captures both the interference and mobile traffic precisely. Suppose cells i = 1,.. . ,6 are located in the first
ring and cells i = 7 , . . . ,18 are located in the second
ring. We denote by Ni the number of mobiles in cell i,
and make the following observations.
1. The out-cell interference generated by the mobiles in
the first or second ring has the equivalent interference
Ni mobiles
effect as that generated by E:=, Ni or
that are uniformly distributed in the first or second ring.
2. The out-cell interference, in terms of mean and variance, is directly proportional to the number of mobiles
in other cells, given that these mobiles are uniformly
distributed.

cfil

where
E&,N,, N2=A 18 Ni,and I e , 1 and I e , 2
are the out-cell interferences from the first ring and the
second ring, respectively. The mean and variance of le,l
or l e , p are readily calculated by Equations (6) and (7)
except that the integration area is now only the first ring
or the second ring, as discussed below.
The smallest distance principle, Equation (5), basically states that the mobile belongs to either basestation
i or basestation 0. Thus, the out-cell interference from
a first-ring neighboring cell is only caused by mobiles 10cated in that cell and mobiles located in the 60" sector
of cell 0 that is closest to that cell. The out-cell interference from a second-ring neighboring cell is only caused
by mobiles located in that cell. However, the intra-cell
interference is caused by mobiles located in any cell as
long as they belong to cell 0. Hence, the integration
area for EIIe,l]and var[le,l]covers cells 0,1,. ..,6, and
covers cells
the integration area for E[le,2]and va~-[l,,~]
7,. . ., 18. Hence, given No mobiles in cell 0, NI mobiles
in the first ring, and N2 mobiles in the second ring, the
conditional outage probability at cell 0 is given by

The probability distributions of Nl and N2 can be


obtained via their characteristic functions, denoted by
@1(z) and @ 2 ( z ) , respectively. Assume that { N i } are
independent for i = 1,. . .,18. Given Au/pu, we determine the minimum K such that p~ < (Pb)o, and given
K , calculate the characteristic function

341

180

Po

200

180

240
Edanp Capadly

280

280

300

Plots of Erlang capacity versus outage probability with

Fig. 1. Plots of Erlang capacity versus outage probability with


(%)O
= 2%.

Fig. 2.

where p j is given by Equation (2). Hence, we have

In Figure 2, the above numerical example is repeated, except that the requirement of blocking probablity, ( p b ) O , is tightened t o 0.2%. Clearly, the differences
among the results of the three analysis methods are increased. Overall, we note that method I11 provides the
most accurate Erlang capacity results and that method
I can be used as a simplified way of roughly determining
the admission control threshold K .

(Pb)O= 0.2%.

6K

Q1(z)

[@O(Z)]~:

hNl,lZN1

and

Ni =O
12K

@2(~)=

[@o(z)]~~ h

~ ~ , 2 ~ ~ .

Nz =O

Therefore, the overall outage probability is given by


K
Pout

6K

12K

( K )=

p a t (NO,

N2)PNo

hNi ,lhNZ 92.

No=O Ni=O Nz=O

For the purpose of comparison, we also show the Erlang capacity results of the traditional channel inversion
scheme in Figures 1 and 2. Clearly, the truncated power
control scheme significantly improves the Erlang capacity.

We adjust Au/pu and repeat the above procedure until


the outage probability requirement is satisfied.

V. NUMERICAL
RESULTS
In this section, we numerically investigate the three
methods of determining the Erlang capacity. We consider the following set of parameters: 0 = 8, 77 = 4,
v = 30 miles/hr, a = 0.1 s-l, b = 1 s - l , X = 2 packets/sec, L, = 600 bytes, W = 5 MHz, R = 28.8 kbps,
and ( %)o = 10 dB. The blocking probability requirement is ( p b ) O = 2%.
In Figure 1, we plot the Erlang capacity versus the
outage probability requirement. From the figure, we observe that the results given by methods I and I1 are very
close results, and are about 10% below those obtained
by the exact calculation of method 111. Numerical results
given by the three analysis methods are relatively close
partially because the blocking probability requirement
is not stringent. As a result, the number of mobiles in a
cell is close to K with large probability, and, therefore,
traffic fluctuation among cells is not very significant.

VI. CONCLUSION
Methods have been compared for calculating Erlang
capacity in a cellular CDMA system. The most detailed
method shows that the other methods give somewhat
pessimistic results. The power control scheme of truncated channel inversion is shown to result in a larger
Erlang capacity than the power control scheme of traditional channel inversion.

REFERENCES
[l] A. M. Viterbi and A. J . Viterbi, Erlang capacity of a power

controlled CDMA system, IEEE Journal On Selected Areas


I n Communications, vol. 11, no. 6, pp. 892-890, August 1993.
[2] L. Ding and J. S. Lehnert, Performance analysis of an u p
link power control using truncated channel inversion for data
traffic in a cellular CDMA system, in Pmc. IEEE Vehicular
Technology Conference (an extended version was submitted to
IEEE Transactions on Vehicular Technology), May 2000.
[3] K. S. Gilhousen, I. M. Jacob, R. Padovani, A. J . Viterbi,
Jr L. A. Weaver, and C. E. Wheatley 111, On the capacity of
a cellular CDMA system, IEEE Transactions on Vehicular
Technology, vol. 40, no. 2, pp. 303-312, May 1991.

342

You might also like