Improving the Accuracy of SINR Threshold Lower
Bound for SINR-based Call Admission Control in
CDMA Networks
Mohamed H. Ahmed
Faculty of Engineering and Applied Science
Memorial University of Newfoundland
St. Johns, Canada
mhahmed@engr.mun.ca
Abstract-Call admission control (CAC) is used in CDMA
networks to guarantee the signal quality in terms of the signal-to-
interference-and-noise ratio (SINR). SINR is employed as the
criterion for user admission by comparing the SINR with a
threshold value (SINRth). A small threshold level is desirable to
reduce the blocking rate. However, a lower bound of SINRth is
essential to keep the outage probability (Pout) below a maximum
value. A lower bound of SINRth (SINRth-lb) has been derived in [6]
by finding the relationship between Pout and SINRth. Then,
SINRth_lb is determined as the lowest SINRth that keeps Pout below
a certain Pout_max. In this paper, we improve the accuracy of
SINRth-lb by modeling the admitted traffic using a Markov model
(instead of the inaccurate Poisson model used in [6]). In addition,
we improve the accuracy by choosing the transition value of the
step function that approximates the conditional outage probability
distribution using the minimum mean square error criterion.
Keywords: Quality of service; call admission control; imperfect
power control.
I. INTRODUCTION
Call admission control (CAC) is used in CDMA networks
to maintain the signal quality in terms of the signal-to-
interference-and-noise ratio (SINR). SINR-based CAC schemes
that use SINR as the call admission criterion have been widely
studied [e.g., 1, 2]. In the reverse link of CDMA systems,
perfect power control is usually assumed such that received
power and SINR at the base station (BS) are fixed regardless of
the users location or the channel condition (including
shadowing and fast fading). In reality, however, a number of
factors including power control (PC) command errors and
response delays cause fluctuations in the received signal.
Hence, it has been shown by simulation and from field
measurements that the received power and SINR at BS is well
approximated by the lognormal distribution [3, 4].
In [1, 2] the only constraint imposed on the SINR threshold
value (SINRth) is the following condition: SINRth>SINRmin,
where SINRmin is the minimum SINR level for acceptable signal
quality. Obviously, setting a high value for SINRth maintains
the signal quality but increases the blocking probability (Pb)
and reduces the network utilization. Therefore, an upper bound
Mahmoud Elsayes
MacDonald, Dettwiler and Associates Ltd. (MDA)
Toronto, Canada
melsayes@hotmail.com
of SINRth, SINRth_ub, has been derived in [5] in order to
guarantee that Pb is kept below a maximum value (Pb_max).
On the other hand, lowering SINRth is desirable for better
resource utilization. However, small SINRth allows more users
to be admitted and might render the PC infeasible if the number
of admitted users per cell exceeds a certain limit. If PC turns
out to be infeasible, the outage probability (Pout =
P(SINR<SINRmin)) increases since SINR converges to a smaller
level than the target value (SINRtrg). Outage might also take
place even if power control is feasible due to SINR fluctuation
around the target value [6].
A lower bound of SINRth (SINRth_lb) that keeps the outage
probability limited (less than a maximum value Pout_max) in the
reverse link of a single-class CDMA system has been derived
in [6]. An exact value (based on numerical solution) and an
approximate closed form of the lower bound have been
obtained in [6]. However, the lower bound is obtained
assuming the number of active users per cell is modeled by
Poisson distribution. This assumption is not necessarily
accurate since the admitted traffic rate is not constant because
of its dependence on the number of existing users as shown in
Fig. 1. In addition, the exact form derived in [6] has been
obtained by approximating the conditional outage probability
(as a function of the number of users) by a unit step function as
shown in Fig. 2. However, it has not been shown how the
transition value of the step function (Ntr) is determined.
In this paper we improve the accuracy of SINRth_lb by
calculating the probability of the number of users using a
Markov chain model instead of the inaccurate Poisson
distribution assumption used in [6]. Furthermore, we derive an
optimum value of Ntr based on minimizing the mean square
error between the exact and the approximate values.
II. ANALYSIS
Actually SINRth_lb can be determined as the lowest value of
SINRth that keeps Pout below the maximum acceptable outage
probability (Pout_max). Hence, the dependence of Pout on SINRth
is found first. Using the law of total probability, Pout can be
expressed as

SINR dB SINR thdB when the number of existing users is N.
Pout = P out N
P (N ) (1)
N =1 Hence, Pb|N can be expressed as

where Pout|N is the conditional outage probability and P(N) is SINR thdB m (6)
Pb N = 1 Q .
the probability that there are N admitted users per cell. When

imperfect PC is considered, SINR is modeled as a lognormally-
distributed random process. Hence, Pout|N is given by From (1), (2), (5) and (6), it can be shown that Pout is
expressed as a function of SINR thdB as follows
SINR min
dB
m (2)
Pout = 1 Q

N
N
SINRthdB m
N Q

(7)

SINRmin m
dB
.
Pout = 1 Q i =1

where m and are the mean and standard deviation of SINR dB

i i SINRthdB m
respectively. Actually is a constant that depends only on the N =1 Q



i =1 j =1
delay and the PC errors. when PC is feasible, m converges to
dB ). If PC is infeasible, m starts to
the target value ( SINR trg dB can be obtained from (7) by equating P
Thus, SINRth_lb out to
degrade and its value becomes a function of the number of
active user (N) [6]. Hence, m can be given by Pout_max and then solving for SINR th_lb
dB . Since a closed form

expression for SINR th_lb

dB
cannot be obtained, SINR th_lb
dB
has to be
SINR dB
N N max
trg determined by solving (7) numerically.
(3)
m= 1 The authors have determined in [6] a closed form
10 log10 N > N max
( ) ( ) ( ) expression for SINR th_lb
dB
N + f + W S by adopting the following two
1 1 o /
approximations:
where Nmax is the maximum number of admitted users for
feasible power control, o is the noise power spectral
density, S is the targeted received power level, W is the 1) N is modeled as a Gaussian random variable.
spreading bandwidth, and f is the ratio of the inter-cell 2) Pout|N is approximated by a step function (instead of
interference to the intra-cell interference. In fact, Nmax is a the continuously-increasing function given by (2)) as
function of the target SINR (SINRtrg) and it can be shown that it shown in Fig. 2.
is equal to
Using these two approximations, it can be shown that the
closed form expression of SINR th_lb
dB
1 W 1 is given by [6]
N max = o + 1 (4)
S (1 + f )
SINR trg 2
2 Pout _ max 2 Pout _ max


+ 2Ntr + 2Ntr 4Ntr2
1
where x is the largest integer less than or equal to x. In [6] 1 .
SINRth_lb = m +
dB

2
we have determined P(N) by assuming that N follows Poisson
distribution. However, this assumption is found to be inaccurate

since the arrival rate, as shown in Fig. 1, is not constant as it
(8)
depends on the blocking rate (Pb|N), which in turn depends on
the number of the admitted users N. Instead, P(N) is determined
where Ntr is the transition value of the step function as shown in
here using the birth-death model [7] shown in Fig. 1 as follows
Fig. 2, is the inverse Q-function such that
y = Q ( x ) x = ( y ) , and = Pout N< N max . The derivation of
(1 P )
N
N bN
(5) this closed form is provided in [6]. However, it has not been
P (N ) = i =1
,
shown in [6] how Ntr can be chosen. In fact Ntr has been found
(1 P )
i
i
bN in [6] by trail and error. In this paper we use the minimum
i =0 i =1
mean square error (MMSE) as a criterion for finding Ntr. The
MMSE is defined as
where is the average traffic intensity in Erlang per cell, which
is equal to /, where is the average arrival rate while 1/ is
(SINR ) d
max
1 2
the average call residence time. The conditional blocking rate
(Pb|N) is equal to the probability of the event:
min
th_lb SINRth_lb (9)
where SINR th _ lb is the approximate value of the SINRth_lb while
min and max are the minimum and maximum values of the
arriving traffic rate respectively.
III. RESULTS
The system parameters are chosen as follows:
Pout_max = 1%, SINR trg
dB
= -12 dB, SINR min
dB
= -18 dB, oW/S =
0.01, f = 0.3 and = 1 dB, min = 20 Erlang per cell, and max
= 50 Erlang per cell.
Fig. 3 shows SINRth_lb (exact and approximate) versus
arrival traffic intensity in Erlang per cell using the Markov
model, while Fig. 4 shows the normalized mean square error
(difference) between the exact value and the approximate value
versus Ntr.
As shown in Fig. 3, there is a significant difference (>1
dB) between the exact solution using the Markov model and the
exact solution using the Poisson distribution. This proves that
the Poisson distribution assumption used in [6] is not accurate
enough especially at high loading values. In addition, Fig.3
shows the good agreement between the exact and approximate
solutions (both obtained using the Markov model) and it is
evident that the maximum difference is less than 0.5 dB.
As depicted in Fig. 4, the normalized mean square error
between the exact SINRth_lb value (obtained numerically) and
the approximate value obtained from the closed form
approximation is highly dependent of the transition value of the
step function (Ntr). It is clear that the optimum value of Ntr is
[1-Pb|0] [1-Pb|1]
0 1
2 (+1)
Fig. 1. Markov Chain Model Representing the Admission Process.
40, which corresponds to the minimum mean square error
(MMSE).
IV. CONCLUSIONS
In this paper, we have improved the accuracy of the lower
bound of SINR threshold of call admission control in CDMA
networks. We used the birth-death model to represent the
network loading instead of the inaccurate Poisson assumption.
Also, we have shown how to find (using MMSE criterion) the
optimal transition value of the unit step function approximating
the conditional outage probability.
REFERENCES
[1] Z. Liu and M. El Zarki, SIR-based call admission control for DS-CDMA
cellular systems, IEEE Journal on Selected Areas in Commun. (JSAC),
vol. 12, no. 4, May 1994, pp. 638-644.
[2] I.-M. Kim, B.-C. Shin, and D. Lee, SIR-based call admission control by
inter-cell interference prediction for DS-CDMA systems, IEEE
Communications Letters, vol. 4, no. 1, Jan. 2000, pp. 29-31.
[3] A. Viterbi and A. Viterbi, Erlang capacity of a power controlled CDMA
system, IEEE Journal on Selected Areas in Commun. (JSAC), vol. 11,
no. 6, Aug. 1993, pp. 892-900.
[4] S. Ariyavisitakul and L. Chang, Signal and interference statistics of a
CDMA system with feedback power control, IEEE Trans. on Commun.,
vol. 41, Nov. 1993, pp. 1626-1634.
[5] D. Kim, On upper bounds of SIR-based call admission threshold in
power-controlled DS-CDMA mobile systems, IEEE Communications
Letters, vol. 6, no. 1, Jan. 2002, pp. 13 -15.
[6] Mohamed H. Ahmed and Halim Yanikomeroglu SINR Threshold Lower
Bound for SINR-based Call Admission Control in CDMA Networks with
Imperfect Power Control, IEEE Communication Letters, April 2005.
[7] A. Leon-Garcia, Probability and Random Processes for Electrical
Engineering, 2nd ed., Addison-Wesley Inc., 1994 (Section 8.3, pp. 476-
477)
[1-Pb|(N-1)] [1-Pb|N] [1-Pb|(N+1)]
N N+1 ..
(+2)
 0
10

-2
10

-4
10
out | N
P

-6
10

-8
10

Exact
Approximate
-10
10
0 20 Ntr40 60 80 100 120 140
Number of users/cell (N)

Fig. 2. Conditional Outage Probability.

2
-14 10
Exact-Poisson
Approximate-Closed Form
Exact-Markov
-14.5

1
10
-15
Mean Square Error%

-15.5 0
10
SINRth-lb

-16

-1
10
-16.5

-17
-2
10

-17.5

-3
-18 10
20 25 30 35 40 45 50 10 20 30 40 50 60 70 80

Mean Arrival Rate (Erlang/Cell) Ntr

Fig. 3. Lower Bound of SINRth. Fig. 4. Normalized Mean Square Error.