Impact of Measurement Errors on the Closed Loop Power Control for CDMA Systems

Li-Chun Wang and Chih-Wen Chang National Chiao Tung University, Taiwan Email : lichun@cc.nctu.edu.tw

AbstractIn this paper we present a simple analytical model to evaluate the impact of measurement errors on the closedloop power control for code division multiple access (CDMA) systems. By introducing a new performance measurement, the probability of false command in power control, PF C , a closed form formula for calculating PF C with consideration of measurement errors is presented. Furthermore, we drive a bit error rate (BER) performance bound in terms of PF C for the CDMA system with the closed-loop power control (CLPC). The proposed analytical approach can quantitatively evaluate the performance of the CLPC taking into account of the effects of measurement errors and Doppler frequency under a Rayleigh fading channel. Through simulation and analysis, we show that the proposed analytical BER bound can accurately estimate the BER performance of the CLPC under the impact of measurement errors. Interestingly, we nd that the CLPC is less sensitive to measurement errors due to the non-linear operation in the one step up/down power control scheme compared with the variable-step size power control. Index TermsCDMA System, Power Control, Measurement Error.

I. I NTRODUCTION C curate power control is one of key technologies to achieve high capacity code division multiple access (CDMA) systems. Power control errors may be resulted from many factors [1], such as loop delay [2], quantization errors [3], multi-path fading [4], [5], link-quality measurement errors [6], and feedback errors [7]. Although the closed-loop power control (CLPC) in the CDMA system has been studied extensively in the literature [3], [6], [8], [9], [10], fewer papers have analyzed the performance of power control scheme subject to signal-to-interference ratio (SIR) measurement errors. Previous works about the impact of measurement errors on power control can be summarized as follows. The impact of measurement errors on the open-loop power control was studied in [6]. The authors in [1] discussed the ltering effect in the measurement scheme for the WCDMA sysThis work was supported jointly by the Lee and MIT center for networking research, and the National Science Council R.O.C under the contract 90-2213-E-009-068, 89-E-FA06-2-4, and EX-91-E-FA06-4-4.

tems. In [10], [11], [12], the issue of joint minimization of SIR measurement errors and power control errors are investigated in the form of a stochastic control problem, but the SIR measurement errors in [10], [11], [12] are modeled as white Gausssian noise. However, in [13], it has been found that measurement errors tend to be log-normal distributed in cellular channel with Rayleigh fading and shadowing. The goal of this paper is to develop a simple and accurate analytical model to evaluate the impact of log-normal distributed SIR measurement errors on the CLPC of CDMA systems. The contributions of this work can be summarized in two folds. First, to evaluate the impact of measurement errors on the closed-loop power control, we introduce a new performance measurement called the probability of false power control command, PF C . The motivation of introducing the new parameter, PF C , is because the feedback power control command is the key to the accuracy of power control scheme. We will present a close-form formula for calculating PF C in terms of measurement errors. Second, we propose a simple BER bound in terms of PF C . Hence, by using the proposed analytical approach, the performance for different SIR measurement schemes on the closed-loop power control in the CDMA system can be easily obtained. The rest of this paper is organized as follows. In Section II, we briey introduce power control and dene the probability PF C . Section III derives the closed-form formula of PF C in terms of measurement errors. In Section IV, we derive a simple BER bound for the CLPC in CDMA systems with the parameter PF C . Section V shows some analytical and simulation results. Section VI gives concluding remarks. II. P ROBABILITY OF FALSE P OWER C ONTROL C OMMAND A. Background Generally speaking, we can categorize power control models into two kinds. The rst one is the open-loop power control and the other one is the closed-loop power control. In

0-7803-7701-X/03/$17.00 (C) 2003 IEEE

the former scheme, a mobile terminal decides its own transmission power by comparing the desired received SIR and the target SIR to compensate path loss and shadowing. On the other hand, the CLPC makes the base station determine the up/down power control command by comparing the received SIR with the target SIR. The block diagram of the CLPC scheme is shown in Fig.1. As shown in this gure, the CLPC has two feedback loops where the inner loop is for fast power adjustment, while the outer loop is for setting the target Eb /No . B. Probability of False Power Control Command Because the accuracy of power control commands will strongly inuence the performance of the closed loop power control, it is important to investigate the impact of measurement errors on the false power control command. Figure 2 illustrates how measurement errors will inuence the power control command. For example, as shown in Fig. 2, if the actual SIR value is below the target SIR value, the CLPC will issue a power up command to increase the transmission power. For a small measurement error, the power control command will still be correct since the measured SIR is below the the target SIR as shown in the left hand side of the gure. However, if a large measurement error causes the received SIR to become larger than the target SIR, then power control command may change to be a wrong power down command, as shown in the right hand side of the gure. In order to evaluate the effect of measurement errors on the closed-loop power control scheme, we dene the probability of false command (PF C ) in power control as follows. PF C = P rob{ sgn (SIRT SIRM ) = sgn (SIRT SIRA )}

Received Signal Channel Match Filter Rake Combiner Channel Decoder CRC Detection

T Eb /No Measurement

+
Power Control Step Size Target Eb/No Setting

Fig. 1. The block diagram of the closed-loop power control scheme.

can be expressed as the summation of SIRT and power control error, X , i.e. SIRA = SIRT + X . (2)

The measured SIR, denoted as SIRM , can be expressed as the summation of SIRA and the measurement error, Y , i.e, SIRM = SIRA + Y . (3)

From (2) and (3), we know (1) SIRM = SIRT + X + Y . (4)

where sgn(x) is the operator to choose the sign of x, SIRM is the measured SIR, SIRA is the actual SIR value, and the SIRT is the target SIR. From this denition, one can see that the measurement error makes the sign of (1) change, thereby issuing a false command in power control. III. A NALYSIS In this section, we will derive an analytical formula to evaluate the probability of false command in power control. To begin with, we model the power control error (PCE) and measurement errors as independent log-normal distributed random variable as [9], [13]. Denote the SIR measurement error by a random variable, Y, and the power control error by a random variable, X. In the following, we will consider all the variables in the dB domain. Denote SIRT and SIRA as the target SIR and the actual SIR, respectively. Then SIRA

Substituting (2) and (4) into (1), we nd that a false power control command occurs when the following condition is sustained: sgn(SIRT (SIRT + X + Y )) sgn(SIRT (SIRT + X )) or sgn(X + Y ) = sgn(X ) . (6) = (5)

Hence, the probability of false power control command (PF C ) can be written as PF C = P (X + Y < 0|X > 0)P (X > 0) + P (X + Y > 0|X < 0)P (X < 0) (7)

= P (X + Y < 0, X > 0) + P (X + Y > 0, X < 0) .

0-7803-7701-X/03/$17.00 (C) 2003 IEEE

Measured Eb/No < Target Eb/No =>Issue a Power up command

Measured Eb/No > Target Eb/No => Issue a Power down command
Measured Eb/No

Let x = x/X and y = y/Y . Then we can express PF C as PF C = 2

0
X Y

1 2 1 exp( x ) 2 2 (11)

Target Eb/No Measured Eb/No Measurement error PCE Actual Eb/No Measurement error

exp(

1 2 y )dy dx . 2

Target Eb/No PCE Actual Eb/No

Next, we change the coordinate axis to polar coordinate axis by letting x = r cos and y = r sin . Then PF C can be written as PF C = = 2
/2
tan1 X Y

1 1 r 2 e 2 rdrd 2 (12)

No False Command

However, actual Eb/No < Target Eb/No. We need to increase power. False Command occurs.

1 2

X tan1 Y

Fig. 2. Occurrence of false command in power control.

Now we want to analyze the impact of PF C on the BER performance. From [14], we know that for the received bit energy to noise density ratio b , the bit error rate of BPSK modulated signals in Rayleigh fading channel is P b1 = 1 2 1 b 1 + b . (13)

Note that in the dB domain both X and Y are Gaussian random variables with zero means and standard deviations of X and Y , respectively. Because P (X + Y < 0, X > 0) = P (X + Y > 0, X < 0), we can simplify PF C as PF C = = = 2P (X + Y < 0, X > 0) 2P (Y < X, X > 0) 2
x 0

In the AWGN channel, the BER of BPSK modulated signals is Pb2 = Q( where 1 Q(x) = 2
x

(8)

2b )

(14)

e 2 dx .

t2

(15)

fXY (x, y )dydx ,

(9)

where fXY (x, y ) is the joint probability density function of random variables X and Y . Because measurement errors are majorly inuenced by the measurement scheme rather than power control method, we assume the measurement error X is independent of power control error Y . Then, we have PF C = 2
0 x 2

Since the purpose of power control is aimed to remove Rayleigh fading, we expect that the BER performance of the CLPC will be closed to (14) if the power control perfectly removes the impact of Ralyeigh fading; and on the other hand, the BER performance of the CLPC will be close to (13) if there is no power control. Consequently, we propose an upper bound on the BER performance for the closed-loop power control subject to measurement errors as follows. Pb,F C 2PF C Pb1 + (1 2PF C )Pb2 PF C 1 b 1 + b 2b ) , + (16)

1 x2 exp ( 2 ) 2X Y 2X

exp ( = 2
0

y 2 )dydx 2Y
x

(1 2PF C )Q(

y 2 exp ( 2 )dydx . 2Y

1 x2 exp ( 2 ) 2X Y 2X (10)

where Pb1 and Pb2 are dened in (13) and (14). Note that because the maximum of PF C is 0.5, we use 2PF C to approximate the case without power control. We will evaluate the accuracy of the proposed upper bound (16) via simulation in the next section.

0-7803-7701-X/03/$17.00 (C) 2003 IEEE

TABLE I
S IMULATION PARAMETERS .

0.5

0.45

Spreading Factor Doppler Frequency Power Control Period Power Control Step Size Target Eb/No Modulation Scheme
0.12

4 5-30Hz 0.667mSec 1 dB 5 dB BPSK

fd=5 Hz
0.4

0.35

f =15 Hz d
FC

0.3

0.25

fd=30 Hz
0.2

Theory
0.15

Simulation

0.1 PCE (Simulation) LogNormal Fitting 0.08

0.1 1 2 3

(STD. of measurement error in dB)

Y

10

Probility

0.06

Theoretical P =0.301
FC

Fig. 4. The probability of false power control command ,PF C , subject to measurement errors for different Doppler frequencies, where fd = 30, 15, and 5 Hz, and target Eb /No = 5 dB.

Simulative PFC=0.309
0.04

0.02

0 30

20

10

10

20

30
2

40

PCE in dB ( The measurement error is modeled by LogN(0,5 ) )

Fig. 3. The power control error statistics with a measurement error of standard deviation equal to 5 dB, i.e. Y =5 dB, with fd = 20 Hz, and target Eb /No = 5 dB.

IV. N UMERICAL R ESULTS In this section, we simulate the uplink performance for a single user under a at Rayleigh fading channel. We assume the measurement errors are log-normal distributed random variables [13]. Other parameters used in the simulation are listed in Table I. Through simulation, Figure 3 shows the probability density function of power control errors (PCE) subject to measurement errors with standard deviation of 5 dB. As shown in the gure, the PCE (expressed in dB domain) exhibits the similar characteristics of normal distribution. This result conrms our assumption that the PCE is lognormal distributed random variable even with the inuence of measurement errors. Figure 4 shows the impact of measurement errors for different Doppler frequencies on the the probability of false power control command, PF C . One can note that the higher the Doppler frequency, the lower the PF C . This phenomenon can be explained as follows. Because the radio channel will change faster at a higher Doppler frequency, the power control mechanism usually cannot follow fast channel variations. Thus, in this situation a larger power con-

trol error will occur. Due to a larger gap between the target SIR and the actual SIR, it is less likely to change the power control command from power up to power down or vice versa. In other words, a larger PCE due to higher Doppler frequency can tolerate large measurement errors, thereby having a smaller PF C as shown in the gure. In Figure 4, we also verify the accuracy of the analytical results of PF C derived from (12) by comparing with simulation. As shown, simulation results of PF C are close to the analytical results derived from (12). Figure 5 shows the corresponding BER performance of Fig. 4 including the impact of PF C . Although in Fig. 4 it is demonstrated that a higher Doppler frequency leads to a smaller PF C , the BER performance at a higher Doppler frequency is still larger than that at a lower Doppler frequency as shown in Fig. 5. It is implied that even with a larger probability of false power control command, the closed-loop power control scheme under a slower fading channel still performs better than that under a fast fading channel. Figure 6 shows the impact of very large measurement errors on the probability of false command PF C in the closedloop power control. One can see that as a measurement error increases up to 30 dB, the probability of false power control command tends to be saturated at 0.45. Note that in (16), the maximum of PF C approaches to 0.5 for the innite measurement error. Figure 7 illustrates the BER performance of the closedloop power control subject to very large measurement errors. As shown in the gure, the theoretical values according to formula (16) provide a reasonable bound on the BER performance of the closed-loop power control subject to measurement errors. Because PF C will be bounded to be 0.5 even with large measurement errors, one can nd that the

0-7803-7701-X/03/$17.00 (C) 2003 IEEE

10

0.45

0.4

0.35

0.3

0.25

BER

10

FC

0.2

0.15

0.1

fd=5Hz fd=15Hz fd=30Hz

10
3

0.05

Simulation Theory
0 5 10 15 20 25 30

(STD. of measurement error in dB)

Y

10

Y (STD. of Measurement error in dB)

Fig. 5. The BER performance comparison of the the closed-loop power control subject to measurement errors for different Doppler frequencies, where fd = 30, 15, and 5 Hz, and target Eb /No = 5 dB.

Fig. 6. Probability of false power control command, PF C , with a large standard deviation of measurement error, Y , fd = 30 Hz, and target Eb /No = 5 dB.
1

10

BER of the up-down closed-loop power control will also be bounded under a large measurement error. In the gure, with the target Eb /No equal to 10 dB, the BER is limited to below 0.02 even with a measurement error with standard deviation of 30 dB. Thus, it is implied that the up/down power control scheme is robust to measurement errors. Figure 8 compares the up/down closed-loop power control with the variable-step size closed-loop power control from the standpoint of the sensitivity of power control errors with respect to measurement errors. In our simulation, the variable-step size power control is a power control mechanism with higher resolution in power adaptation steps. As shown, unlike the variable-step power control has a strong correlation between measurement errors and power control errors, the impact of a larger measurement error on the up/down closed-loop power control is less sensitive. V. C ONCLUSIONS In this paper we have presented a simple and accurate analytical expression for the probability of false command (PF C ) in the closed-loop power control (CLPC) subject to measurement errors. Moreover, we also propose a BER performance bound in terms of PF C . It is found that the larger the Doppler frequency, the smaller the PF C , while for the BER performance, a larger Doppler frequency results in a poor BER performance even with a smaller PF C . Interestingly, we nd that compared with the variable-step power control, the up/down CLPC is more robust to measurement errors due to the non-linear operation in the up/down power control scheme. The proposed analytical approach can be easily applied to analyze the performance of the closed-loop power control with different SIR measurement schemes.

Target Eb/N0=5 dB

10

BER

Target Eb/N0=10 dB

Upper bound Simulation

10
3

10

Y (STD. of measurement error in dB)

15

20

25

30

Fig. 7. BER performance of the closed-loop power control against measurement errors with stand-deviation, Y , and fd = 30 Hz.

Thus, an interesting future research extended from this work is to apply the proposed analytical framework to incorporate different SIR measurement schemes to evaluate the performance of the closed-loop power control in CDMA systems. R EFERENCES
[1] F. Gunnarsson, Fundamental limitations of power control in WCDMA , IEEE Vehicular Technology Conference, Vol. 12, pp. 630-634, Fall 2001. [2] F. Gunnarsson, F. Gustafsson, adn J. Blom, Dynamical effects of time delays and time delay compesation in power controlled DSCDMA, IEEE Jouranl on Selected Areas in Communications, Vol. 19, No. 1, pp. 141-151, Jan. 2001. [3] A. Abrardo, G. Giambene, D. Sennati, Optimization of Power Control Parameters for DS-CDMA Cellular Systems, IEEE Trans. on Commmunications, Vol. 49, No. 8, pp. 1415-1424, Aug. 2001.

0-7803-7701-X/03/$17.00 (C) 2003 IEEE

12

10

X (STD. of PCE in dB)

fd=10Hz fd=20Hz fd=30Hz fd=20Hz(Variantstep size)

Y (STD. of Measurement error in dB)

10

Fig. 8. Comparison of the up/down closed-loop power control and the variable step closed-loop power control in terms of the correlation of power control errors and measurement errors for

[4] N. Kong, L. B. Milstein, Error probability of multicell CDMA over frequency selective fading channels with power control error, IEEE Trans. on Communications, Vol. 47, No. 4, pp. 608-617, Apr. 1999. [5] J. M. Romero-Jerez, M. Ruiz-Garcia, and A. Diaz-Estrella, Effects of multipath fading on BER Statistics in cellular CDMA networks with fast power control, IEEE Communications Letters, Vol. 4, No. 11, pp. 349-351, Nov. 2000. [6] Y. W. Leung, Power control in cellular networks subject to measurement error, IEEE Trans. on Communications, Vol. 44, No. 7, pp. 772-775, July 1996. [7] A. Abrardo, G. Giambene, and D. Sennati, Performance analysis of SIR-based closed-loop power control with feedback errors, IEICE Transn. Commun., Vol. E85-B, No. 5, pp. 872 - 881, May 2002. [8] L. Song, N. Y. B. Mandayam, and Z. Gajic, Analysis of an up/down power control algorithm for the CDMA Reverse link under fading, IEEE Journal of Seletive Area on Communications, Vol. 19, No. 2, pp. 277-286, Feb. 2001. [9] Andrea Abrardo, D. Sennati, On the Analytical Evaluation of Closed-Loop Power-Control Error Statistics in DS-CDMA Cellular Systems, IEEE Trans. on Vehicular Technology, Vol. 49, No. 6, pp. 2071-2080, Nov. 2000. [10] L. Qian and Z. Gajic, Joint Optimization of Mobiles Transmission Power and SIR Error in CDMA Systems, Proceeding of American Control Conference, Vol. 5, pp. 3767-3772, May 2001. [11] L. Qian and Z. Gajic, Variance minimization stochastic power control in CDMA systems, IEEE Internation Conference on Communications, pp. 1763-1767, May 2002. [12] S. Ulukus and R. D. Yates, Stochastic power control for cellular radio systems, IEEE Trans. on Commun., Vol. 46, No. 6, pp. 784798, Jun. 1998. [13] A. J, Goldsmith, L. J. Greenstein, G. J. Foschini, Error Statistics of real-time power measurements in cellular channels with multipath and shadowing, IEEE Trans. on. Vehicular Technology , Vol. 43, No. 3, pp. 439-446, Aug. 1994. [14] John G.Proakis, Ditital Communications, Fourth Edition, pp. 726794, New York: McGraw-Hill, 2001.

0-7803-7701-X/03/$17.00 (C) 2003 IEEE