1160

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004

Power-Adaptation Strategies for DS/CDMA

Communications With Successive Interference
Cancellation in Nakagami-Fading Channels
Ye Hoon Lee, Member, IEEE, and Yeheskel Bar-Ness, Fellow, IEEE

AbstractIn this paper, power adaptation for direct-sequence

code-division multiple-access communications that employs a successive interference cancellation (SIC) receiver is considered. The
transmission power is adapted so that, with the channel variations,
the received power levels of each user have appropriate disparities.
Under the constraint of average transmission power, we consider
two strategies in adjusting the disparity between received signal
powers. With the first strategy, the average bit-error rate (BER)
for a given user averaged over channel fading statistics is minimized, while with the other, the instantaneous BER is equal for
all users. We find that the performance difference between the two
strategies becomes negligible as the average transmission power
or line-of-sight component increases. We also discuss the impact of
appropriate disparity in received power levels on the BER performance of SIC receivers.
Index TermsCode-division multiple-access (CDMA), Nakagami-fading channels, power adaptation, successive interference
cancellation.

I. INTRODUCTION

HE RADIO LINK for either a portable or vehicular

unit can be characterized by time-varying multipath
fading, which causes the link quality to vary with time. When
the transmitter and receiver are provided with channel state
information (CSI), the transmission schemes can be adapted
to this information, enabling more efficient use of the channel.
In current code-division multiple-access (CDMA) cellular systems (IS-95), transmission power is adapted so as to maintain
received power from all mobiles at the same desired level [1].
This conventional way of controlling the signal strength at the
base station (BS) is well known to be optimal for the matched
filter receiver [2].
In recent years, there has been much interest in improving
the capacity of direct-sequence CDMA (DS/CDMA) systems
through the use of multiuser detectors [3], [4]. Successive
Manuscript received July 8, 2002; revised July 9, 2003 and February 18, 2004.
This work was supported in part by the New Jersey Center for Wireless Telecommunications (NJCWT) and by the National Science Foundation under Grant
CCR-9903381.
Y. H. Lee was with the with the Center for Communications and Signal
Processing Research, Department of Electrical and Computer Engineering,
New Jersey Institute of Technology, Newark, NJ 07102 USA. He is now with
i-Networking Laboratory, Samsung Advanced Institute of Technology, Suwon
440-100, Korea (e-mail: [email protected]).
Y. Bar-Ness is with the Center for Communications and Signal Processing
Research, Department of Electrical and Computer Engineering, New Jersey
Institute of Technology, Newark, NJ 07102 USA (e-mail: [email protected]).
Digital Object Identifier 10.1109/TVT.2004.830152

interference cancellation (SIC) is a form of multiuser detection

in which the strongest signal is the first detected and cancelled, while the weakest is the last detected [5][8]. While
the matched-filter receiver attains its best performance under
equal received power from all users, the SIC receiver favors
unequal received signal powers. Studies were conducted for
finding the received signal strengths that give, after successive
cancellation, equal bit-error rate (BER) performance for all
users [7][9]. It was particularly shown in [7] that, by ignoring
error propagation, a geometric distribution of received powers
will provide such equal BER. The optimal received powers for
equal BER for nonlinear SIC was obtained in [9] and power
distribution yielding equal BER for linear SIC was found in [8].
A geometric distribution of received powers that minimizes the
average BER, averaged over all active users, was considered in
[10].
We consider power adaptation for DS/CDMA systems that
employ a SIC receiver in Nakagami-fading channels. In this
paper, unlike some of the aforementioned references that
focused on geometric power profile, the transmission power
is adapted relative to channel variations to make the received
powers at BS having an arbitrary power profile. That is, the
received signal powers are distributed with some factor s to
be chosen
and

(1)

is the received power

where is the number of users and
of the user having the th strength, when user strengths are
ranked in the order of estimated channel gains. For users
strength ranking, the correlation of the received signal with
each users spreading sequence was used in [6]; however, we
will exploit separate channel estimates as done in [11]. The
latter approach might be more adequate for next-generation
mobile cellular systems, as separate channel estimation with
transmitted pilot symbols is performed in both wide-band
CDMA (WCDMA) [12] and cdma2000 [13] standards. In this
paper, we assume that the channel can be perfectly estimated at
both the transmitter and receiver.
The factor s in (1) give a measure of the disparity between the received power levels. Clearly, the conventional
power-adaptation method corresponds to a special case of
and the geometric distribution of
,
received powers corresponds to
) is a constant. Under fixed average transwhere (
mission power constraint, we will first consider choosing s

0018-9545/04$20.00 2004 IEEE

LEE AND BAR-NESS: POWER-ADAPTATION STRATEGIES FOR DS/CDMA COMMUNICATIONS

such that the average BER for each user is minimized. We will
also consider choosing s such that after SIC, an equal BER
independent of channel variations for all received signals, is
obtained. Analyzing the performance of these power-adaptation
strategies with average transmission-power constraint, we show
that the proposed adaptation has a significant performance gain
over the conventional method.
This paper is organized as follows. In Section II, we introduce the system model and formulate the signal-to-interference-plus-noise ratio (SINR) of the received signals for a
SIC receiver. The proposed power-adaptation schemes are described and the resulting BER performances are analyzed in
Section III. In Section IV, we present numerical results, all
obtained from the analysis done in Section III, and discuss the
performance improvements provided by proposed adaptation
schemes. Finally, the conclusion is made in Section V.
II. SYSTEM MODEL AND SINR
We consider a DS/CDMA communication system with binary phase-shift-keying (BPSK) signaling. We assume that
the channel variation due to fading is slow relative to the
bit duration and that the channel fading is characterized by
the Nakagami- probability density function (pdf). The Nakagami- fading model fits experimental data from a variety
of fading environments, including urban and indoor multipath
propagation [14]. We look only at a single-cell system for
easier analysis. The implications of a multiple-cell system can
be accounted for by the out-of-cell interference coefficient [1].
We assume that the channel is frequency nonselective. This
assumption is valid for channels with a small delay spread, such
at the BS is
as indoor systems. Then, the received signal
expressed as

1161

Fig. 1. Successive interference cancellation for the ith cancellation stage.

We assume perfect knowledge of the spreading sequences

and CSI of all the users at the receiver. The received signals
are ranked in the order of channel gains. The basic process is
decoding the strongest signal (having the highest channel gain)
and then canceling its effect from the received signal after regeneration. The remaining signals are cancelled in the decreasing
order of channel gains. The th cancellation stage is presented
in Fig. 1, which is repeated until the last decision for the weakest
signal (having the lowest channel gain) is completed.
At the th cancellation stage, a coherent correlator recovering
the signal of the th strongest signal forms a decision statistic
as

(6)
where

(2)
and
where is the transmission power of user and
are the random binary data sequence at a rate of
and the
random signature sequence at a rate of
for user , respecand
are the channel-induced delay and phase, retively.
spectively, of user , which are assumed to be independent and
uniformly distributed, the former over a bit interval and the latter
.
represents the white Gaussian noise with mean
over
.
is the indezero and two-sided power spectral density
pendent and identically distributed (i.i.d.) gamma random variable representing the channel power gain for user and its pdf
is given by [15]

(7)
is the ordered set of the channel gain

, so that
.
and
are the transmission power and data bit of the signal having the
th strength (the th strongest signal), respectively.
is the
cross-correlation between the th strongest signal and the th
strongest one, given by

(3)
(8)
where

is the Nakagami fading parameter

(4)

and

is the gamma function defined as

(5)

In (7), the first term is the multiple-access interference (MAI),

caused by uncancelled user signals; the second term is the
; and the
Gaussian noise with mean zero and variance
third term is the cumulative contribution due to imperfect
. For the
cancellations of the th signals,
detailed derivations of (6) and (7), see [5].

1162

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004

According to [16],
. By using the result
obtained in [8], we can formulate the variance of as

is the cumulative distribution function (cdf) of

where
, given by

(17)
(9)
where

where

is the incomplete gamma function defined as

is the spreading gain defined as

(18)
(10)

Substituting (14) into (13) yields

and we assume that in (7), the first and third terms and
and
, respectively, are independent. Hence, from (6) and (9), as
a function of the channel-fading realization, the instantaneous
for the th strongest signal is given by
SINR

(19)

is adjusted to equal a given

. That is,
and, with (13), this equals
. Hence, together with (19), for the proposed power adaptation (that ensures the received power of the th strongest to
be at given
), the instantaneous SINR for the th strongest
signal is
From (11),

(11)

III. POWER ADAPTATION

The power adaptation considered in this paper ensures the received power of the th strongest users signal to be kept at given
. That is, the transmission power
of the user having the
th strongest channel gain is adjusted to compensate for fading
is equal to
. Since the channel gain
for
such that
user can be any of the th strongest gains, the probability of
being the th strongest among all users is equally probable
(i.e.,
) for all
. Therefore, the average
for user is given by
transmission power

(20)
To calculate the BER, we will use the Gaussian approximation [16], i.e., we assume that the interference plus noise is
Gaussian with mean zero and variance .1 Then, the conditional error probability, given that the user s channel gain is
. Therefore, the average
the th strongest, is given by
(over the channel statistics) BER for every user is given by

(12)
(21)
It follows from (1) that the received power level
strongest user signal is expressed as

of the th
(13)

Note that the average BER of the conventional power adaptation can be obtained by substituting (20) with
into (21).

Then, (12) can be rewritten by

A. Minimal Average BER Power Adaptation
(14)
where

is defined as
(15)

Using order statistics [17], the pdf of

tribution of
as

We now consider the problem of finding the best transmission

power assignment that can minimize the average BER in (21).
This minimization problem can be formulated as
minimize

(22)

is related to the disClearly, the minimization in (22) is over

unknown re, which are constrained
ceived power factors
(16)

1This assumption is commonly made in SIC receivers [6], [8], although it can
give optimistic results if the BER is investigated for low values [18].

LEE AND BAR-NESS: POWER-ADAPTATION STRATEGIES FOR DS/CDMA COMMUNICATIONS

to be positive. To determine the optimal value of

with respect to s
partial derivatives of

1163

s, we take the

(23)
. A set of necessary conditions for
where
an optimal received power profile that minimizes the average
.
BER can be obtained by setting (23) to zero for
We use a multidimensional Newton method [19] to find optimal s in our work. We will call this power-adaptation policy,
which employs an optimal power profile in adjusting the transmission power of each user, the minimal BER power adaptation.
B. Equal BER Power Adaptation
In some applications, it is required to maintain the link quality
as a fixed value regardless of channel variations. This corresponds to making the SINR at each cancellation stage the same
value for all received signals, i.e.,

Fig. 2.

S T =N

Received power distribution for SIC: K

= 10 dB; and
= 1.

= 7; N = 32; m = 2;

C. No Power Adaptation
(24)
where is the desired fixed SINR. It is well known that the received power levels should be geometrically distributed in order
to obtain equal SINR for all signals after SIC [8]. That is, for
equal SINR, in (13) can be given by

For the purpose of performance comparisons, we consider, in

this section, the SIC scheme without power adaptation. When
CSI is not available at the transmitter, the transmission power is
(i.e.,
for every and ). In this
fixed at
case, we have
(29)

(25)

where we also used the definition in (15). Substituting (29) in

(11), we have the instantaneous SINR of the th strongest signal
with such a nonadaptive scheme

and corresponding instantaneous SINR is

(30)
where
(26)
The power ratio required to obtain the same SINR for all received signals is given by [8]
(27)
We will call the power-adaptation strategy that employs the geometric power profile with as equal BER power adaptation.
(or ) can be obtained by substituting (27) in (26) toThe
gether with (24) and solving with a numerical technique. The
(or ).2 Since the equal
secant method [19] is used to find
BER power adaptation ensures the same SINR for all received
signals, the (average) BER is given by
(28)
calculating (or r ), we can use any of 3 ; i 2 f1; 2; . . . ; K g in (24),
because all 3 s yield the same (or r ). Also, with various attempts made by
authors, a solution was found.
2In

(31)
in (30) is a function of
(fluctuates with channel
Since
variations), the conditional BER expression, given that the
user s channel gain is the th strongest, is then given by
. Thus, the average BER for user is given
by
(32)

IV. NUMERICAL RESULTS AND DISCUSSIONS

Fig. 2 shows a received power distribution for several adaptation schemes. We note that with the minimal and the equal BER

1164

Fig. 3. BER versus user number i: K

10 dB; and
= 1.

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004

= 7;

= 32;

= 2;

T =N

power-adaptation schemes, we have decreasing received power

levels for users in order of their cancellation. We also note that
between the earlier and later cancelled signals, the power difference with minimal BER power adaptation is greater than that
of equal BER power adaptation. This indicates that, in order to
minimize the average BER, higher received power is required to
achieve reliable symbol decision at earlier cancellation stages.
Fig. 3 depicts a BER distribution for several schemes. This
shows that, with the minimal BER power-adaptation strategy,
the BER of the earlier cancelled signal is lower than that of the
later cancelled signal. That is, such a strategy makes the symbol
decision at earlier cancellation stages more reliable than later
cancellations, so the error propagation is protected and the resulting average BER can be minimized. In fact, with the minimal BER power-adaptation strategy, the average BER for each
user is the same even though the instantaneous BER is different;
hence, it is appropriate for nonreal-time applications where the
use of forward error correction with a large interleaving span
having a large number of fades are allowed. On the other hand,
the no-power-adaptation strategy, which has a natural disparities
between received power levels, results in a relatively high BER
at later cancellation stages, hence dominantly degrading the average BER. Finally, the conventional power-adaptation strategy,
which maintains the received signal powers for all users at the
same level, fails to yield a reliable symbol decision at earlier
cancellations, leading to significant error propagation to later
cancellation stages and hence results in the degradation of average BER.
The average BERs for the different adaptation schemes
are compared in Fig. 4. It shows that the conventional power
adaptation cannot yield significant performance improvements
over the nonadaptive scheme, while the other two power-adaptation strategies (minimal and equal BER power adaptation)
have significant power gains over the conventional-power or

Fig. 4.

Average BER versus

= 1.

T =N

= 7;

= 32;

= 2; and

no-power adaptation. This indicates that appropriate disparities

in received power levels incurred by transmission power adaptation can be utilized effectively when using SIC. The power
gain, which translates into a power reduction at the transmitter,
results in reduced interference to other users and leads to a capacity increase. We note that the equal BER power-adaptation
strategy yields comparable performance to the minimal BER
power adaptation and that the performance difference becomes
increases. This indicates that, at
negligibly small as
), keeping
an interference-limited region (for higher
equal SINR after SIC for all received signals is the way to
attain the minimum average BER.
Fig. 5 depicts the average BER versus the number of users
. At an average BER of
, the maximum number of
users supportable with the equal BER power adaptation and the
minimal BER power adaptation increase by factors of 14/9 and
16/9, respectively, over the conventional power adaptation. This
capacity gain translates into a reduction of transmitter power
required to support a target system loading, which in turn means
a prolongation of the battery life at a mobile handset.
Fig. 6 depicts the average BER as a function of Nakagami-fading parameter ,3 which shows that, for higher
(with stronger line-of-sight component), the performance
difference between the minimal BER and the equal BER power
adaptation becomes negligible. It also is interesting to note that,
for some range of system parameters, the no-power-adaptation
strategy exhibits better BER performance than the conventional
power adaptation. With a weaker line-of-sight component
(for smaller ), transmitted signals undergo relatively severe
3When m = 1 (i.e., Rayleigh fading), some weaker received signals after
ordering have worse channel statistics than Rayleigh, which results in zero received power level in (19), since (k; 1) goes to infinity for some k . Thus,
the (channel inverting) power adaptation considered in this paper cannot be employed in Rayleigh channels.

E 0

LEE AND BAR-NESS: POWER-ADAPTATION STRATEGIES FOR DS/CDMA COMMUNICATIONS

1165

the conventional power-adaptation strategy, since the channel

gains for all users asymptotically go to unity.
V. CONCLUSION
We considered power adaptation for SIC in DS/CDMA systems, where transmission power is adjusted to creat appropriate
disparities in the received signal powers. In this paper, we considered two adaptation strategies. With the first, the transmission
power is adapted such that the average BER is minimized,
while with the other, equal BER after SIC for all received
signals is obtained. The two adaptation strategies were shown
to have significant power gains over the conventional power
adaptation. It was also found that the performance difference
between the two strategies becomes negligible as
or
increase (better channel conditions). The power gain
provided by controlling disparities between received powers
translates into a capacity increase and a prolonged battery life
at the transmitter.

Fig. 5. Average BER versus K : S T =N

= 1.

= 10 dB; N = 32; m = 2; and

ACKNOWLEDGMENT
The authors would like to thank the anonymous reviewers and
the Edior for their valuable comments.
REFERENCES

Fig. 6. Average BER versus m: S T =N

= 1.

= 10 dB; N = 32; K = 7; and

channel fading, which results in higher disparity between

received power levels, leading to much weaker signal powers
for later cancellation stages; hence, later cancelled signals
dominantly degrade the average BER. On the other hand, as
the line-of-sight component increases, the signal fluctuation
over the fading channel is reduced and the natural disparities
incurred by a no power-adaptation scheme can be exploited
more effectively in SIC, yielding better performance than the
conventional power-adaptation scheme. In the limiting case
where approaches infinity (no fading), the no-power-adaptation scheme has exactly the same received power distribution as

[1] K. S. Gilhousen et al., On the capacity of a cellular CDMA system,

IEEE Trans. Veh. Technol., vol. 40, pp. 303312, May 1991.
[2] R. Cameron and B. Woerner, Performance analysis of CDMA with imperfect power control, IEEE Trans. Commun., vol. 44, pp. 777781,
July 1996.
[3] S. Moshavi, Multi-user detection for DS-CDMA communications,
IEEE Commun. Mag., vol. 34, pp. 124136, Oct. 1996.
[4] S. Verdu, Multiuser Detection. Cambridge, U.K.: Cambridge Univ.
Press, 1998.
[5] P. R. Patel and J. M. Holtzman, Analysis of a DS/CDMA successive
interference cancellation scheme using correlations, in Proc. IEEE
GLOBECOM93, Nov. 1993, pp. 7680.
[6]
, Analysis of a simple successive interference cancellation scheme
in a DS/CDMA system, IEEE J. Select. Areas Commun., vol. 12, pp.
796807, June 1994.
[7] A. J. Viterbi, Very low rate convolutional codes for maximum theoretical performance of spread-spectrum multiple-access channels, IEEE
J. Select. Areas Commun., vol. 8, pp. 641649, May 1990.
[8] R. M. Buehrer, Equal BER performance in linear successive interference cancellation for CDMA systems, IEEE Trans. Commun., vol. 49,
pp. 12501258, July 2001.
[9] G. Mazzini, Equal BER with successive interference cancellation
DS-CDMA systems on AWGN and Ricean channels, in Proc. IEEE
PIMRC95, Sept. 1995, pp. 727731.
[10] J. H. Kim and S. W. Kim, Combined power control and successive interference cancellation in DS/CDMA communications, in Proc. IEEE
WPMC02, Oct. 2002, pp. 931935.
[11] P. Dent, B. Gudmundson, and M. Ewerbring, CDMA-IC: A novel code
division multiple access scheme based on interference cancellation, in
Proc. IEEE PIMRC92, Oct. 1992, pp. D4.1.1D4.1.5.
[12] H. Holma and A. Toskala, WCDMA for UMTS. New York: Wiley,
2000.
[13] V. K. Garg, IS-95 CDMA and CDMA2000. Englewood Cliffs, NJ:
Prentice-Hall, 2000.
[14] H. Suzuki, A statistical model for urban multipath propagation, IEEE
Trans. Commun., vol. COM-25, pp. 673680, July 1977.
[15] M. Nakagami, The m-distribution A general formula of intensity distribution of rapid fading, in Statistical Methods in Radio Wave Propagation. New York: Pergamon, 1960, pp. 336.
[16] M. B. Pursley, Performance evaluation for phase-coded spread spectrum multiple access communicationPart I: System analysis, IEEE
Trans. Commun., vol. COM-25, pp. 795799, Aug. 1977.
[17] H. A. David, Order Statistics, 2nd ed. New York: Wiley, 1981.

1166

[18] R. K. Morrow and J. S. Lehnert, Bit-to-bit error dependence in slotted

DS/SSMA packet system with random signature sequences, IEEE
Trans. Commun., vol. 37, pp. 10521061, Oct. 1989.
[19] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flanner, Numerical Recipes in C, 2nd ed. Cambridge, U.K.: Cambridge Univ. Press,
1997.

Ye Hoon Lee (M02) received the B.S., M.S., and

Ph.D. degrees in electrical engineering from Korea
Advanced Institute of Science and Technology
(KAIST), Daejeon, Korea, in 1990, 1992, and 2000,
respectively.
From 1994 to 1997, he was an Invited Member
of Research Staff at the Electronics and Telecommunications Research Institute (ETRI). In 1998,
he was a Visiting Research Engineer with SK
Telecom, Daejeon, Korea. From June 2000 to July
2001, he was a Senior Research Engineer with LG
Electronics, Anyang, Korea. From August 2001 to December 2002, he was with
the Department of Electrical and Computer Engineering, New Jersey Institute
of Technology, Newark, NJ, as a Postdoctoral Researcher. Since 2003, he has
been with the i-Networking Lab, Samsung Advanced Institute of Technology,
Suwon, Korea, as a Senior Member of Technical Staff, where he works on
transceiver design for next-generation wireless mobile and ultrawide-band
(UWB) communications. His research interests include spread-spectrum
communications, adaptive resource allocation for wireless mobile communications, communication theory, and adaptive modulation/coding and diversity
techniques for wireless communications.

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004

Yeheskel Bar-Ness (M69SM79F89) received

the B.Sc and M.Sc degrees in electrical engineering
from TechnionIsrael Institute of Technology,
Haifa, Israel, and the Ph.D. degree in applied
mathematics from Brown University, Providence,
RI.
After working in the private sector, he joined
the School of Engineering, Tel-Aviv University,
Tel-Aviv, Israel, in 1973. From September 1978
to September 1979, he was a Visiting Professor at
Brown University, (on leave from the University
of Pennsylvania, Philadelphia, and Drexel University, Philadelphia). He
joined the New Jersey Institute of Technology, Newark, NJ, from AT&T
Bell Laboratories in 1985, where he currently is a Distinguished Professor of
Electrical and Computer Engineering and Foundation Chair of Communication
and Signal Processing Research. He also is the Executive Director of the Center
for Communication and Signal Processing Research (CCSPR). His current
research interests include the design of OFDM and MC-CDMA, adaptive
processing for UWB, adaptive array, and spatial interference cancellation
and signal separation for multiuser personal, indoor, and mobile wireless
communications. He has published numerous papers in these areas.