- Research Article
- Open access
- Published:
Iterative Soft Decision Interference Cancellation for DS-CDMA Employing the Distribution of Interference
EURASIP Journal on Wireless Communications and Networking volume 2010, Article number: 214393 (2010)
Abstract
A well-known receiver strategy for direct-sequence code-division multiple-access (DS-CDMA) transmission is iterative soft decision interference cancellation. For calculation of soft estimates used for cancellation, the distribution of residual interference is commonly assumed to be Gaussian. In this paper, we analyze matched filter-based iterative soft decision interference cancellation (MF ISDIC) when utilizing an approximation of the actual probability density function (pdf) of residual interference. In addition, a hybrid scheme is proposed, which reduces computational complexity by considering the strongest residual interferers according to their pdf while the Gaussian assumption is applied to the weak residual interferers. It turns out that the bit error ratio decreases already noticeably when only a small number of residual interferers is regarded according to their pdf. For the considered DS-CDMA transmission the bit error ratio decreases by 80% for high signal-to-noise ratios when modeling all residual interferers but the strongest three to be Gaussian distributed.
1. Introduction
The demands on the data rates provided by future mobile communications systems are further increasing especially for the downlink. Higher data rates in the downlink require for example, higher-order modulation schemes and more efficient receiver algorithms to overcome intersymbol interference (ISI) and, additionally for direct-sequence code-division multiple access (DS-CDMA) systems, multiple access interference (MAI).
In this paper we consider the downlink of a DS-CDMA system. Although the optimum solution for the receiver of a user terminal is known [1], its application is prohibitively complex, because the computational effort increases exponentially with the number of users. Therefore, when applying more powerful receiver algorithms than the standard Rake receiver [2], one has to consider suboptimum schemes.
A promising approach was presented in [3], where successive interference cancellation is proposed. This scheme was refined in for example, [4–7] by utilizing soft decisions for cancellation. For calculation of soft decisions, the distribution of residual interference is commonly assumed to be Gaussian. This assumption is accurate according to the central limit theorem [8], when a successive interference cancellation algorithm starts and the number of noteworthy residual interferers is high. But the Gaussian model gets inaccurate as the algorithm converges and the number of relevant residual interferers decreases.
A general discussion of the accuracy of the Gaussian assumption can be found in [9, 10]. In the noniterative approach [11] the distribution of interference is not approximated as Gaussian but modeled via uniform triangular densities which, however, leads to an only minor performance gain. The benefit of employing the actual probability density function (pdf) is for example, recognized in [12, 13] where the distribution of interference is approximated by kernel smoothing, modeling the pdf of interference by a Gaussian mixture density (sum of Gaussian densities) which is then adjusted according to the occurrence of interference. Approximating the pdf of interference with a Gaussian mixture is also proposed in [14], where the approximation is fixed for the entire transmission.
In contrast to these approaches we derive an approximation of the pdf of interference based on probabilities of interfering symbols calculated in the receiver. Our approach uses the matched filter-based iterative soft decision interference cancellation (MF ISDIC) algorithm which was introduced in [6] and extended in [15]. In [6], the algorithm is derived for synchronous DS-CDMA systems with random spreading sequences and binary phase-shift keying (BPSK) symbols considering transmission over an additive white Gaussian noise (AWGN) channel. In [15], MF ISDIC is designed for quadrature amplitude modulation (QAM) transmission over multipath channels. Extensions to the ISDIC algorithm for linear modulation are proposed in [16].
The paper is organized as follows. First, we introduce the system model in Section 2. We review the MF ISDIC receiver for transmission over multipath channels with general square QAM constellations using the Gaussian assumption according to [15] in Section 3.1 and extend it for use of an approximation of the actual pdf of residual interference in Section 3.2. In Section 3.3, a graphical comparison of the pdf calculated via the Gaussian assumption and the approximation of the actual interference distribution is given. A hybrid scheme is derived in Section 3.4. In Section 4, performance of the introduced algorithms is compared by means of simulations.
2. System Model
In the following, all signals and systems are represented by their complex-valued baseband equivalents. A DS-CDMA transmission with general square QAM constellations and Gray mapping is considered. The discrete-time transmit signal 
at chip time 
of a base station is expressed as
where 
is the number of served users, 
is the number of transmitted QAM symbols per user during the considered time interval, and 
is the spreading factor which is assumed to be identical for all users. 
denotes the spreading sequence of user 
at symbol time 
which is nonzero for 
. 
is the transmitted QAM coefficient of user 
at symbol time 
which is taken from a square QAM set of cardinality 
. The average power of the transmitted coefficients 
(
: expectation) is denoted by 
. The coefficients 
of the QAM set are assumed to be equiprobable and the values 
and 
are taken from the set 
of cardinality 
. If a user is served with several spreading sequences simultaneously we denote this as multicode transmission. Obviously, this case is also covered by the system model because one user may comprise several virtual users which are served with one spreading sequence each.
The discrete-time received signal is given by
where 
denotes the causal discrete-time channel impulse response of order 
including the effects of transmit filtering, channel, and continuous-time receiver input filtering. 
is additive complex white Gaussian noise with variance 
.
For the ISDIC algorithm we use a matrix vector notation for simplicity, which is introduced in the following. First, we define a convolution matrix 
of size 
whose entries in the 
th row and 
th column are 
. The received signal vector r
(
: transposition) can be expressed as
where 
which may also be written as 
with
and 
. The matrix 
contains the accordingly stacked spreading sequences and 
is abbreviated by 
.
Furthermore, a truncated version of the model according to (3) is used to derive the sliding window filters of MF ISDIC. For the truncated system model we consider a time interval of the received signal in vector 
which exactly matches the influence of a transmitted symbol 
, 
, respectively, according to (4),
with 
and 
and 
denote the according parts of the matrix 
and the vector a, respectively. As 
we denote the index of the entry 
in the vector 
. Therefore, the 
th column of matrix 
is equivalent to the effective spreading sequence of the symbol 
. The indices 
and 
lead to the first symbol 
and the last symbol 
, respectively, in vector a which has any influence on the time interval covered by the effective spreading sequence of the symbol 
. Note that this system model also comprises transmission with linear modulation for one user (
) employing spreading factor 
. Hence, the algorithm analyzed in the following can also be utilized as a receiver for transmission with linear modulation over dispersive channels.
3. Matched Filter-Based Iterative Soft Decision Interference Cancellation (MF ISDIC)
In each iteration of MF ISDIC, soft-decision feedback is performed for cancellation of ISI and MAI in a sequential manner starting from symbol index 
up to 
. 
is the index of the current iteration and 
, cf. (4), denotes the soft decision on 
calculated in iteration 
.
The soft decisions 
are initialized according to 
for 
. For derivation of MF ISDIC, we introduce the vector 
, which is used for calculation of the soft estimate 
.
In each iteration 
and for each 
the following steps have to be done. In order to obtain an estimate for the desired coefficient 
, ISI and MAI caused by other coefficients are removed from the vector 
in the best possible way using the latest soft estimates in vector 
The resulting vector 
contains significantly less interference compared to 
when the soft estimates in 
get better from iteration to iteration.
Subsequently, the vector 
is filtered with a vector 
(
: Hermitian transposition) acting as a matched filter adjusted to the effective spreading sequence of symbol 
Here, 
is the 
th column of the matrix 
which is equal to the effective spreading sequence of symbol 
and 
is the energy of the effective spreading sequence of symbol 
. The output of the matched filter is, compare with (6) and (7),
where we have used
is an abbreviation for residual ISI and MAI and noise.
After the current iteration has been finished with processing of the last data symbol 
, a new iteration starts. The algorithm stops if soft decisions remain essentially unchanged from one iteration to the next, that is,
with a small constant 
, or the iteration number exceeds a prescribed limit 
.
3.1. MF ISDIC with Gaussian Assumption
For MF ISDIC with Gaussian assumption, we model the residual ISI and MAI and noise 
as a random variable with a complex Gaussian pdf [8] with zero mean and variance 
in real and imaginary part and zero correlation coefficient between them. The power 
of 
can be calculated via conditioned expectations of the latest matched filter outputs for refinement of estimation as
where we have used
which minimizes the mean-squared error 
, cf. [8, 17]. 
denotes complex conjugate of a complex number. The vector 
is defined as 
. The initialization of 
is done according to 
. With the Gaussian assumption we can evaluate the expectation values in (12). 
is valid, where 
and 
denote the real part and the imaginary part of a complex number, respectively. The first term can be calculated to
Similarly we get
Finally, using the mentioned assumptions, the soft estimate 
of (13) can be derived as
The calculation of soft estimates for general complex-valued symbol alphabets according to (16) is also given in [7]. When assuming symbols of a general phase-shift keying (PSK) or cross QAM constellation for transmission instead of a square QAM constellation, only (14)-(15) have to be modified.
For 4QAM transmission with 
the expectation values in (12) simplify to 
and (16) reads
which was also utilized in [18].
For the Gaussian assumption, the pdf 
of residual ISI and MAI and noise is
with 
according to (12).
3.2. MF ISDIC Employing an Approximation of the Actual Interference Distribution
Assuming the residual interference to be Gaussian distributed is well justified according to the central limit theorem [8], when a successive interference cancellation algorithm starts and the number of relevant residual interferers is high. However, the assumption gets inaccurate as the algorithm converges in course of the iterations and the number of relevant residual interferers decreases. In this subsection, we modify MF ISDIC for employment of an approximation of the actual pdf of residual interference which has to be derived first.
Hence, the pdf 
of residual ISI and MAI and noise 
in (9) which was assumed to be a complex Gaussian pdf with zero mean and power 
in Section 3.1 is now calculated approximately, resulting in a more accurate expression. For this, we first derive the pdf of a term 
in the first sum in (8), keeping in mind that the pdf 
is the convolution of several such pdfs and a normal pdf. We assume that 
and 
are already available and calculate the conditional probability of all symbols 
given 
using Bayes' theorem [8]
Here, 
denotes the probability that 
equals 
.
For derivation of the pdf 
we have to define the Dirac pulse in the complex plane for a complex variable 
in dependence of the Dirac pulse 
for real numbers
Because the transmitted symbols are equally probable, any transmitted symbol 
has the pdf
The conditional pdf 
can be expressed as
where (19) can be used for 
. From (22), the conditional expectation value for the symbol 
is obtained
The subtraction of the corresponding expectation value in (8) leads to a modified symbol
whose pdf is given by
Obviously, the pdf 
has zero mean due to the subtraction of the expectation value in (24). Weighting 
with a factor 
according to (8) yields 
. The pdf of 
is, cf. [8],
Notethat the squared value 
in (26) is founded by the fact that the considered random variables are complex-valued, cf. for example, (20). Therefore, the pdf 
of residual ISI and MAI and noise 
in (9) is the convolution (symbol "
") of a complex Gaussian pdf and pdfs according to (27) (the convolution of two pdfs of complex-valued random variables, 
and 
with 
, is defined as 
)
where 
denotes (
). Here we have used, that the power of the noise is 
as can be seen from the last term in (12) and assumed independence of random variables corresponding to the pdfs to keep mathematical tractability. The latter assumption is an approximation which is fulfilled perfectly only for the a priori pdfs of different symbols but not for their a posteriori pdfs conditioned on the computed soft symbols. With help of the pdf 
a new soft estimate 
according to (23) can be calculated utilizing (19).
3.3. Comparison of the Gaussian Model with the Approximation of the Actual Interference Distribution
To visualize the difference between the pdf with Gaussian assumption, cf. (18), and the approximation of the actual distribution of interference according to (29), some graphs are given in the following. We assume that no prior knowledge is available, that is, MF ISDIC starts with all prior estimates being zero, 
and 
. The crosscorrelation values according to (10) are selected to 
and 4QAM (
) is used.
Figure 1 gives the pdf of interference and noise with Gaussian assumption, cf. (18), and in Figure 2 the approximation of the actual pdf of interference and noise, cf. (29), is shown. For both figures, the assumed signal-to-noise ratio (SNR) is 
dB. Here, 
is the average received energy per bit and 
stands for the single-sided power spectral density of the Gaussian channel noise. Obviously, the Gaussian assumption is not justified for the considered SNR although both pdfs have the same variance. However, the approximation of the actual pdf gets closer to a Gaussian pdf for lower SNR as can be seen from Figures 3 and 4 for 
dB. The Gaussian assumption is also justified for a higher number of interfering symbols because in this case the number of convolved pdfs according to (28) increases. However, in MF ISDIC the number of residually interfering symbols is supposed to be low with increasing number of iterations 
.
Pdf of ISI, MAI, and noise with Gaussian assumption for
dB.
Approximation of actual pdf of ISI, MAI, and noise for
dB.
Pdf of ISI, MAI, and noise with Gaussian assumption for
dB.
Approximation of actual pdf of ISI, MAI, and noise for
dB.
3.4. Hybrid MF ISDIC
Calculation of 
according to (29) is very complex. To reduce complexity we propose a hybrid scheme, where the approximation of the pdf is simplified by considering only the strongest 
residual interferers according to their pdf and applying the Gaussian assumption to the weak residual interferers. Obviously, for the hybrid scheme, the number of terms in (29) decreases and 
additionally includes the power of the weak residual interferers. The strongest interferer is defined as the one that has the highest value 
.
First, the power 
of each pdf according to (27) in (28) has to be calculated for each 
It is sufficient to approximate the pdfs 
which have the weakest power with Gaussian pdfs. As all of these pdfs have zero mean, the Gaussian approximation can be performed in (28) by substituting the corresponding pdf 
by a Dirac pulse 
and incrementing 
by 
. The number of pdfs 
which are taken into account according to their pdf is denoted with 
.
The resulting pdf 
for hybrid MF ISDIC consists of a sum of Gaussian densities and therefore has the same form like the pdfs in [12–14]. But in contrast to these approaches the resulting pdf of interference is derived based on probabilities of interfering symbols calculated in the receiver. Thus, a better approximation is expected.
Figure 5 shows the hybrid pdf approximation of the pdf in Figure 2. Here, the strongest 
interfering symbols have been taken into account. It gets apparent that the complex pdf of Figure 2 is approximated quite well by that of Figure 5. The hybrid pdf approximation of the pdf in Figure 4 is shown in Figure 6. Again, the strongest 
interfering symbols have been taken into account. In this case the complex pdf of Figure 4 is approximated almost perfectly by that of Figure 6. In both cases the scenario of Section 3.3 has been considered.
Hybrid pdf approximation of ISI, MAI, and noise for 
dB with 
.
Hybrid pdf approximation of ISI, MAI, and noise for 
dB with 
.
To judge the accuracy of the hybrid scheme in comparison to the scheme with the Gaussian assumption, an analysis according to the Kullback Leibler distance (KLD) [19] is conducted in the following. The KLD is a measure for the similarity of two pdfs, where a value of 
corresponds to a perfect match. First, 
dB is considered. The KLD of the approximation of the actual pdf in Figure 2 and the corresponding pdf with Gaussian assumption in Figure 1 is 
. In contrast, the KLD of the approximation of the actual pdf in Figure 2 and the corresponding pdf with hybrid approximation in Figure 5 is 
. Obviously, the approximation of the actual pdf is matched more accurately by the hybrid approach than by the Gaussian assumption. For 
dB, the following observations can be made. The KLD of the approximation of the actual pdf in Figure 4 and the corresponding pdf with Gaussian assumption in Figure 3 is 
. In contrast, the KLD of the approximation of the actual pdf in Figure 4 and the corresponding pdf with hybrid approximation in Figure 6 is 
. Here, the approximation of the actual pdf is matched almost perfectly by the hybrid scheme—much better than with the Gaussian assumption.
4. Numerical Results and Discussion
MF ISDIC employing the common Gaussian assumption and hybrid MF ISDIC as introduced in Sections 3.1 and 3.4, respectively, are compared in the following by means of simulations. The results are shown in Figure 7. We consider a Rayleigh multipath channel consisting of ten chip-spaced paths with decreasing average powers according to 
, 
, 
, 
, 
, 
, 
, 
, 
, and 
.
BER versus 
for MF ISDIC employing the Gaussian assumption and hybrid MF ISDIC.
, 
, 4QAM with 
, uncoded DS-CDMA downlink transmission with 
spreading sequences having spreading factor 
over a Rayleigh multipath channel, ideal power control.
This power delay profile is an approximation of the power delay profile of the vehicular 
test channel [20] for chip-spaced paths. An ideal power control algorithm is assumed resulting in a normalization of the sum of all instantaneous tap powers to one. Uncoded multicode transmission is applied using 
spreading sequences with spreading factor 
, representing the whole load of the base station. The channel is constant during the transmission of one block, that is, a block fading channel model is used. For simulations, 
has been chosen to 
, the number of iterations maximally tolerated to 
, and 
(
for 4QAM). Note that this scenario is related to a high speed downlink packet access (HSDPA) transmission with UMTS.
In Figure 7 simulation results for MF ISDIC employing the common Gaussian assumption and hybrid MF ISDIC are shown for 4QAM transmission. For hybrid MF ISDIC, all but the 
strongest residual interferers are modeled to be Gaussian distributed. Obviously, the bit error ratio (BER) can be lowered by 80 for high 
by applying the proposed hybrid MF ISDIC, assuming that all residual interferers but the strongest three are Gaussian distributed.
5. Concluding Remarks
In this paper, a receiver algorithm performing matched filter-based iterative soft decision interference cancellation (ISDIC) which was proposed recently has been analyzed for a DS-CDMA downlink transmission over multipath channels when employing general square QAM constellations. The commonly used Gaussian assumption for the pdf of residual interference has been replaced by a better approximation of the exact pdf. Additionally, a hybrid scheme has been proposed, which provides less computational complexity by considering only the strongest residual interferers according to their pdf while the Gaussian assumption is applied to the weak residual interferers. The algorithms have been compared by means of simulations for an UMTS scenario, and it has been shown, that the bit error ratio decreases already noticeably when only a small number of residual interferers is processed according to their pdf. In fact, for the considered DS-CDMA transmission we have been able to lower the bit error ratio by 80
for high signal-to-noise ratios when modeling all residual interferers but the strongest three to be Gaussian distributed.
References
Verdú S: Minimum probability of error for asynchronous Gaussian multiple-access channels. IEEE Transactions on Information Theory 1986, 32(1):85-96. 10.1109/TIT.1986.1057121
Proakis J: Digital Communications. 3rd edition. McGraw–Hill, New York, NY, USA; 1995.
Patel P, Holtzman J: Analysis of a simple successive interference cancellation scheme in a DS/CDMA system. IEEE Journal on Selected Areas in Communications 1994, 12(5):796-807. 10.1109/49.298053
Frey T, Reinhardt M: Signal estimation for interference cancellation and decision feedback equalization. Proceedings of Vehicular Technology Conference (VTC '97), May 1997, Phoenix, Ariz, USA 155-159.
Divsalar D, Simon MK, Raphaeli D: Improved parallel interference cancellation for CDMA. IEEE Transactions on Communications 1998, 46(2):258-268. 10.1109/26.659484
Müller R, Huber J: Iterated soft–decision interference cancellation for CDMA. In Broadband Wireless Communications. Edited by: Luise M, Pupolin S. Springer, London, UK; 1998:110-115.
Sgraja C, Teich W, Engelhart A, Lindner J: Multiuser/multisubchannel detection based on recurrent neural network structures for linear modulation schemes with general complex–valued symbol alphabet. COST Workshop 262, Technical Report ITUU-TR-2001/01, January 2001, Ulm, Germany, 45-52.
Papoulis A: Probability, Random Variables, and Stochastic Processes. 3rd edition. McGraw–Hill, New York, NY, USA; 1991.
van Rooyen P, Solms F: Maximum entropy investigation of the inter user interference distribution in a DS/SSMA system. Proceedings of the 6th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC '95), September 1995, Toronto, Canada 1308-1312.
Poor H, Verdú S: Probability of error in MMSE multiuser detection. IEEE Transactions on Information Theory 1997, 43(3):858-871. 10.1109/18.568697
Mailaender L: Non-linear detectors for multiuser CDMA exploiting non-Gaussianity. Proceedings of the 31st Asilomar Conference on Signals, Systems & Computers, November 1997, Pacific Grove, Calif, USA 1410-1414.
Li Y, Li KH: Iterative PDF estimation and turbo-decoding scheme for DS-CDMA systems with non-Gaussian global noise. Proceedings of the IEEE Global Telecommunication Conference (Globecom '01), November 2001, San Antonio, Tex, USA 3262-3266.
Luschi C, Mulgrew B: Nonparametric trellis equalization in the presence of non-Gaussian interference. IEEE Transactions on Communications 2003, 51(2):229-239. 10.1109/TCOMM.2003.809249
Zhang Y, Blum RS: Iterative multiuser detection for turbo-coded synchronous CDMA in Gaussian and non-Gaussian impulsive noise. IEEE Transactions on Communications 2001, 49(3):397-400. 10.1109/26.911444
Rößler J, Huber J: Iterative soft decision interference cancellation receivers for DS–CDMA downlink employing 4QAM and 16QAM. Proceedings of the Asilomar Conference on Signals, Systems and Computers, 2002, Pacific Grove, Calif, USA
Rößler JF, Gerstacker WH, Huber JB: Iterative equalization with soft feedback with a subsequent stage employing error search and correction. IEEE Transactions on Vehicular Technology 2008, 57(1):335-344.
Tarköy F: MMSE-optimal feedback and its applications. Proceedings of the IEEE International Symposium on Information Theory (ISIT '95), September 1995, Whistler, Canada 334.
Rößler J, Gerstacker W, Lampe A, Huber J: Matched–filter– and MMSE–based iterative equalization with soft feedback for QPSK transmission. Proceedings of the Zurich Seminar, February 2002, Zurich, Switzerland 191-196.
Cover T, Thomas J: Elements of Information Theory. John Wiley & Sons, New York, NY, USA; 1991.
T.-R. 3GPP, TR 101 112 : Selection procedures for the choice of radio transmission technologies of the UMTS (UMTS 30.03 version 3.2.0). V3.2.0, April 1998
Acknowledgment
This work has been presented in part at the Vehicular Technology Conference (VTC Spring 2003), Jeju, Korea.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Cite this article
Rößler, J., Gerstacker, W. & Huber, J. Iterative Soft Decision Interference Cancellation for DS-CDMA Employing the Distribution of Interference. J Wireless Com Network 2010, 214393 (2010). https://doi.org/10.1155/2010/214393
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1155/2010/214393