IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO.

4, OCTOBER 2002

781

Mitigating Multiple Access Interference and Intersymbol Interference in Uncoded CDMA Systems With Chip-Level Interleaving
Ravishankar H. Mahadevappa and John G. Proakis, Life Fellow
AbstractThe presence of intersymbol interference (ISI), in addition to multiple access interference, severely hampers the performance of a code-division multiple-access (CDMA) communication system. In such a situation, channel coding can be used to obtain better performance, but at the cost of a reduction in rate of flow of information. In this paper, it is shown with the help of simulation results that the chip-Interleaved CDMA system effectively combats ISI without requiring additional channel coding. The system differs from the conventional CDMA system in the sense that, the chip sequence resulting from the pseudo noise (PN) sequence modulation is interleaved before transmission. Two receivers are proposed, one based on the turbo equalization method which employs a Maximum a posteriori equalizer of exponential complexity and the other based on minimum-mean square erroroptimized iterative interference cancellation principles which is of linear complexity. Simulation results are provided which show that error rates close to the no-ISI single-user case can be obtained. The shortcomings of the coded CDMA with turbo detection system in the presence of ISI are also discussed. Index TermsIntersymbol interference (ISI), multiple access interference (MAI) algorithm, multiuser detection (MUD), serially concatenated systems, turbo detection.

I. INTRODUCTION IRECT-sequence spread-spectrum code-division multiple-access (DS-SS-CDMA) systems have found application in several communication scenarios, prominent among them being wireless communications. In such a scenario, several users share a given bandwidth and time period to communicate with each other, using pseudo noise (PN) sequences assigned to them. The low cross-correlation values of the PN sequences enable them to communicate fairly reliably even in the presence of multiple access interference (MAI) due to the other users, using a simple conventional matched filter at the receiver. Further, in the presence of strong MAI, more complicated multiuser detectors (MUDs) can be used to improve performance. However, in a multipath fading channel which leads to significant intersymbol interference (ISI), the error rates are often much worse compared with the case where there is no ISI.

Manuscript received July 25, 20001; revised March 21, 2001 and November 16, 2001; accepted November 16, 2001. The editor coordinating the review of this paper and approving it for publication is L. Hanzo. The authors are with the Department of Electrical and Computer Engineering, Northeastern University, Boston, MA 02115 USA (e-mail: [email protected]; [email protected]). Digital Object Identifier 10.1109/TWC.2002.804163

The turbo equalization method, presented by Douillard et al. [2] and also studied by Bauch et al. [3], is shown to be effective in recovering the loss due to ISI for a binary communication system. At the transmitter, the data is first coded using a convolutional encoder, then interleaved and transmitted. The multipath channel is seen as a second encoder. The two encoders and the interleaver together form a serially concatenated system. In the detector, as in turbo decoders [1], information is exchanged between two soft-inputsoft-output (SISO) maximum a posteriori (MAP) decoders and decisions are made after several iterations. It has been demonstrated that error rates close to that of the no-ISI case can be obtained with such a system. The turbo decoding idea has been extended to coded CDMA systems by Wang and Poor [5], Alexander et al. [6], and Valenti [8], among others. In coded CDMA, the data of each user is convolutionally encoded, interleaved, and modulated with a PN sequence before transmission. The combination of PN sequence modulation and the multipath channel is considered as the second encoder. It is claimed in [5] that the system is able to mitigate the detrimental effects of both MAI and ISI. However, this is achieved by normalizing the composite sequence obtained by convolving the PN sequence with the channel impulse response, before transmission. This is not always possible, since, in general, the channel coefficients are not known at the transmitter. In addition to that, different receivers see different channels and it is not possible to normalize the composite sequence as per the requirements of all channels, at the transmitter. On the other hand, the chip-Interleaved CDMA (cI-CDMA) system does not require this normalization and yet is effective in mitigating the MAI and ISI effects. The cI-CDMA system differs from the conventional CDMA system in the sense that the chip sequence resulting from the PN sequence modulation is interleaved before transmission. Chiplevel interleaving was introduced in the early 80s by Tachikawa and Marubayashi, [19] (and references therein), as a means of mitigating burst impulsive noise disturbances. Further, in [20], its effectiveness in the presence of wideband on-off jamming and tone jamming, has been investigated. Frenger et al. in [21], have compared performance of cI-CDMA with random signature sequences with low rate code-spread CDMA in single path and two-path fading channels. In this paper, the effectiveness of chip-level interleaving coupled with iterative (turbo) detection based receivers, in combatting ISI and MAI, is studied. The PN sequence modulation is seen as the outer encoder and no additional encoding is required. The multipath channel is seen as the second encoder. A receiver

1536-1276/02$17.00 2002 IEEE

782

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

Fig. 1.

Block diagram of a CDMA system with different multipath channels.

is designed along the lines of the turbo equalization method, involving two MAP decoders working iteratively on the a posteriori probability (APP) outputs of each other. Simulation results are presented which show that the system performs well in the presence of ISI and error rates close to that of the no-ISI case are obtained. The interleaved CDMA system proposed in users [9] considered the case where the channels for all the were identical. Further, the interleavers were also assumed to be identical. Such a model could possibly be used to represent the downlink scenario. However, in general, the channels for the different users could be different. Here such a general scenario is considered. The interleavers for the users are also assumed to be different. The turbo equalization method based receiver has computational complexity which is exponential in the number of users and lengths of the channels. It is, therefore, not practical when there is a large number of users in the system. An alternative low-complexity receiver based on minimum-mean square error (MMSE)-optimized iterative interference cancellation principle is presented here. The detector is a modified version of the MMSE-optimized iterated soft-decision interference canceller by Mller and Huber [17]. It has been modified to work at the chip level and to cancel the ISI in addition to MAI. There are also some differences in the way the soft estimates are obtained. Simulation results are provided which show that the low-complexity receiver too performs well, yielding error rates close to the no-ISI, single-user error rates. The paper is organized as follows. Section II introduces the notation used and illustrates the effect of ISI on a CDMA system with an example. Section III discusses the shortcomings of the coded CDMA system in more detail. The cI-CDMA system transmitter and the turbo-equalization based receiver structures are presented in Section IV. The MMSE-optimized chip-level interference canceller (cIC) based receiver is presented in Section V. Conclusions and possible future extensions are pointed out in Section VI. II. CONVENTIONAL CDMA SYSTEM AND ISI In this section, it is shown that the conventional CDMA system is prone to severe loss in performance in the presence of ISI, with the help of an example. The optimal MAP decoder is used and it is assumed that the channel coefficients are

known at the receiver, so that there is no loss due to suboptimal detection or errors in channel estimation. The block diagram of a -user DS-SS-CDMA system with associated multipath channels is shown in Fig. 1. The data is assumed to be transmitted in blocks of length . Each multipath channel is represented by a tapped delay line filter with delays equal to the chip period. The received signal is passed through a chip-matched filter, the output of which is sampled at chip rate to yield samples

where

(1) is a white Gaussian noise sample. is an arbitrary delay and for each user, which represents the bit-asynchronous nature of , asthe system. It takes values from the set suming the transmission is chip synchronous. It must be noted that although chip synchronization may be harder to achieve in practice, it is a closer approximation to an asynchronous system compared with the bit-synchronous assumption and simplifies analysis to a large extent. represents the PN sequence of user . The sequences are assumed to be of equal length and are , normalized. They are binary valued, i.e., . are the coeffiwhere paths in the multipath channel of the cients of the th user. is the th user energy. represents the th bit transmitted by the th user and takes values from the bi. is the unique value which satisfies nary alphabet for a given . Due to the bit-asynchronous nature of the transmission and bits of the th user also the multipath channel, the past contribute to the transmitted signal in a given bit period. Here (2) represents the smallest integer greater than or equal where to . An example is shown in Fig. 2. An optimal MAP receiver which does joint equalization and multiuser detection can be designed to detect the data.

MAHADEVAPPA AND PROAKIS: MITIGATING MAI AND INTERSYMBOL INTERFERENCE IN UNCODED CDMA SYSTEMS

783

Fig. 2. Grouping of chip-matched filter output samples at the receiver. 2;  = 0;  = 3; N = 7; L (1) = L (2) = 1 and L = L = 1.

K=

Vasudevan and Varanasi [10], Fawer et al. [11] and [12] have considered the problem of multipath fading channels and presented optimal and suboptimal receiver structures to solve the problem. Channel estimation methods and their effects on decoding are studied in [11] and [12]. However, since the purpose of this article is to compare the performance of the conventional CDMA system with the cI-CDMA system in the presence of ISI, the channel coefficients are assumed known and an optimal receiver is designed for the system. A. MAP Receiver for CDMA Multipath The MAP detector for the combined CDMA and multipath which maximizes the APP channel finds the value of , for each and . An optimal decoding algorithm to compute these probabilities has been presented by Bahl et al. [4]. A modified version of it, which is more suitable for recursive calculations, is applied to decode convolutional codes in [1]. In this paper, the version in [1] is used with appropriate modifications. are grouped into The chip-matched filter output samples for ease of implementation. The grouping vectors of length is done with reference to the desired user which is assumed to be user 1 in the rest of the paper. is set to zero. Then the samples are grouped as

Fig. 3. Trellis diagram for a K = 2;  = 0;  = 3; N = 7 system with L (1) = L (2) = 1 and L = L = 4 channel.

Using such a trellis, the APP can be written as

(4) where and , and are defined as follows:

Note that the definition of is slightly different from the definition in [1], in the sense that there is a factor in the denominator. This makes it easier to write the recursion equations for and . The forward and backward probabilities and are calculated recursively using (5) (6) The transition probabilities using are computed

for . Fig. 2 shows an example of the grouping operation. The optimal algorithm requires a trellis with states and transition paths starting at each state. Fig. 3 shows . Each the trellis for a two-user case with denote the state at the ellipse depicts a state. Let the variable takes values of the form end of the th bit period.

(7) (3) Since each bit can take two different values can take different values. In Fig. 3, since takes different combinations of . The transitions from each state correspond to possible combinations of the -users transmitted bits in the th bit period. (8) is the th eleIn (7), is the white Gaussian noise variance, and is an estimate of the transmitted value in the ment in and the conditional probability

the

784

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

same chip period. The estimate is computed from the informa. The probabilities tion specified by the state pair take value 0 if there is no path between and 1 if there are also computed is one in the trellis. The probabilities for each of the pairs based on the trellis. They take values which are connected in the trellis, since there are paths from and the paths are equiprobable. The recursion equations can also be expressed in matrix form, of order and transiby defining column vectors of order for each stage in the trellis. tion matrix is the cardinality of the state space. The transition mahas elements . Equations trix (4)(6) can be given in matrix form as (9) (10)
Fig. 4. Comparison of error rates obtained for two-user CDMA system with f [0:227 0:460 0:688 0:460 0:227]. multipath channel coefficients f SIR is fixed at 3 dB. Optimal MAP detector is used.

(11) has only those elements which correspond to input where , . It should be noted that the dendepends on the density of the trellis. If sity of the matrices the matrices are sparse, it would be computationally efficient to use the original expressions (4)(6) for implementation. The log APP ratios are computed using (12) bits in a block of bits are set to The last for each user. This ensures that the trellis terminates at the state. To summarize, the MAP Algorithm can be given by these steps: , for 1) Initialize and for other . , compute and 2) For each observation using (5), (7), and (8). using (6) for all and . 3) Compute for all and using (4) and (12). 4) Compute The values are passed through a sign detector to detect the unknown data. B. Simulation Results A CDMA system with two-users transmitting asynchronously on a multipath channel given by , for , was simuwere used. Gold sequences of lated. Blocks of length length 7, used in [5], were used for modulation. The sequences assigned to user 1 and user 2 were (13) Fig. 4 shows the error rates obtained. They are compared with the error rates obtained when the ISI is removed. It can be seen that there is a loss of about 4 dB in signal-to-noise ratio (SNR). This is due to the fact that the composite sequence obtained by

= =

convolving the PN sequence with the channel has about 3.5-dB lower power than normal. It is shown in Section IV that the cI-CDMA system is successful in recovering almost the entire loss, without additional coding. The coded CDMA system has been suggested in [5] as a method to mitigate ISI and MAI. The discussion in the next section shows that the coded CDMA system as presented in [5], has a practical limitation. III. CODED CDMA/TURBO DETECTION RECEIVER Turbo detection methods have been applied to coded CDMA systems in [5], [6], [8], and others. The general approach is that the data of each user is passed through a convolutional encoder, then interleaved and modulated with PN sequences and transmitted, as shown in Fig. 5. Here, the PN sequence modulation and the channel are together considered as a second encoder and Turbo Decoding principle is applied. Valenti [8] shows that error rates close to single-user error rates can be obtained with coded CDMA systems in the presence of Rayleigh fading. However, there is considerable loss compared with the no-fading case. Wang and Poor [5] have studied the behavior of the coded CDMA system with a multipath channel and have also proposed less complex SISO detectors as a substitute for the highly computationally intensive combined MAP equalizer and MUD. A four-user case with signature sequences of length 7 and a complex three-path channel is used to show the effectiveness of the detector. It is argued that the detrimental effects of MAI and ISI can be almost completely overcome by iterative processing with the system in Fig. 5. However, this is achieved by normalizing the composite sequence before transmission, which may not be possible to achieve in practice. , Consider the channel in (13), used in and [3], [7], and the PN sequence given by which is normalized to have unit energy. The composite sequence obtained after convolution with the channel, is shown , in (14), at the bottom of the next page, has 3.5 dB lower than that of the original sequence. Therefore, this

MAHADEVAPPA AND PROAKIS: MITIGATING MAI AND INTERSYMBOL INTERFERENCE IN UNCODED CDMA SYSTEMS

785

Fig. 5. Coded CDMA system.

would result in a deterioration in performance. In the convolution process, the overlapping parts of the PN sequence might add constructively too, depending on the values of the PN sequence and the channel coefficients. This would result in an improvement in performance. However, if path gains are modified such that the composite sequence has unit norm, as is done in [5], then there is practically no loss. This requires knowledge of the channel coefficients at the transmitter, which is not always available. In addition to that, there may be multiple receivers which see different channels and it is not possible to normalize the composite sequence as per the requirement of each channel at the transmitter. In the absence of such a normalization operation, coded CDMA systems, as shown in Fig. 5, are not effective in mitigating ISI. In fact, the 4-dB loss seen in the previous section remains. The loss can also be explained intuitively by the observation that the interleaver in Fig. 5 is not effective in dispersing the data information throughout the block, since it operates at the bit level. In other words, the redundancy provided by the spreading sequence protects the correct bits and the errors resulting from the channel equally so that the outer decoder cannot effectively recover the loss due to ISI. It may be theoretically possible to find particular PN sequences which result in constructive overlapping for a given ISI channel. However, if the channel is time-varying, then the transmitter will have to adapt to the varying conditions and change the PN sequences appropriately. Further, the transmitter requires feedback from the receiver in order to realize the gain, which then would require more bandwidth and lead to other associated expenses. On the other hand, by introducing chip-level interleaving, the chips are

redistributed in the block so that not all combinations of chips result in destructive overlapping. In the next section, it is seen that this recovers most of the loss due to ISI. IV. cI-CDMA A block diagram of the cI-CDMA system is shown in Fig. 6. The data is transmitted in blocks of length as before, by each of the -users. The transmission is bit-asynchronous and chipsynchronous. Each users data is modulated by its respective PN . This stream sequence, resulting in a chip stream of length is interleaved and then transmitted. Here the channel alone is considered as the inner encoder, as by Douillard et al. [2] and by Bauch et al. [3]. The PN sequence modulation is seen as the outer encoder. Although for each user the modulation appears as a simple repetition code, the -user combined modulation is actually an block code with nonbinary output. The optimal receiver for the system is exponentially complex, not only in terms of the number of users, but also in terms of the length of the frame. If one has to which result in minimum error, one has find bit sequences possible sequences. This complexity is due to to consider the interleaving operation. The problem is alleviated by using the turbo equalization method presented in [2], where the data estimation is done iteratively. Although the method is suboptimal, it has been shown that, for large interleaver lengths the performance is very good. A receiver can be built along the lines of the turbo equalization method, with a MAP equalizer and single-user detectors, as shown in Fig. 6. The symbol detector in [2] is replaced with a MAP equalizer and the channel decoder is replaced with single-user detectors.

(14)

786

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

Fig. 6. Chip-Interleaved CDMA system.

The received chip-matched filter output samples can be written in terms of the chip streams transmitted by the users, as (15) is the interleaved version of the chip stream generwhere ated by the modulation of the th users bit stream with its PN sequence, i.e.,

for

(16)

is the th interleaver and is of length . As in (1), is . Therefore, each such that takes binary values in , where element . The interleaving operation distributes the chips of a bit in the entire frame, so that there is no destructive overlapping of chips from a bit, as seen in Section III. Using different interleavers for different users ensures that the interfering chips in a particular bit period are independent over . The case where same interleaver is used for all - users is studied in detail, in [9]. The differences in the structure of the receivers in the two cases are pointed out in Section IV-C. At the receiver, shown in Fig. 6, the equalizer computes the , given a priori information on the log(APP) of the chips and the received sequence . Extrinsic indistribution of formation from the equalizer is computed by subtracting the a priori information from the output a posteriori information. This information is deinterleaved and fed to the single-user detectors as a priori values. The th single-user detector (SUD) then computes the log(APP) and log APP ratio estimates for , by correlating the extrinsic information of the chips in the th bit period with the binary PN sequence of the th user. are obtained Extrinsic information values for the chips

by subtracting the a priori information fed to the SUD from the log(APP) estimate of the corresponding bit. This information is then interleaved and fed to the equalizer as a priori values for the second iteration. After a sufficient number of iterations, the unknown data bits are decided based on the log APP ratios. The operations are discussed in detail in the next two sections. Some quantities are defined to facilitate the explanation ( and unless noted otherwise). Log of APP of event , estimated by equalizer. Log of APP of event , estimated by th SUD. Log APP ratio estimates for , for . a priori value for , fed to equalizer. extrinsic info. for , computed by equalizer. a priori value for , fed to SUD. extrinsic info. for , computed by SUD. denote the deinterleaving operation. Then the folLet lowing relations hold: (17) (18) The computation of sections. A. MAP Equalizer The equalizer computes . is given by given and a priori information and is explained in the next two

(19)

MAHADEVAPPA AND PROAKIS: MITIGATING MAI AND INTERSYMBOL INTERFERENCE IN UNCODED CDMA SYSTEMS

787

is the length of the received sequence and is equal to . These quantities are computed using a trellis which has states, with transitions from each state. Transitions in the trellis occur at the chip rate. Each state is chips of all the users, that is, the represented by the past vector

where

The s, s, s, and s are computed for all indices using (21), (22)(23), and the recursion equations (5)(6) with apand are propriate modifications as mentioned earlier. are normalized across the computed using (18). The outputs , for all , and columns, so that fed as a priori information to the SUDs. B. SISO SUD and log APP ratio estimates APP estimates computed by the SUD using the a priori information tained from the equalizer. is given by are ob-

(20) The transitions from each state corfor respond to the different combinations of . It is assumed that the trellis starts and terstate. This can be ensured in minates at the number of s before and after practice by appending transmission of the sequence . The equations pertaining to the computation of (19) are similar to (4), (5), and (6), with transition probabilities computed appropriately. Equation (4) is reproduced here with suitable modifications to the variables involved

for Here,

(24)

denotes the probability of the event given the a priori information . The probability can be written as a joint probability of the chips in that bit period. Let be the chip sequence of the th user, corresponding to the th period before interleaving. Then

(21) where and , and are defined as

Ideally, the joint distribution can be found from the MAP equalizer by considering paths which satisfy the condition . However, the process is complicated by the presence of the interleaver due to which the chips corresponding to a bit are distributed throughout the frame. On the other hand, the interleaver provides an alternate way of computing an estimate for the joint probability. Since the chips are spread out in the frame far apart from each other and different interleavers are used for different users, it is reasonable to assume that the noise samples added to the chips are independent of each other, i.e., the joint distribution can be approximated by the product of marginals, (25) The terms in the product are the a priori probabilities . Therefore

The forward and backward probabilities and are calculated recursively using equations similar to (5) and (6) replaced by . with The transition probability is given by

(26) (22) is the a priori information. The state pair The term gives a combination for the past values of for is estimated using the equation each , from which (23) and The quantities as explained in Section II-A. are computed from the trellis, of a bit proNow, the log APP ratio estimate vides a new APP estimate for the chips in the particular bit period, i.e., for

(27)

788

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

Fig. 7. Comparison of error rates obtained for the cI-CDMA system with turbo f [0:227 0:460 0:688 0:460 0:227]. equalization based receiver. f SIR = 3 dB.

Fig. 8. Comparison of error rates obtained for the cI-CDMA system with those of the conventional CDMA system. SIR = 3 dB. f as before and f = [ 0:45 0:25 0:15 0:1 0:05].

Extrinsic information and a priori input to the equalizer for are computed using (17). the next iteration Final decisions are made, after several iterations, by passing through a sign detector. the log APP ratio estimates C. When s and Channels Are Identical

The particular case of identical interleavers and identical channels for the -users is explained in detail in [9]. When the interleavers are identical, one cannot make the assumption are independent over , given . that the variables Therefore, the joint APPs were computed by the MAP equalizer in [9], instead of the . Further, since the channels marginal APPs were assumed to be identical, the a posteriori distribution of the variable for different combinations of the vector was computed by the equalizer. A MAP MUD was required to process the joint APPs and compute the log APP . ratios The complexity of the equalizer could be reduced by comfor fewer than the possible values, puting the APP of without much loss in performance. In the general case where the interleavers are not identical, such a method for reducing complexity cannot be used. However, there is reduction in complexity since the MAP MUD is replaced by SUDs. D. Simulation Results/Observations The error rates obtained with the cI-CDMA system, for the same specifications used in Section II-B are plotted in Fig. 7. Random interleavers were generated for each frame and each user. The performance of the system is compared with that with no ISI. It is seen that the error rates with ISI are close to that obtained when there is no ISI as SNR increases. It is seen that there is a slight degradation compared with the results given in [9], where the interleavers were identical, for the same channel coefficients for both users. The reason for this

is that the two equalizers are different, as pointed out in previous section. A posteriori distributions of joint events are used in [9], whereas here they are approximated by marginal distributions. This approximation is more accurate for larger interleaver lengths. In fact, it was observed that the error rates for the fifth-iteration were even closer to the no-ISI case with frames of . length Fig. 8 shows the error rates obtained for the conventional CDMA system and cI-CDMA system with the channel coefused in [2]. ficients These results show that the system is indeed effective against both ISI and MAI. One of the disadvantages of the system is the complexity of the receiver. The order of complexity is for the equalizer, which is prohibitive for a large number of users. Alternative detectors, which work with a smaller number of states, like the M-algorithm could be used for the equalizer. Alexander et al. have presented a SISO version of the M-algorithm in [6]. In the next section, a SISO detector of linear complexity is presented. V. MMSE-OPTIMIZED CHIP-LEVEL INTERFERENCE CANCELLER (cIC) An interference canceller is a SISO device which takes in soft estimates of interfering bits, cancels their effect in the received signal and provides a soft estimate for the desired bit. They are found to be a very good low complexity alternative for detecting data in CDMA systems, especially when used in an iterative fashion. The iterative detectors can be classified into two categories, based on the way the ICs are used: parallel and serial ICs. PICs have been studied by Divsalar et al. [13], Buehrer and Woerner [14]. Here, bits of all users are estimated in parallel at each stage and their effect is cancelled from the received samples for the next iteration of estimation. In serial ICs [15][17], the most recent estimates of the interferers are used to cancel the interference. The detector of Patel and Holtzman [15] is not iterative since it stops when all users bits are estimated once, but it could be extended. The detectors based on the expectation

MAHADEVAPPA AND PROAKIS: MITIGATING MAI AND INTERSYMBOL INTERFERENCE IN UNCODED CDMA SYSTEMS

789

maximization (EM) algorithm, presented in Nelson and Poor [16] are iterative SICs. Here, the interference cancellers are derived using different variations of the EM algorithm. In particular, the space alternating generalized EM (SAGE) based IC was found to perform well. The MMSE-optimized iterated soft-decision interference canceller presented by Mller and Huber [17] is yet another SIC. Although similar in structure to the SAGE based algorithm, the soft estimates are obtained differently. The cIC used for detecting data in the cI-CDMA system, uses a soft estimator which is slightly different from the MMSEISDIC and SAGE-IC estimators. In addition to that, the cIC operates at the chip level and cancels both ISI and MAI, whereas the others operate at bit level and cancel MAI. Before going into the cIC based cI-CDMA receiver, the soft estimators in SAGE-IC and MMSE-ISDIC and the proposed cIC are studied here briefly. A. Soft Estimators in SAGE-IC, MMSE-ISDIC and cIC Consider a conventional CDMA system with synchronous transmission and additive white Gaussian noise (AWGN) channel, so that MAI is the only form of interference. For such a system, the sufficient statistics can be obtained by matched-filtering with the PN sequences at the receiver. Since the transmission is synchronous, symbol-by-symbol detection can be done without loss in performance. The matched filter outputs for each bit period can be given by

In contrast to the above estimators, the proposed cIC estimator takes the form

where

(31)

The estimator can be derived by applying the MMSE criterion from , assuming that the combination of for estimating residual interference and noise follows a Gaussian distribution , as outlined in [18, pp. 175]. The variance estimate along the lines of given by (31) is an approximation for equal to . (30), with variance of B. cIC Based cI-CDMA Receiver A cI-CDMA receiver can be built using the cICs as building blocks. The received sequence from (15) is passed through -maximal ratio combiners (MRCs). The th MRC is basically a filter matched to the th users channel. The output of the th , is fed to the th cIC. The structures of the MRC MRC, and cIC for the th user are shown in Fig. 9. Essentially, the MAP equalizer in Fig. 6 is split into blocks of cICs, resulting in loops, with each loop consisting of a cIC and a SUD. The soft estimates of the bits produced by the SUD are fed back to cICs of all the loops so that they can cancel the interference and produce new chip estimates. These estimates are processed by the SUD to get new soft estimates for the bits. The soft estimates are converted to hard estimates after several iterations. The various operations are explained in detail below. th MRC output can be given by The (32) be the soft estimate of the th chip transmitted by the Let th user, computed by the th SUD. An estimate of the ISI and MAI can be obtained using these soft estimates and subtracted . Let be the resulting output which consists of from the required chip information, residual ISI and MAI, and noise. Then

where the noise is distributed as being the cordenote the soft estimate for the th user relation matrix. Let bit. Let it be initialized to zero to start with. For each , MAI can and can be cancelled from to get be estimated using

(28) is the sum of residual interference and noise. where The SAGE based estimator is of the form (29) where This estimator doesnt take into account the noise due to residual interference and the correlation in the Guassian noise samples due to the front-end matched filtering. The MMSE-ISDIC incorporates the effect of the residual interference in computing the noise variance. The soft estimator here is of the form is the correlation between the filters and (33) , i.e.

where

(30)

In (33), the resulting noise sequence is zero-mean Gaussian, corresponds to the combut correlated. The first term and, hence, is not included in the bination double summation. Since the filter coefficients are normalized, . Assuming the combined residual interference and noise to be is given by Gaussian, the MMSE estimate for (34) where is the normalizing factor for the PN sequence and is for a binary PN sequence and is the equal to

, since Here, the noise variance is an approximation for are the most recent available estimates. the estimates Further, the correlation in is ignored.

790

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

Fig. 9. MRC and SISO equalizer for the k th user receiver.

variance of the combined noise and residual interference in (33). It is estimated using the past soft estimates of the interfering chips obtained from the SUDs

(35) can be considered as the The MMSE estimate mean value for a particular distribution on the alphabet , i.e. for some From (34) and (36), we get and (36)

(37) is in the form of a log APP ratio. Note that is deinterleaved to get which is fed to the SUDs for further processing. The SUD is essentially similar to the one described in Section IV-B, with additional computations for obtaining soft estimates of the bits and chips. The relevant opera, for convenience. New tions are described here in terms of and chips are computed using estimates
Fig. 10. Comparison of error rates obtained for the cI-CDMA system with cIC based receiver. f f [0:227 0:460 0:688 0:460 0:227]; SIR = 3 dB.

= =

(38) and where for (39) To summarize, the cIC based cI-CDMA receiver can be given by the following steps. using (32). 1. Obtain MRC outputs , for all and . 2. Initialize chip estimates 3. For each , , For each using (33) and required obtain compute new log APP ratio estimates (37).

End by deinterleaving . Obtain , For each using (38) compute estimates using (39) compute estimates End to get new estimates . Interleave estimates End 4. Jump to step 3 for the next iteration. after sufficient iterations. 5. Make hard decisions on C. Simulation Results/Observations A two-user cI-CDMA system with the low complexity receiver described above was simulated. The channels considered in Section IV-D were used for these simulations too. Fig. 10 shows the error rates obtained for and as above Fig. 11 shows the error rates obtained with . By comparing and with the corresponding optimal error rates for conventional CDMA system shown in Figs. 4 and 8, it can be inferred that the cIC based receiver also effectively removes ISI and MAI.

using

MAHADEVAPPA AND PROAKIS: MITIGATING MAI AND INTERSYMBOL INTERFERENCE IN UNCODED CDMA SYSTEMS

791

REFERENCES
[1] C. Berrou and A. Glavieux, Near optimum error-correcting coding and decoding: Turbo codes, IEEE Trans. Commun., vol. 44, pp. 12611271, Oct. 1996. [2] C. Douillard, M. Jzquel, C. Berrou, A. Picart, P. Didier, and A. Glavieux, Iterative correction of intersymbol interference: Turbo equalization, Eur. Trans. Telecomm., vol. 6, pp. 507511, Sept.Oct. 1995. [3] G. Bauch, H. Khorram, and J. Hagenauer, Iterative equalization and decoding in mobile communication systems, in Proc. Eur. Personal Mobile Commun. Conf., Sept. 1997, pp. 307312. [4] L. Bahl, J. Cocke, F. Jelinek, and J. Raviv, Optimal decoding of linear codes for minimizing symbol error rate, IEEE Trans. Inform. Theory, vol. IT-20, pp. 284287, Mar. 1974. [5] X. Wang and H. V. Poor, Iterative (turbo) soft interference cancellation and decoding for coded CDMA, IEEE Trans. Commun., vol. 47, pp. 10461061, July 1999. [6] P. D. Alexander, M. C. Reed, J. A. Asenstorfer, and C. B. Schlegel, Iterative multiuser interference reduction: Turbo CDMA, IEEE Trans. Commun., vol. 47, pp. 10081014, July 1999. [7] J. G. Proakis, Digital Communications. New York: McGraw-Hill, 1995. [8] M. C. Valenti, Iterative Detection and Decoding for Wireless Communications, Ph.D. Dissertation, Dept. Elect. Eng., Virginia Polytechnical Inst., Blacksburg, VA, July 1999. [9] R. H. Mahadevappa and J. G. Proakis, Turbo equalization method for mitigating multiple access interference and intersymbol interference in uncoded CDMA systems, in Proc. 2000 Conf. Information Sciences Systems, Princeton, NJ, Mar. 2000, pp. FA6.1FA6.6. [10] S. Vasudevan and M. K. Varanasi, Optimum diversity combiner based multiuser detection for time-dispersive Rician fading CDMA channels, IEEE J. Select. Areas Commun., vol. 12, pp. 580591, May 1994. [11] U. Fawer and B. Aazhang, A multiuser receiver for code division multiple access communications over multipath channels, IEEE Trans. Commun., vol. 43, pp. 15561565, Feb./Mar./Apr. 1995. [12] S. E. Bensley and B. Aazhang, Subspace-based channel estimation for code division multiple access communication systems, IEEE Trans. Commun., vol. 44, pp. 10091020, Aug. 1996. [13] D. Divsalar, M. K. Simon, and D. Raphaeli, Improved parallel interference cancellation for CDMA, IEEE Trans. Commun., vol. 46, pp. 258268, Feb. 1998. [14] M. Buehrer and B. D. Woerner, Analysis of adaptive multistage interference cancellation for CDMA using an improved Gaussian approximation, IEEE Trans. Commun., vol. 44, pp. 13081321, Oct. 1996. [15] P. Patel and J. Holtzman, Analysis of a simple successive interference cancellation scheme in a DS/CDMA system, IEEE J. Select. Areas Commun., vol. 12, pp. 796807, June 1994. [16] L. B. Nelson and H. V. Poor, Iterative multiuser receivers for CDMA channels: An EM-based approach, IEEE Trans. Commun., vol. 44, pp. 17001710, Dec. 1996. [17] R. R. Mller and J. B. Huber, Iterated soft-decision interference cancellation for CDMA, in Broadband Wireless Communications, M. Luise and S. Pupolin, Eds. New York: Springer-Verlag, 1998, pp. 110115. [18] A. Papoulis, Probability, Random Variables and Stochastic Processes, 3rd ed. New York: McGraw-Hill, 1991. [19] S. Tachikawa and G. Marubayashi, Spread time spread spectrum communication systems, in PROC.IEEE GLOBECOM87, vol. 16.5, Nov. 1987, pp. 617619. [20] X. Gui and T. S. Ng, A novel chip-interleaving DS SS system, IEEE Trans. Veh. Technol., vol. 49, pp. 2127, Jan. 2000. [21] P. Frenger, P. Orten, and T. Ottosson, Code-spread CDMA using maximum free distance low-rate convolutional codes, IEEE Trans. Commun., vol. 48, pp. 135144, Jan. 2000.

Fig. 11. Comparison cI-CDMA system with

cIC based receiver. f as = [p0 45 p0 25 p0 15 p0 1 p0 05] SIR = 03 dB.

of

error

rates

obtained

for before

the and

It is also seen that the error rates close to the no-ISI, single-user error rates are reached with as few as five iterations, in both cases. Further, the receiver complexity was linear in number of users and length of the channel, whereas the complexity of the MAP equalizer was exponential in the same parameters. From these results, it appears that the cI-CDMA, even with a simple linear complexity receiver, is a very good alternative to the conventional CDMA system, when there is significant ISI in the channel. The cI-CDMA system is effective in single path fading channels too. Although such channels do not introduce ISI, it is seen that the conventional CDMA system suffers heavy losses, due to deep fading. In the cI-CDMA system, by using sufficiently large interleavers, one can ensure on an average that all the chips of a particular bit do not undergo severe fading, thus, improving the average performance. The actual amount of loss recovered would depend on the length of the interleaver, the spreading , the Doppler spread and the number of users in the factor system.

VI. CONCLUDING REMARKS A modified version of the CDMA communication system, called the cI-CDMA system was presented. It was shown that this system is effective in mitigating both ISI and MAI. Two receivers were presented. The turbo equalization method based receiver is computationally intensive, since it requires a MAP equalizer whose complexity is exponential in the number of users and the lengths of the channels. The MMSE-optimized cIC based receiver is able to achieve performance close to that of the turbo equalization method, with computational complexity linear in number of users and channel lengths. The results presented assumed that the channel coefficients were known at the receiver. The effect of channel estimation, convergence characteristics of the receiver, effect of Doppler spread, and capacity of the system are topics which require some study.

Ravishankar H. Mahadevappa received the B.Tech. degree in electrical and computer engineering from the Indian Institute of Technology, Madras, INDIA, in 1994 and the M.S.E.E. degree from Northeastern University, Boston, MA, in 1998. His research interests are in multiuser communications, signal processing, and information theory.

792

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

John G. Proakis (S58M62SM82F84LF97) received the B.S.E.E. degree from the University of Cincinnati, OH, in 1959, the M.S.E.E. degree from Massachusetts Institute of Technology (MIT), Cambridge, in 1961, and the Ph.D. degree from Harvard University, Cambridge, MA, in 1967. He is currently an Adjunct Professor at the University of California, San Diego, and Professor Emeritus at Northeastern University, Boston, MA. From 1969 to 1998, he was a Faculty Member at the Electrical and Computer Engineering Department, Northeastern University and held the positions of Department Chair (19841997), Associate Dean and Director of the Graduate School of Engineering (19821984), and Acting Dean (19921993). Prior to joining Northeastern University, he was with GTE Laboratories and the MIT Lincoln Laboratory. His professional experience and interests are in the general areas of digital communications and digital signal processing. He is the author of Digital Communications (New York: McGraw-Hill, 1983, 1st ed.; 1989, 2nd ed.; 1995, 3rd ed.; 2001, 4th ed.) and coauthor of Introduction to Digital Signal Processing (New York: Macmillan, 1988, 1st ed.; 1992, 2nd ed.; 1996, 3rd ed.); Digital Signal Processing Laboratory (Englewood Cliffs, NJ: Prentice-Hall, 1991); Advanced Digital Signal Processing (New York: Macmillan, 1992); Algorithms for Statistical Signal Processing (Englewood Cliffs, NJ: Prentice-Hall, 2002); Discrete-Time Processing of Speech Signals (New York: Macmillan, 1992, IEEE Press, 2000); Communication Systems Engineering, (Englewood Cliffs, NJ: Prentice-Hall, 1994, 1st ed.; 2002, 2nd ed.); Digital Signal Processing Using MATLAB V.4 (Boston, MA: Brooks/Cole-Thomson Learning, 1997, 2000); Contemporary Communication Systems Using MATLAB (Boston, MA: Brooks/Cole-Thomson Learning, 1998, 2000). He holds five patents and has published over 150 papers.