Iterative CDMA Multiuser Receiver With Soft Decision-Directed Channel Estimation

We develop an iterative code-division multiple-access (CDMA) receiver for multipath block-fading channels in the uplink. The mean-square error (MSE) of the channel estimates is reduced iteratively by using soft data estimates as additional pilots. The multiuser detector performs parallel interference cancellation (PIC) with subsequent linear filtering. We consider a single-user matched filter (SUMF) and develop a low-complexity linear minimum MSE (LMMSE) filter for intersymbol interference (ISI) channels that allows for 33% more users than the SUMF. We investigate two choices of training sequences for initial channel estimation. The first is a random pilot symbol sequence modulated by a short spreading sequence (RAND), and the second is a perfect root of unity sequence (PRUS) used as long sounding chip sequence. We observe that the RAND scheme approaches the single-user bound up to 2 dB and in case of PRUS up to 1 dB at a target bit error rate (BER) of 10^-3. Further, several channel estimators are developed that use various combinations of soft decision symbols and statistical a priori knowledge about the channel. These are the approximated least-squares (ALS), approximated linear MMSE (ALMMSE), and linear MMSE (LMMSE) estimators. The iterative estimators show an improvement of 2.5 and 5 dB with PRUS and RAND pilots, respectively, over estimators that do not exploit soft feedback symbols at a BER of 10 ^-3. The ALS is sacrificing less than 0.6 dB to obtain a BER of 10^-3 in a system at load 1.5 when compared with the ALMMSE and LMMSE schemes. The load is understood as the ratio of the number of users to the spreading factor. This paper shows that significant improvements in system capacity can be achieved in multipath scenarios with low-complexity choices of multiuser detectors and channel estimators