Performance improvement of coded CDMA receiver with weighted parallel interference cancellation based on symbol reliability

Abstract

A novel weighted parallel interference cancellation (PIC) detector with symbol-by-symbol weight determination is proposed to improve the performance of PIC detectors for convolutionally coded CDMA systems. It has been known that the performance of the PIC detector is highly dependent on reliability of the data decisions at the first stage. In coded CDMA systems, the post-decoding PIC has been proposed, in which channel decoding is utilized to improve the performance of the first stage data decisions. In convolutionally coded systems, the Viterbi algorithm is generally employed as a channel decoder. Although the Viterbi algorithm is used for improving the performance of the first stage data decisions, the performance of the second stage Viterbi decoding is degraded due to error dependence introduced from the first stage Viterbi algorithm. Hence the PIC performance gain through the enhanced first stage decisions is limited. In this work, it is shown that the weighted PIC detector can overcome the problem of the post-decoding PIC detector. Even though effectiveness of the weighted PIC detector has been stated in previous works, most works have used empirical methods for weight determination because of difficulty in optimum weight determination. The empirical weight takes only one optimum value on the average for given channel environments because the empirical method can not benefit by the decision reliability of each symbol used in the interference cancellation, and even the method can hardly be implemented in real systems. In this work, a new weighted PIC detector is proposed together with symbol-by-symbol weight determination based on symbol reliability. For generation of the reliability information, soft-output Viterbi algorithm (SOVA) is employed as the first stage channel decoder in the post-decoding PIC detector. Investigations are carried out into the characteristics of the error probability and the log-likelihood ratio generated from SOVA. Two methods o...