Maximum Likelihood Multiuser Detection of DS/CDMA Signals in Improper Noiset

The improper signals are often encountered in communications and signal processing. Since the known maximum-likelihood (ML) multiuser detection problem is only for proper noises, we derive the improper version of this expansion. We show that the proposed scheme improves the near-far resistance (NFR) for any spreading sequences and channel conditions. This gain comes from appropriate management of the additional information contained in nonzero pseudo-covariance matrix. The average NFR is obtained in a random channel environment and random spreading sequence. The proposed scheme halves the effective number of users. Although the ML multiuser detection gives us the optimum bit error rate (BER) performance, the computational complexity that is exponential in the number of users makes it impractical. In this paper, an efficient ML multiuser detection is developed. First, we relieve the combinatorial constraint of ML detection and obtain the initial decision of the symbols. Then, the most error probable symbols are chosen by referring the reliability measures of the initial symbols. The ML searching is accomplished with only the chosen symbols. Computer simulations demonstrate the results of the paper and show the error rate performance of the proposed near-ML multiuser detection