Least Bit Error Rate Adaptive Multiuser Detection

Abstract

Linear detector required at direct-sequence code division multiple access (DS-CDMA)communication systems is classically designed based on the minimum mean squares error (MMSE) criterion, which can efficiently be implemented using the standard adaptive algorithms, such as the least mean square (LMS) algorithm. As the probability distribution of the linear detector's soft output is generally non-Gaussian,the MMSE solution can be far away from the optimal minimum bit error rate (MBER) solution. Adopting a non-Gaussian approach naturally leads to the MBER linear detector. Based on the approach of Parzen window or kernel density estimation for approximating the probability density function (p.d.f.), a stochastic gradient adaptive MBER algorithm, called the least bit error rate (LBER), is developed for training a linear multiuser detector. A simplified or approximate LBER (ALBER) algorithm is particularly promising, as it has a computational complexity similar to that of the classical LMS algorithm. Furthermore, this ALBER algorithm can be extended to the nonlinear multiuser detection.