Abstract: The existing neural network multi-user detectors are often trapped in the local minima, resulting in the performance degradation. In this paper, a timevarying scaling-parameter transiently chaotic neural network (TSTCNN) is proposed. The TSTCNN network has powerful capability to escape from getting into the local minima. The TSTCNN network is applied to the optimum detection problem in DS/CDMA systems. Numerical results show that the TSTCNN-based detector can perform better than the existing neural network detectors. The proposed detector can approximate to the optimum detector closely.

Key words: multi-user detection, optimum detection, chaotic neural network, near-far problem