Code division multiple access communications: multiuser detection based on a recurrent neural network structure

Abstract

A multiuser detector based on a recurrent neural network structure (MU-RNN) is derived for a direct sequence code division multiple access communication system with multi path propagation. Contrary to other neural network approaches the MU-RNN has the advantage, that the network size as well as the coefficients of the network can be derived from parameters which characterize the communication system. The energy function of the MU-RNN is identical to the log-likelihood function of an optimum multiuser detector. The performance of the MU-RNN is compared to other optimal and suboptimal multiuser detectors. The MU-RNN can achieve the same or almost the same BER as the optimal multiuser detector while at the same time the complexity is much lower.