Multiuser detection in CDMA based on the annealed neural network

Abstract

Recently, several neural network based multiuser detectors in code division multiple access (CDMA) have been studied. The Hopfield neural network based detector showed good performance by properly selecting the parameters of the network and restricting initial states to the outputs of a conventional detector. Inspired by these results, we propose an annealed neural network approach to coherent multiuser detection in a synchronous code division multiple access channel. This technique combines good features of the Hopfield neural network and simulated annealing. This technique eliminates restriction of initial states for the neural network. The performance of this technique is near optimal and close to that of the optimum multiuser detector in the presence of a Gaussian noise and the near-far problem. We will present theoretical and simulated performance enhancement of this technique to the existing methods.