Modified Hopfield Neural Network for CDMA Multiuser Detection

Abstract

We investigate the application of modified Hopfield neural networks (HNNs) based on Annealing Techniques to the problem of multiuser detection in spread spectrum CDMA communication systems. It is shown that the NP-complete problem of minimizing the objective function of the optimal multiuser detector (OMD) can be translated into minimizing an HNN “energy” function, thus allowed to take advantage of the ability of HNNs to perform very fast gradient descent algorithms in analog hardware. The performance of the proposed HNN receiver is evaluated via computer simulations and compared to that of the general HNN as well as to that of the OMD for CDMA transmission cases. It is shown that the modified HNN detection scheme exhibits a number of attractive properties and that it provides in fact more powerful performance than the general HNN scheme or the OMD scheme.