Adaptive Decision Feedback Multiuser Detectors with Recurrent Neural Networks for DS-CDMA in Fading Channels

Abstract

In this work we propose adaptive decision feedback (DF) multiuser detectors (MUDs) for DS-CDMA systems using recurrent neural networks (RNN). A DF CDMA receiver structure is presented with dynamically driven RNNs in the feedforward section and finite impulse response (FIR) linear filters in the feedback section for performing interference cancellation. A stochastic gradient (SG) algorithm is developed for estimation of the parameters of the proposed receiver structure. A comparative analysis of adaptive minimum mean squared error (MMSE) receivers operating with SG algorithms is carried out for linear and DF receivers with FIR filters and neural receiver structures with and without DF. Simulation experiments including fading channels show that the DF neural MUD outperforms DF MUDs with linear FIR filters, linear receivers and the neural receiver without interference cancellation.