A decision feedback recurrent neural networks equalizer for digital communication

Two decision feedback equalizer structures employing recurrent neural networks (RNN) are proposed, which put the traditional decision feedback structure for linear channels equalization skilfully into the RNN, substitute the training signal for a decision feedback signal in the learning process and adaptively adjust the learning step. The first type of two new equalizer structures is used for non-linear channels with severe intersymbol interference (ISI) and mild non-linear distortion, its simulation results have shown that it has better equalization performance than traditional recurrent neural networks equalizer (RNNE) with the same number of parameters.