Fast nonlinear channel equalisation using generalised diagonal recurrent neural networks

Abstract

A generalised diagonal recurrent neural network (GDRNN) for nonlinear channel equalisation is proposed. The hidden nodes of the GDRNN have recurrent weights to capture the dynamic characteristics of the communication channels. The learning algorithm of the proposed GDRNN is derived, based on constrained optimisation. The proposed neural network gives faster learning speed and has better convergence properties than do conventional channel equalisers.