A novel concept of embedding orthogonal basis function expansion block in a neural equalizer structure for digital communication channel

Abstract

The proposed neural equalizer structure is based on a novel orthogonal basis function (OBF) expansion technique, motivated by genetic evolutionary concept, which utilizes a self-breeding approach to evolve new information to consolidate the final output. Here, the decision at a feedforward neural network (FNN) node termed as expert opinion of a generation undergoes an orthogonal expansion in two dimensions, where one of the outputs possessing the knowledge base for that generation participates in taking the final decision. Hence, a collective judgment based on the expert opinions evolved from decisions of individual generations gives a more rational and heuristic solution compared to a conventional feedforward neural network (CFNN) structure. Propagation of output error backwards and calculation of local gradients at each node become a difficult task as the OBF block is positioned in between the neurons of different layers. In order to circumvent such situation, a new technique has been evolved. The developed equalizer structure using this concept has outperformed the CFNN equalizer with wide margins. Further their bit-error-rate performances are close to that of Bayesian equalizer, which is optimal in the theoretic sense. Application of this proposed technique also reduces the structural and computational complexity of conventional neural equalizers. Hence, this efficient equalizer structures suitable for digital communication channels have the potential for real-time implementation in DSP, FPGA processors also.