Adaptive equalization for PAM and QAM signals with neural networks

Abstract

The authors investigate the application of neural networks to adaptive and blind equalization problems. The purpose is twofold: (1) to introduce a new realization structure of a multilayer perceptron (MLP) with a backpropagation training algorithm and show that it works well for both PAM and quadrature amplitude modulation (QAM) signals of any constellation size, and (2) to demonstrate the performance of self-organizing maps (SOMs) as blind equalizers and establish that they are simply not powerful enough for this problem, especially when the intersymbol interference is large. A new MLP structure for adaptive equalization of PAM and QAM signals is described and its performance, along with the simulation results of SOMs as blind equalizers, is demonstrated. >