Performance improvement of neural network equalizers

Nonlinear equalizers find use in applications where the channel distortion is too severe for a linear equalizer to handle. Unfortunately, most nonlinear equalizers are usually too complicated to meet real-time processing demands. Neural networks offer a computationally-efficient alternative to currently-used nonlinear filter realizations. However, because the backpropagation training algorithm of MLP is a generalized LMS algorithm, it suffers the same problem as the linear LMS, i.e., slow convergence. We introduce two techniques to improve the performance of MLP equalizers. One is based on the normalization of the adaptation term to the derivative of the output activation function, the other is using the pre-orthogonalization technique. Significant improvement in both the convergence and stationary error performance has been shown in the simulation results