Neural network channel equalization

The authors present computer simulation results of the performance of conventional and perceptron-based equalizers in the presence of intersymbol interference (ISI), additive noise and co-channel interference (CCI). They show that a three-layer perceptron equalizer which is trained and updated by a complex-valued backpropagation adaptive algorithm can achieve acceptable bit error rate performance and demonstrate the effect of the perceptron configuration, size, and adaptation parameters on the equalizer performance. It is shown that the perceptron equalizer can essentially match the performance of a conventional equalizer under all noise and interference conditions. It is observed that the convergence rate of the perceptron-based equalizer is much slower than that of the conventional equalizer. With a sufficiently small adaptation step size, both of these equalizers are robust to decision-directed error propagation during data transmission