Neural Equalizer Performance Evaluation Using Genetic Algorithm

Artificial Neural Networks (ANN) have been successfully applied to deal with linear or nonlinear problems. The best ANN architecture choice is not a trivial task to be performed and requires some a priori knowledge. In this work, we propose a Genetic Algorithm (GA) evaluation approach to determine the best combination of ANN and learning algorithm for equalization propose. A comparative analysis, using well known neural architectures, is presented in order to accomplish a 4-QAM equalization of signals submitted to Inter Symbol Interference (ISI), inherent in typical mobile communication channels. MLP, FLANN, PPN and three RNN based ANN structures, trained using backpropagation algorithm and others, have been evaluated.