OFDM pulse shape generation using artificial neural networks

Abstract

This paper presents an experiment of using neuronal networks as a pulse shape generator for CP-OFDM. A set of pulse shapes were generated using genetic algorithm that minimizes the mean square error of the timing offset estimator. These pulse shapes were used to train function approximation neural networks. Such neural networks make the use of adaptive pulse shaping in OFDM systems feasible. Results from neural networks simulation, which have shown the ability of neural networks to fulfill such function, are presented.