Uma aplicação do algoritmo genético baseado em tipos abstratos de dados ao problema de separação cega de fontes

Abstract

This dissertation presents a model of a genetic algorithm based on abstract data types, applied to problems of blind source separation. In this sense, it is through comparisons of approaches using genetic algorithm Holland and Genetic Algorithm Based on Abstract Data Types, in order to show the efficiency of this algorithm in the characterization of a combined solution of complex optimization problems, using so parsimonious computational resources. The problems in applications blind source separation are concentrated in two parts, one in the recovery of an electrocardiographic signal in noisy environment, from the source separation process itself, and the other in the characterization of the inverse filter, a problem of blind deconvolution . These two approaches, seeks to, through the evolutionary process, adjusting the weighted coefficients of the systems from consideration as the cost function, some techniques for the separation of sources, such as kurtosis, mutual information, negentropy and especially their combination. The results obtained by simulations show a good compromise between performance and computational cost.