Aprendizado de Dicionários para a Separação Semi-Cega de Fontes por Matching Pursuit

The Sparse Component Analysis is one of the most used techniques for handle the source separation problem. In this paper, a technique based in atomic decompositions, is used in the identification of a mixture system. If isolated excerpts of the sources are available, we show that using adapted dicionaries can imply a better (with a inferior MSE in, approximately, 15 dB) and less computationally intensive identification procedure.