INVESTIGADORES
CAIAFA Cesar Federico
congresos y reuniones científicas
Título:
A non-Gaussianity measure for blind source separation
Autor/es:
CESAR F. CAIAFA; ARACELI N. PROTO
Lugar:
Rennes, FRANCE
Reunión:
Workshop; Signal Processing with Adaptative Sparse Structured Representations - SPARS 05; 2005
Institución organizadora:
Istitut de Recherche en informatique et Systemes Aleatoires - IRISA
Resumen:
Blind Source Separation (BSS) problem has been extensively studied for signals composed by statistical independent components. As it is well known, the applied methods usually fail when sources exhibit some degree of dependence. Our work points to solve the BBS problem removing the condition of statistical independence looking for sources estimates that maximize a measure of nonGaussianity. We present a measure of nongaussianity based on the L2Euclidean distance using a nonparametric technique for the estimation of probability densities. These mathematical tools have allowed us to build new algorithms for BSS which may have good performance for dependent as well as independent nonGaussian real world sources
