FREQUENCY-DOMAIN INFOMAX FOR BLIND SEPARATION OF CONVOLUTIVE MIXTURES
Cristina Mejuto, Adriana Dapena, Luis Castedo
Tel: 34 981 167150, Fax: 34 981 167160, e-mail: cris@des. .udc.es, adriana@des. .udc.es
Blind Source Separation (BSS) is a well-known
problem that arises in a large number of signal process-
ing applications. In this paper we present a new ap-
proach developed in the context of unsupervised learn-
ing of Neural Networks (NN) and based on the In-
fomax Principle for the separation of linear mixtures
of sources with memory (convolutive mixtures). The
problem is solved in the frequency domain, turning
the convolution operation in the time domain into a
multiplication in the frequency domain. The simula-
tion results show the performance of the proposed al-
gorithm for the separation of convolutive mixtures of
white sources.