BLIND SOURCE EXTRACTION IN GAUSSIAN NOISE

Sergio Crucesy  , Andrzej Cichockix, Luis Castedoz, x Lab. for Open Information Systems, Brain Science Institute, RIKEN, Japan. E-mail: cia@brain.riken.go.jp, zElectronic and Systems, University of La Coruņa, Spain. E-mail: luis@sol.des..udc.es
ySignal Processing Group, University of Seville, Spain. E-mail: sergio@viento.us.es, x Lab. for Open Information Systems, Brain Science Institute, RIKEN, Japan. E-mail: cia@brain.riken.go.jp, zElectronic and Systems, University of La Coruņa, Spain. E-mail: luis@sol.des..udc.es

In this paper we address the problem of recovering a subset of sources from a noisy linear mixture. We pro- pose a novel blind source extraction algorithm that is robust with respect to the noise. This robustness is two-fold: on the one hand the algorithm does not lead to biassed estimates and, on the other, it minimizes the amount of signal and noise interference on the es- timated sources. In addition, the asymptotic conver- gence of the algorithm to the extraction solution is demonstrated for almost any source distribution. Fi- nally, the proposed algorithm is shown to be a gener- alization of the powerful kurtosis FAST-ICA algorithm that enables us to sequentially extract the sources in an arbitrary number of groups.