A FAST ALGORITHM FOR BLIND SEPARATION OF NON-GAUSSIAN ANDTIME-CORRELATED SIGNALS

摘要

In this article we propose a computationally efficientrnmethod (termed FCOMBI) to combine the strengths of non-rnGaussianity-based Blind Source Separation (BSS) and crosscorrelations-rnbased BSS. This is done by fusing the separationrnabilities of two well-known BSS algorithms: EFICArnand WASOBI. Simulations show that our approach is atrnleast as accurate and often more accurate that other stateof-rnthe-art approaches which also aim to separate simultaneouslyrnnon-Gaussian and time-correlated components.rnHowever, in terms of computational efficiency and stability,rnFCOMBI is the clear winner which makes it specially suitablernfor the analysis of very high-dimensional datasets likernhigh-density Electroencephalographic (EEG) orMagnetoencephalographicrn(MEG) recordings.