An MA model based blind source separation algorithm

We give a maximum entropy or maximum likelihood approach for blind source separation (BSS). This approach model the sources as filtered versions of white zero-mean signals by a class of moving average (MA) filters. We consider not only the effect of instantaneous measurements, but also the effect of delayed measurements. Compared to the dynamic component analysis (DCA) algorithm, we do not need to estimate the large number of the parameters of the probability density function (PDF) model of white signals. The proposed algorithm can separate the mixture of Gaussian sources.