An information-maximization approach to blind separation and blind deconvolution