Blind separation of convolved sources using the information maximization approach

In a number of real-world signal processing applications, signals from various independent sources may get distorted by environmental factors that can be represented as convolutive mixtures of original signals received at the sensors. In this paper, the effects of environmental factors and modeling assumptions on the performance capabilities of independent component analysis-based techniques are investigated. The so-called blind source separation feedback network architecture that is capable of coping with convolutive mixtures of sources is derived using Bell and Sejnowski´s information maximization principle. We develop ideal solutions for separation of independent source signals from the convolutive mixtures that is applicable to an arbitrary N × N feedback network architecture. A number of simulation case studies corresponding to various types of environment filters are presented using synthetically generated data.