Minimal distortion principle for blind source separation

Abstract

In blind source separation, the number of sensors is usually assumed to be equal to the number of sources. In this case, an indeterminacy appears with which any linear transform of an estimated source signal can also be considered another estimation of the source signal. Moreover, in the case that the number of sensors is greater than the number of sources, another indeterminacy arises due to the redundancy of the sensors. Although these indeterminacies are often considered unsubstantial and have been eliminated without definite bases, an appropriate normalization of the separator is important to enhance the accuracy of the separation result, particularly in the case of a convolutive mixture. The paper shows two principles for eliminating these indeterminacies: (i) minimal distortion principle; (ii) inverse minimal distortion principle.