A blind separation algorithm that preserves signal quality

Abstract

In blind source separation the definition of sources has an inherent indeterminacy; any linear transform of a source signal can also be considered another form of source signal. Due to this indeterminacy, the determination of the separator has a certain freedom. This paper proposes an idea for eliminating the indeterminacy, i.e., a principle for uniquely normalizing the separator. The output of the separator based on the principle undergoes the least distortion relative to the sources signals observed at the sensors. The obtained normalization has some good properties: easy implementation, no fluctuation, and no signal distortion.