A Blind Separation Algorithm with a Linear Constraint

It is known that the task of blind source separation has an inherent ambiguity that is called scaling indeterminacy or filtering one. Namely, since the only prior knowledge about the sources is that they are statistically independent, any linearly filtered version of a source signal can be considered another form of the source. As an idea for eliminating the indeterminacy, one of the authors proposed a principle named the minimal distortion principle (MDP). The principle designs the separator so that its output may be the least subjected to distortion. This paper addresses a new idea for eliminating the indeterminacy. While the separator based on the method preserves signal quality as MDP does, its implementation is much easier than MDP. Moreover we describe a local minimum problem in the algorithm and show a solution to it