Simultaneous Blind Separation of Instantaneous Mixtures With Arbitrary Rank

This paper presents a gradient-based method for simultaneous blind separation of arbitrarily linearly mixed source signals. We consider the regular case (i.e., the mixing matrix has full column rank) as well as the ill-conditioned case (i.e., the mixing matrix does not have full column rank). We provide one necessary and sufficient condition for the identifiability of simultaneous blind separation. According to our identifiability condition and the existing general identifiability condition, all source signals are separated into two categories: separable single sources and inseparable mixtures of several single sources. A sufficient condition is also derived for the existence of optimal partition of the mixing matrix which leads to a unique maximum set of separations. One sufficient condition is proved to show that each maximum partition of the mixing matrix corresponds to a unique class of separated signals and as a result we can determine the number of maximum partitions from the classes of outputs under different separation matrices. For sub-Gaussian or super-Gaussian source signals, a cost function based on fourth-order cumulants is introduced to simultaneously separate all separable single sources and all inseparable mixtures. By minimizing the cost function, a gradient-based method is developed. Finally, simulation results show the effectiveness of the present method