Minimum Distance Index for BSS, Generalization, Interpretation and Asymptotics

Abstract

We consider complex valued linear blind source separation, where the signal dimension might be smaller than the dimension of the observable data vector. In order to measure the success of the signal separation, we propose an extension of the minimum distance index and establish its properties. Interpretations for the index are derived through connections to signal-to-noise ratios and correlations. The interpretations are novel also for the real valued original case. In addition, we consider the asymptotic behavior of the extended minimum distance index. This paper is an invited extended version of the paper presented at the CDAM 2019 conference.