On the indeterminacies of convolutive blind signal separation based on second order statistics
Daniël W. E. Schobben, P.C.W. Sommen · 2003
Several blind signal separation algorithms have been developed which are based on second order statistics. Little has been published however on whether second order statistics are sufficient to obtain a unique solution. Especially for applications that involve convolutive mixing and unmixing of signals that are correlated in time, there is a lack of knowledge on why and in what cases second order statistics suffice. This paper investigates the indeterminacies that are introduced when second order statistics are used and presents a theorem for the unmixing system to be uniquely found using second order statistics.