Time-Scale Referenced Cumulant for Blind Source Separation

Blind source separation consists of recovering a set of signals of which only linear mixtures are observed. Several techniques based on statistical or time-frequency distributions have been developed to solve this problem. This paper introduces a new separation technique, in case of instantaneous mixtures, that exploits statistical information and wavelet analysis for independent sources. A proposed Time Scale Referenced Cumulant (TSRC) characterizes the evolution of high-order statistics (HOS) in the time scale domain. The estimation of the sources or the mixing matrix can be performed by non-unitary joint diagonalization of a set of matrices constructed by using the proposed TSRC. A numerical simulations are presented to demonstrate the effectiveness of the suggested approach. Keywords- Blind source separation, Referenced Cumulants, Time Scale Analysis, Non-Orthogonal Joint Diagonalization.