Underdetermined Source Separation for Non-Stationary Signal

Abstract

This paper proposes a new blind source separation (BSS) technique for underdetermined MIMO system using space-time-frequency distributions. The sparseness in time-frequency domain is exploited for underdetermined BSS. The paper also proposed a new detection technique for auto sources time-frequency points, which is incorporated into the BSS technique. The proposed BSS technique applicable to the situations where sources' time-frequency signatures overlap at certain points in the time-frequency plane and even in situations where there are sources with multicomponent signals. In these situations, where time-frequency planes cannot be partitioned (masked) into groups of fewer or equal number sources than antennas, the existing overdetermined time-frequency BSS techniques could not be applied. Simulation results are included to show its effectiveness.