DESPRIT-histogram based blind source separation of more sources than sensors using subspace methods

A blind source separation (BSS) technique is presented which combines the subspace processing and associated DOA capability of ESPRIT with the weighted histogram of DUET. The method allows for a weakened version of the disjoint signal assumption of DUET and can be seen as a natural extension of DUET to more than two mixtures. Instead of only allowing at most one source be active at any time-frequency point, up to M-1 sources can be active where M is the number of anechoic mixtures. The combination of these techniques creates a DUET-ESPRIT (DESPRIT) blind source separation algorithm which can demix an arbitrary number of sources N, even when N > M.