Blind separation and identification of cyclostationary processes

This paper deals with the blind separation of convolutive mixtures when the sources are uncorrelated and wide-sense (quasi) cyclostationary random processes with the same cyclic frequencies. The proposed method relies on the successive diagonalisation of the spectral matrix and the cyclic spectral matrix of multiple measurements. It is shown how the combined use of the spatial diversity - as offered by sensors - and the frequency diversity - as offered by cyclostationarity - enables the resolution of the problem in a general non-parametric way. Unlike other methods, the solution is attained´ up to a constant permutation matrix which does not depend on frequency. The approach is extended to the blind identi · cation of the mixing transfer matrix when the sources are white.