On the behavior of current second order blind source separation methods for first and second order cyclostationary sources-application to CPFSK sources

Most of the current Second Order (SO) and Higher Order (HO) blind source separation (BSS) methods aim at blindly separating statistically independent sources, assumed zero-mean, stationary and ergodic. However in practical situations, such as in radiocommunications contexts, the sources are non stationary and very often cyclostationary. In a previous paper [5] the cumulant-based BSS problem for cyclostationary sources has been analysed assuming zero-mean sources (linear modulations). The purpose of this paper is to analyse the behavior and to propose adaptations of the current SO BSS methods for cyclostationary sources, assumed cyclo-ergodic, which are not zero-mean but first order and SO cyclostationary, such as some CPFSK sources (non linear modulations).