A new approach to blind separation of cyclostationary sources

This paper studies the problem of blind source separation (BSS) under the assumption that the source signals are cyclostationary. Attention is restricted to methods based on second-order cyclostationary statistics (SOCS). Necessary and sufficient conditions for SOCS-based identifiability and SOCS-based separability are presented. Algorithms are developed to achieve BSS using only SOCS.