Blind source separation using second-order cyclic-statistics

This paper addresses the problem of blind separation of cyclostationary sources. By exploiting the cyclostationarity property of the source signals, a new approach based on second-order cyclic statistics is proposed for identifying the parameter matrix and estimating the source signals. The new approach does not impose any restriction on the cyclic frequencies of the signals and yields waveform-preserving estimates of the source signals. Simulation examples are presented to illustrate the effectiveness of this approach.