A Blind Source Separation Method based on the Fourth-order Cyclic Cumulant

Blind source separation is how to recover a set of signals from a set of their observations coming to multi-sensors, without any prior knowledge of sources. In this paper, a simple blind source separation method of cyclostationary signals based on the properties of forth-order cyclic cumulant is proposed. In the case of two mixed cyclostationary signals, firstly the observed signal matrix is cyclic whitened in order to change the cyclic autocorrelation matrix to a unit matrix. Thus the separation matrix is turned into a unitary, which can be described by one parameter. Then the optimal value of this parameter can be achieved by the judge function based on the characteristic of cyclic statistics, through which the separating matrix can be determined. This paper makes simulations of mixed BPSK signal and AM signal separately, and the analysis of the result figure of the signal separation, parameter determining and crosstalk error show the effectiveness of the proposed method. Besides, the proposed method is compared with natural gradient algorithm and cyclostationary natural gradient algorithm, and shows great advantages specially in the case of AM signals. At last, the run time of the proposed method is discussed.