Extraction of a cyclostationary source using a new cost function without pre-whitening

The proposed method aims to extract a cyclostationary source, whose cyclic frequency is a priori known, from a set of additive mixtures. The other sources may be either stationary or cyclostationary as long as their cyclic frequencies are different from that of the source to be extracted. The method does not require pre-whitening and consists in minimizing a criterion based on stationary and cyclostationary second order statistics of the observations; this method is labeled as Second Order Cyclostationary Statistics Optimization Criterion (SOC2). The relevance of this criterion is proven theoretically in the general case of N sources by P sensors, with P≥N. Other properties of the algorithm such as its accuracy and its robustness against additive noise or strong interferences are studied through a set of simulations.

► We propose a new method for the extraction of a cyclostationary source.
► The method is based on simple matrix calculations.
► The method is compared to equivalent algorithms and leads to better extraction.