A semi-analytical method for blind separation of cyclostationary sources in digital communications

This work puts forward the problem of blind mixing matrix identification in the case of linearly mixed signals of cyclostationay sources whose cyclic frequencies are unknown and different. The identification is achieved using a semi-analytical solution. It takes advantage of the Eigenvalue Decomposition (EVD) of a set of algebraically particular matrices resulted from the application of the cyclic autocorrelation function on the mixed signals and rank-one selection criteria combined with a hierarchical clustering method. The proposed approach is applied to digital communication signals then numerical simulations are provided to illustrate the proper behaviour of the proposed method in different noise contexts.