Abstract—This article addresses the problem of the blindidentification of the mixing matrix in the case of a possiblyunder-determined instantaneous linear mixture of sources.The considered input signals are cyclo-stationary processeswith unknown cyclic frequencies. We propose a new methodconsisting of the application of a particular linear operatoron the correlation matrix of the observations. Then, takingadvantage of the properties of the above transformed matrix,a set of rank-one matrices can be built. Combined witha classification procedure, it makes it possible to estimatethe different columns of the wanted mixing matrix. Thisapproach is also compared with the classical PARAFACdecomposition approach. Finally, computer simulations areprovided in order to illustrate the behavior and the usefulnessof the two proposed approaches in the context of digitalcommunications.

Index Terms—Blind identification, under-determined mixtures,cyclo-stationary signals, second order statistics,PARAFAC decomposition, classification.

Cite: Saloua Rhioui, Nad`ege Thirion-Moreau, and Eric Moreau, "Two Approaches for the Blind Identification of Cyclo-Stationary Signals Mixtures," Journal of Communications, vol. 2, no. 3, pp. 33-42, 2007.