Blind source separation using time-frequency concepts

In this paper two methods are described for blind signal separation (BSS) based on time frequency distributions TFDs of the received mixed signals. Consequently, they are most suitable to deal with nonstationary mixed signals. The first method determines the decoupling system through setting to zero coefficients of the ambiguity function of the received mixed signals at the edges of its distribution plane. Computations have revealed that this approach is only successful when their TFD signatures, are partially overlapped When the TFD signatures are highly overlapped, an alternate approach is described that determines the elements of the decoupling system through decorrelating the respective rows of the received TFD data matrices. Illustrative examples are given showing that the proposed approaches work when other approaches based on higher order statistics HOS fail to yield satisfactory results.