Sources Separation of Instantaneous Mixtures Using a Linear Time-Frequency Representation and Vectors Clustering

In this paper, we address the problem of separating N unknown sources using as many observed mixtures. The sources considered here are assumed to be of a non-stationary nature, i.e., their spectral contents are assumed to be time-varying. Using linear time-frequency (TF) representations of the mixtures along with a classification procedure based on vector clustering yield an effective way to separate the sources. Compared to other existing TF based separation methods, the proposed one is characterized by its simplicity and ease of implementation. Moreover, it can be applied in situations where others cannot. Specifically, the algorithm can handle monocomponent as well as multicomponent sources and its assumptions about the mixing matrix are more relaxed than other existing algorithms. Example is presented to prove the validity and efficiency of the proposed algorithms