Multiple sparse sources separation based on multichannel frequency domain adaptive filtering

Underdetermined sparse sources separation is a challenge problem especially in adverse environment, where there are often some non-sparse interferences or more than one sparse interferences located closely to the target sources. While in some applications, such as in-car or hands-free environments, references of the interferences (P ≥ 2) coming from loudspeakers are available. Common sparse source separation approaches have not yet used these reference information, we call them traditional approaches in this paper. We propose a FD-MENUET (Frequency domain aDaptive filtering based Multiple sENsor degenerate Unmixing Estimation Technique) approach, in which we get full use of those reference information to help to separate the target sources. Even if no reference is available, the approach would only degenerate to the traditional approaches. The experimental results show that the proposed approach is more general and could achieves better separation performance than the traditional one.