location: Current position: Home >> Scientific Research >> Paper Publications

Blind identification of underdetermined mixing matrix and source separation by finding and solving a row echelon-like form of system in the time-frequency domain

Hits:

Indexed by:期刊论文

Date of Publication:2013-06-01

Journal:ICIC Express Letters, Part B: Applications

Included Journals:EI、Scopus

Volume:4

Issue:3

Page Number:739-746

ISSN No.:21852766

Abstract:This paper presents a novel two step underdetermined blind source separation approach that can apply for non-disjointed source signals. First, the single-sourcepoints (SSPs), each of which is occupied by a single source, are detected in the mixtures; the mixing matrix is then estimated accurately by employing K-means algorithm among those SSPs. In the separation procedure, the time-frequency points that incorporate one source, two sources, and so on, are found out, sequentially, so that they construct a row echelon-like form of system. Then, these sources at the points can be solved out explicitly under weak assumptions. One of the advantages is that algorithm does not rely on the non-stationarity, independence or the non-Gaussianity, as in the conventional ICA algorithms. Experimental results indicate the validity of the method. ? 2013 ISSN 2185-2766.

Pre One:An improved image inpainting algorithm based on multi-scale dictionary learning in wavelet domain

Next One:Detection of audio interpolation based on singular value decomposition