A Separation Algorithm for Sources with Temporal Structure Only Using Second-order Statistics
Alternative metrics PlumXhttp://hdl.handle.net/11012/36937
MetadataShow full item record
Unlike conventional blind source separation (BSS) deals with independent identically distributed (i.i.d.) sources, this paper addresses the separation from mixtures of sources with temporal structure, such as linear autocorrelations. Many sequential extraction algorithms have been reported, resulting in inevitable cumulated errors introduced by the deflation scheme. We propose a robust separation algorithm to recover original sources simultaneously, through a joint diagonalizer of several average delayed covariance matrices at positions of the optimal time delay and its integers. The proposed algorithm is computationally simple and efficient, since it is based on the second-order statistics only. Extensive simulation results confirm the validity and high performance of the algorithm. Compared with related extraction algorithms, its separation signal-to-noise rate for a desired source can reach 20dB higher, and it seems rather insensitive to the estimation error of the time delay.
Document typePeer reviewed
SourceRadioengineering. 2013, vol. 22, č. 3, s. 861-865. issn 1210-2512
- 2013/3