Blind de-mixing real-time algorithm of piecewise time series mixture

Lots of attention has been paid to the instantaneous blind source separation problem (BSS) configuration. However, a piecewise time series model of the BSS is an advance configuration of it. An aim of our control study was to determine if a high-order statistic can detect in real-time a piecewise time change occurrence in the mixing matrix in BSS problem. A preprocess and a postprocess were incorporated with a conventional ICA. The objective of preprocess is to detect the change and of the postprocess is to track over consistency of the permutation and phase shift. Those were achieved by using high-order statistic-kurtosis values. The method was applied to a Laplacian PDF (a synthesised speech) signal and Gaussian PDF note signal which mixed with a mixing matrix having a jump in its intensity and phase values. Results indicated that it could be used in real-time BSS problem when a sudden change is occurred in the recording transmission path and it is formulated in jumps in the entries of the mixing matrix. Still, the algorithm must be improved in order to deal with a slight change in the mixing matrix