Study On Blind Source Separation Algorithm

Blind source separation (BSS) aims to extract independent signals from their linear mixtures captured by a number of sensors without knowing the channel information. Owe to its broad prospect of application in the fields of aerial array signal processing, multiuser's communication, the speech and medical signal processing, this technique has been a focus in signal processing academic recently. The blind source separation algorithm is composed of the off-line batch processing method and the on-line self-adaptive processing method.For the sake of the fact that the off-line batch processing method takes full use of all the samples, the off-line method is superior to the on-line method in separate precision. However, this method gets a limited separate precision in the case of super-gauss and sub- gauss mixed source and insufficient quantity of samples. Some researches about the Fast ICA algorithm have been done in this paper, exactly in regard to improving the flexibility about any source distribution and the precision when few samples can be used.The uncertainty and basic assumption of ICA are analyzed. The measurement criterion of independence and its consistency is studied. Some kinds of optimized algorithm and the performance index are summarized. The simulations on on-line processing and off-line processing at different distribution sources have been run, and the performances are compared afterward.Based on the analysis of the main affective factor of BSS algorithm: step size and active function, a conclusion, the optimum active fuction is the differential coefficient of probability density function (PDF), has been drawn . The PDF estimate methods have been researched, both the advantange and disadvantage are pointed out.Combined with the characteristic of super-gauss and sub-gauss mixed sources situation, a new algorithm of active function estimation has been proposed, which is based on the support vector machine probability density function estimation: do some pre-processing about observed signal first, then estimate the PDF with SVM, and get the sparse expression to create the new active fuction, then applicate it to the Fast ICA algorithm.Comparing with the original algorithm, this algorithm is not merely suitable for the mixed distribution sources separation, but also improves the separate precision under the situation of insufficient quantity of samples.