Applied Research Of Underdetermined Blind Source Separation Method On Phonocardiogram Mixed Signal

Sounds rarely occur in isolation. Speech signals, physiological sounds signals, or other sound signals, more or less will be mixed with other sound signals (or noise). Despite this, some significant applications are in need of a pure sound, such as speech recognizers, Phonocardiogram analysis, Phonocardiogram recognizers and so on. In order to extend such systems to operate on mixtures of many sources, the ability to recover the source signals from the mixture is required. This process is known as source separation.This thesis focus on the problem of underdetermined blind source separation where the number of sources is greater than the number of channels in the observed mixture, especially that the number of channels is 1 or 2. It is often necessary for a separation algorithm to utilize prior information about the sources present in the mixture to constrain possible source reconstructions. This paper uses Phonocardiogram's sparse characteristic to separate such mixture signals.For the beginning, with a single channel mixed signal BSS method and the periodicity of Phonocardiogram, it presents a single channel-single period Phonocardiogram mixed signal underdetermined blind source separation method which based on dwt_ica. First, this method uses dwt_ica getting independent Phonocardiogram sun-band. Then putting the independent Phonocardiogram sun-band into the single channel-single period Phonocardiogram mixed signal to make one mixed signal turn into multi-signals, so the Phonocardiogram signal could be separated by using ICA again. At last, it uses 2 separation experiments on the Phonocardiogram signal proved the feasibility of the method.Then this paper presents an underdetermined blind source separation method for the dual-channel heart sound signal based on a special test platform of two proprietary human voice signals. After depth discussing about underdetermined blind source separation and sparse component analysis, firstly it cluster calculates about the data points in the frequency domain, and analyzes the number of source signals to achieve the estimated mixing matrix by using the sparse signal characteristics shown in the observed signals. Then according to the scatter plot of observed signals, one or multiple source signals are separated, so that the observed signals turn into well-posed or over determined signals. Lastly the remaining source signals are separated by using traditional well-posed or over determined blind source separation method. The validity and the feasibility of this method are confirmed through the separation experiment on the artificial mixed-signals and the actual detected two-channel voice signals respectively.