Owed Under Defined Conditions Mixed-signal Blind Separation Algorithm

Blind Source Separation has been developed since the late 1990s.A lot of algorithms assumed the number of observation signals is no less than the sources.However, there may exist the underdetermined cases in which the number of the sources is more than the observations. The latest progress about the underdetermined algorithms concentrates on the sparse component, but many signals are not strictly satisfied to the sparseness to estimate the mixing matrix and the sources. So it is significant to study the estimation of the mixing matrix when the sources are not sparse.Single channel blind source separation has made a great progress in audio signals, EEG signals, communication signals and so on. The intrinsic mechanism of these algorithms is based on the diversity of the signal components.So it is important to separate the mix signals using the characteristics of the radar signal and digitial modulated signal.The key work and innovations of this thesis mainly include:1) Systematically studied the related theory about the source separation, such as sparseness theory, JADE algorithm, FPICA algorithm, the error performance indexes between the sources and the estimated sources.2) A method of the mixing matrix estimation in underdetermined source separation is proposed, which is based on the linear clustering of sparse component. Then, the source signals can be recovered by the shortest path method and SSDP method. The experiment shows the validity of the algorithm.3) Studied the searching and averaging algorithm, applied it for the radar signal and digitial modulated signal and proposed using TDOA and the pulse train to meet the sparsity.4) A method of the mixing matrix estimation in the underdetermined source separation is proposed in which the sources are not sparse enough to estimate the mixing matrix. The experiment shows,the method have better accuracy and validity than K-means and searching-and-averaging method in the time domain in estimating the mixing matrix.5) Studied two single channel blind source separation algorithms based on carrier phases and Polynomial Phase signals, sufficiently exploring the prior knowedge and the diversity between the signal components to separate the mix signals.