Research On The Algorithms Of Blind Source Separation

Blind source separation (BBS) is the approach to estimate original source signals using only the information of the mixed signals observed in each input channel. There are many applications of blind source separation such as biomedical science, image processing, wireless communication and speech enhancement. This thesis describes in detail the basic theory of blind source separation, then studies blind separation of linearly mixed sources and convolved speech signals. The main works can be summarized as follows:·In this thesis, a fast on-line algorithm based on negentropy is proposed for blind separation of linearly mixed source signals. The algorithm can be used either for blind separation or blind extraction. An adaptive step-size method is adopted to improve its performance. Simulation results prove its better performance and faster convergence rate than the natural gradient adaptive step-size algorithm.·A frequency-domain blind source separation system is proposed to separate convolved speech signals. The preprocessing method, independent component analysis and post-processing method are combined and optimized to achieve fast and effective action. Experimental results using speech signals recorded in a real room shows its fast speed and good performance.