The Research Of Underdetermined Blind Source Separation Under Noise Environment

In the actual scenario, under the influence of the acoustic environment of communication channels, the signals we captured are usually one or several mixtures, or a source contaminated by noise, which deteriorates the quality of speech or communication, making us fail to obtain some valuable information from captured signals. Blind source separation (BSS) is such a technique to solve the separation of the sources from the mixtures, with the mixing procedure and the original source signals unknown to us. Therefore, this technique has been widely applied in the fields such as speech processing, digital communication, image processing, biomedical engineering and radar system, and has also become one of the hotspots in recent years.This paper focus on the basic theory and main algorithms of the underdetermined blind source separation problem under environment noise. The detailed research content mainly consists of the following aspects:(1) Underdetermined blind source separation (UBSS) is a more challenging case in BSS, where the number of sources is more than the mixtures. A layer-by-layer UBSS algorithm based on matrix transformation is proposed in the paper. The proposed method firstly identify the single source points in each layer through constructing separating matrices, then solve the mixing points dominated by all the sources via the idea of pseudo inverse, and furthermore recover the source signals with the help of masking matrices. The proposed method could separate each source signal well, and improve the quality of the recovered signal without too many constraints on the sparsity of the sources.(2) Binary image separation method based on support vector machine (SVM). To tackle the binary image separation under the environment noise, in this paper a binary image separation method via supervised learning is explored. The proposed method uses the mixture data in training set to train the SVM classifier, and then use the trained classifier to classify the noisy mixing image. The proposed could solve the nonlinear UBSS of binary images under noise environment well with low bit error rate.(3) A binary image blind separation method based on maximum a posterior (MAP) is proposed in this paper. The proposed method utilizes the idea of signal estimation and subset adjusting strategy to estimate the nonlinear mixing function, and then separate the source signals layer by layer. The proposed method explored the nature prior of the signals in local area rather than constrained or assumed the distribution of the signals. Comparing with other state-of-art method, the proposed method could achieve better performance in separating binary images, even with the dependent source images, which the comparing method fails to separate. Besides, the proposed method also has other advantages like robust to environmental noise, low time consumption, and so on.Large quantity of computer simulate experiments are conducted to evaluate the proposed method and compared with other state-of-art method. Experimental results show that the proposed method approaches well performance on the UBSS under noise case.