A Study Of Blind Source Separation Algorithm Based On Neural Networks

Blind source separation (BSS) is a new studying community,which is a combination between the signal processing community and the neural network community. It aimed to separate unknown sources from the observed signals which are mixed with unknown channel. BSS becomes more and more important while used in widely many fields such as biomedical signal analysis and processing (EEG,MEG,ECG) , geothysical data processing, data mining, speech enhancement, image recognition and wireless communications.The major contribution of this paper is summarized as follow:(1) The paper introduces purpose and significance of research on Blind source separation, and analyzes its prospect and derivation, and it summarizes and proves the common objective function of high-order cumulants, the negative entropy, the entropy, mutual information and the likelihood function, as well as every kind of optimized algorithm. And it finally presents standard of judging blind source separation algorithm(2) The paper analyzes the principle and structure of blind source separation algorithm based on neural network. In view of blind source separation algorithm based on the minimum mutual information feed-forward neural network, the paper proposes an improved feed-forward neural network blind source separation algorithm which joins the momentum. The previous time power value adjustment quantity is applied in the current weights value adjustment process; it can effectively prevent the network from falling into local minimum and vibration, and then speed up the convergence rate of weights. The super performance of this algorithm is proved by computer simulation.(3) Introduces principle and the structure of blind source separation algorithm based on the biggest entropy recursion neural network, the paper proposes one kind of blind source separation algorithm based on recursion neural network in the view of recursion nerve network blind source separation algorithm proposed by Matsuoka and so on, the algorithm makes the improvement in the separation network architecture and the optimized algorithm to the original algorithm, and especially when the signal is seriously expanded and contracted, namely hybrid matrix is approximate strange, the separation effect is superb. The good separation performance is proved by computer simulation.(4) Introduces the application of blind source separation algorithm in various fields, in particular it presents preliminary discussion in picture processing and in the bio-medicine signal processing.