The Research Of Blind Deconvolution For Mixing And Convolution Signal

Blind signal separation (BSS) is an important topic, it also has used in many applications. Blind sources separation is an emerging technique of array signal processing and analysis, which aims to recover the unobserved signals or sources from the observed mixtures. It only uses some hypothesis of statistic characteristic based on the source signals, so it adapts to the situations when any prior information about the source signal can not be acquired.Till now, BBS has two linear models: one is instantaneous model, the other is convolution model. Instantaneous one is an ideal model; it does not take the delay time of signal in transmission into account.The realization of BSS includes object criterion and optimizing algorithm. The object criterion is based on many theories about information, including Information Maximization, Minimizing Mutual Information and Maximization likelihood, which are all equivalent in most situations. The optimizing algorithm we offen used is gradient descent.Algorithm of instantaneous model can accurately separate signal, but the effects of the algorithm based on convolution model is not ideal.In this thesis, the principles and algorithms of blind source separation are discussed. Firstly, some object criterions are introduced, which are based on the Central Limit Theorem and independence. Secondly, we discusse the algorithm of instantaneous model. Finally, we deduce the Information Maximization algorithm of convolution model, and summarize the defect of convolution model algorithm. We present a new algorithm of decorrelation in the frequency domain based on convolution model, and apply it to the real-time speech signal separation. The simulation results show that the algorithm is effective.