Study On Underwater Acoustic Signal Blind Source Separation With Unknown Source Number

The blind source separation (BSS) is to separate each independent component of the source signal only by the statistical characters of objected signal, without any information about the channel parameters and the source signals. When the source numbers is unknown or dynamic, it is hard for BBS to deal with this case. In the past, the assumption is often held that the source numbers are known and the source signals are mutually independent. However, such assumption can not be held in many advanced applications. Thus aiming at this point, the BBS algorithm is studied deeply when the source numbers are dynamic. The key parts of the study are described as follow:Firstly, the BBS algorithm is summarized. Both the assumption conditions and the model of the BBS are also studied in detail. The knowledge of the information theory is emphasized. And the pretreatment procedure of the objected signal is also discussed deeply, including the signal zero-mean processing and the signal whitening procedure. At the end, how to calculate the cost function of the BSS algorithm is explained fully detailed.Secondly, when the source numbers are unknown, the BBS algorithm is studied. The right estimation of the sour numbers is the basement the following separation of the source signals correctly. The maximized the covariance PCA approach is used to find the principal components using the eigenvalue decomposion method. To estimate the source numbers, then the criterion of information is used to set the threshold of the eigenvalue. On that basis, Infomax is used to achieve the source signal separation. A number of simulations are designed to analyze the effection of the different sensors signal separation, under different information criterion and signal noise ratio. However, it is not an on-line algorithm, and it is not fitted the dynamic source number cases.Thirdly, when the source numbers are dynamic, the adaptive neural algorithm is proposed, which designs and which designs and associates several auto-adjust mechanisms to challenge these advanced BSS problems. The first implementation is the on-line estimator of source numbers improved from the cross-validation technique. The second is the adaptive structure neural network that combines feed-forward architecture and the self-organized criterion. The last is the learning rate adjustment in order to enhance efficiency of learning. The validity and performance of the proposed algorithm are demonstrated by computer simulations, and are compared to other static algorithms. From the simulation results, these have been confirmed that the proposed algorithm performed better separation than others in static BSS cases and is feasible for dynamic BSS cases.