Research Of Source Separation Algorithm Based On Maximum Signal Noise Ratio And Application In Vibration Signal

Blind source separation (BSS) is a signal processing method,which extracts the source from the observed signal, also it is a new research direction in the area of signal processing. Independent component analysis (ICA) is the most important BSS method, which assumes independence between the original signals. For signals which both time-domain and frequency domain are aliasing, traditional filtering methods can not separated them, when the ICA algorithm to highlight their role. Currently, the ICA algorithm has been widely appreciated at home and abroad, and the study in depth up gradually.Most ICA algorithms has large computation and many iterations, also it is easy to converge to local optimum, global optimum ideas changed the situation, the generation of the algorithm which based on maximum signal noise ratio is derived from this concept, it broke the shackles of the past and turn an optimization process into the eigenvalue problem, through this the problem can be simplified.In this paper, we do research on the algorithm which based on maximum SNR and its application, under the project supported by the National Natural Science Foundation of China (No.50275109), which called Research on BSS Method of Mechanical Vibration and Noise Excitation Signal. The main work is as follows:1) Due to the lack of the criterion which can make overall assessment of the algorithm, so this thesis proposed the matrix 2-norm A quantitative to measure the separation results of the algorithm, in addition the only parameter in this algorithm which called moving average length p are studied. Do experiment with different p values and discuss the impact of the separation results. Summarized the general law of the transfer matrix of p, then select the best p value to optimize this algorithm;2) Do research on the separation effect of blind source separation algorithm which based on the maximum signal noise, first put the algorithm on the three-way mixed non-Gaussian signal and studied the result in this general case; then simulated three-way analog signals generated by the main excitation source, both the environment in no noise or with adding noise are concerned; finally, the separation results are compared with the traditional negative entropy-based Fast-ICA algorithm and the algorithm for global optimization, all these prove the advantage of this algorithm;3) The cylinder head vibration signals obtained in 4135 are studied firstly, which focus on the its cyclostationarity,gaining information on running diesel engines via doing research on diesel engine cylinder pressure signal, then put this algorithm based on the maximum SNR into the separation of cylinder head vibration signals, take all p values to discuss the separation results, and compare the result with the prior information, while the global optimal result is considered as a reference. All this indicated that the maximum SNR algorithm has its superiority.