The Algorithms Research Based On Blind Source Separation Of Signal Processing

Blind source separation problem is one of the most popular emerging technologies in signal analysis and processing field in the current era. Blind source separation is under the condition of source signal and the transmission network are unknown, only according to the received mixed signals, under the conditions of statistical independence, the target source signals are separated from the mixing signals. Blind source separation has been widely used in signal processing such as voice signal processing, wireless communications, pattern recognition and image processing, biomedical signal processing, geophysical data processing, array signal processing, data mining, facial recognition and other fields.First, this paper introduces the development status of blind source separation at domestic and international, the basic principles of blind source separation, mathematical models, the purpose and significance, the solution process, pre-treatment methods and evaluation. Then, describes the basic theoretical knowledge of blind source separation such as probability knowledge, statistical knowledge and information theory and other related fields. Finally, focus on the application of blind source separation in the natural gradient algorithm and the fourth-order cumulants algorithm. The solving steps of blind source separation are as follows:First, pre-whitening the signal is to remove the correlation between different signals. Then, determine an objective function, usually separating matrix setting the objective function as the dependent variable. Finally, choose a learning algorithm is to optimize the objective function, such as natural gradient algorithm or fourth-order cumulants algorithm. Through MATLAB simulation program software to achieve the blind source separation process, use the similarity coefficients and power spectral density separation before and after the comparison, that proved the feasibility of the algorithm. In the process of equipment condition monitoring and fault diagnosis, use blind source separation technology, that can extract the fault feature signal for precise positioning in mechanical failure effectively.Blind source separation method has some applications in the mechanical vibration signal analysis and fault diagnosis, but has not yet become a very mature technology. Currently, the major domestic and foreign research is to explore a variety of separation methods and algorithms for signal separation and engineering applications.