Compressed Sensing-based Blind Signal Processing Technology Research

With the fast development of information and computer technologies, people are in need of the information processing. In practice, the signal attained from the sensor often contains interference and noise. How to extract the source signal from the mixed signals is inevitably a problem, and blind source separation (BSS) is a good choice to solve the problem. BSS is related to many subjects such as neural network, information theory and signal processing. Meanwhile, the application of BSS technology is wide. For example, it can be applied in biomedicine, image processing, voice recognition, and communication and radar systems.Generally,the blind source separation algorithm is based on the assumption that source signal and hybrid model are unknown. The source signal is extracted according to a certain amount of prior information. To separate the source signal, it is usually required to satisfy the requirements that all the source signals are statistically independent. When the number of observed signal is less than that of the source signals, it is called underdetermined blind source separation. Under this condition, it is more difficult and complex to separate the source signals successfully. Presently,the main solution of underdetermined blind source separation is based on the sparseness of the signal and statistical probability model.Compressive sensing theory appeared in 2006, and it broke the bottleneck of Nyquist traditional sampling theorem, which made it possible to recover the source signal completely under the condition that the source signal meets the requirements of sparseness.The present situation of the blind source separation technology and compressive sensing theory is introduced firstly, and then the relationship between them is analyzed. Based on the above theories, we put forward a new cost function of BSS by modifying the cost function of traditional blind source separation, and the advantage of the new cost function is proved by simulation. In chapter 4, we use compressive sensing theory to solve the problem of blind source separation under underdetermined condition. In chapter 5, we use the previous method to solve a practical problem, which is blind channel identification.To sum up, we investigate blind source separation deeply in this dissertation. On one hand, we made a modification on the cost function of standard blind source separation algorithm to improve the performance of BSS. On the other hand, we apply compressive sensing theory to solve the problem of blind source separation under underdetermined condition. Furthermore, we combine the proposed method with practical application such as blind system identification. At the end of the dissertation, we point out the future work in this field.