Sound Signal Blind Separation Based On DSP

This thesis is mainly concerned with the Maximum Entropy algorithm in Blind Source Separation and does some research on the practical application of sound signal blind separation of the algorithm based on digital signal processor.Blind Source Separation (BSS) is to separate individual source signal from a set of mixed signals assuming the source signals are independent. At present, methods such as the Maximum Entropy on the basis of Information Theory, Minimization of Mutual Information and so on are mainly used to solve the BSS problem at home and abroad. These methods are of great application foreground in the area of sound signal blind separation, communication and medical signal processing. On the other hand, with the development of microprocessor technology, digital signal processors (DSP) provide ideal physical platform for the hardware realization of real time digital signal processing, which make it possible for the embedded practical application of BSS algorithms.This thesis firstly investigated deeply into the Maximum Entropy algorithm in Blind Source Separation and proved the feasibility of practical use by the theoretical analysis and simulation experiments of sound signal blind separation. Then a two-dimension sound signal blind separation system whose kernel is DSP was constructed, the Maximum Entropy algorithm based on naturalgradient was embedded into DSP and the practical use of the Maximum Entropy algorithm in the sound signal blind separation was explored. In that system, TMS320DM642EVM of TI Company was used for the hardware system.The experiments show that the system can well separate two dimension mixed sound signals under certain conditions and is robust to weak noise, which agree with the theoretical analysis of the Maximum Entropy algorithm.