Research On Adaptive BSS Algorithms And Blind MUD Algorithms

Blind Signal Processing (BSP) is the combination of Artificial Neural Nets (ANN), Statistical Signal Processing (SSP) and Signal Theory (ST), which posses powerful ability of information processing. As the BSP theory cover a lot of subjects and with wide applications, it has attached great importance both in domestic and abroad. BSP has a wide application in a variety of fields, ranging from radio communication, electronic reconnaissance, biomedical signal processing, array signal processing to auto recognizing. As two important branches of BSP, Bind Signal Separation (BSS) and Multi-User Detection (MUD) become the hottest topics in the fields of signal processing and ANN.This paper is divided into two parts mainly: adaptive Blind Signal Separation and Multi-User Detection techonologies. First, we have realized the conflict beteween convergence speed and steady performance based on the analysis of traditonal BSS algorithms, and propose the necessity of application of variable step-size algorithms. Secondly, MUD is a key technique to solve the problem of Multiple Acess Interference (MAI) in the field of communication, but the spreading sequences must be known in the traditonal MUD techonologies. That makes the multi-user detection impractical. The main tasks of the thesis include:(1) Introduces several classic adaptive BSS algorithms, this include: gradient algorithm, EASI algorithm, decor-relating algorithm and iterative-inverse algorithm. Under the frame of natural gradient algorithm, the convergence and stability are simulated. the silmulation experiments verify that: convergence speed and steady performance of traditional adaptive BSS algorithms can not achieve the finest combination. The simulation results of the four algorithms have also been dsicussed.(2) Taking the defect of traditonal BSS algorithms into consideration, a new variable step-size EASI BSS algorithm based on performance index PI has been proposed. when the separation result with low accuracy, the step-size is relatively larger; on the contrary, accuracy increses, the step-size decreases. In this way, performance index PI which represent the separation accuracy can be used to control step-size.(3) Introduces several often-used MUD algorithms, this include: MOE multi-user detection algorithm, CMA multi-user detection algorithm and Kalman filtering multi-user detection algorithm. Limitations of these algorithms in practical are pointed out. Analysis of the models of BSS and DS-CDMA revealed that two models are coherent. That is to say, the problems of MUD can be resolved by BSS algorithms without known the spreading sequences.(4) Taking the limitations of traditional MUD algorithms into consideration, a subspace multi-user detection method are proposed, which is based on MUSIC algorithm. This method can detect the active users'spreading codes, information codes and power, which cast the restrain mentioned above aside.The simulation experiments verify that: the improved variable step-size BSS algorithm is superior to traditonal BSS algorithm, MUD method based on BSS algorithm is operative, the new subspace MUD method based on MUSIC algorithm is operative. At the same time, these two MUD algorithms which are used in complete blind environment provide new methods for spreading codes estimation blindly.