Research On Underdetermined Blind Signal Separation Technology Based On Independent Component Analysis

Blind source separation of signals refers to the process of separating source signals from time-domain aliased observed signals in the absence of signal prior information.Independent component analysis is one of the most typical methods to achieve blind source separation of signals.According to the quantitative relationship between the source signal and the observed signal,it can be divided into three situations: overdetermined,positive definite and underdetermined;according to different channels,it can be divided into both models are instantaneous hybrid and convolutional hybrid.In practical scenarios,on the one hand,due to the influence of multipath transmission effects,the observation signal is usually a mixture of source signals through a delayattenuation convolution model;on the other hand,the number of source signals is usually greater than the number of observation signals.The situation is called an underdetermined situation in blind source separation.Based on the above two backgrounds,this thesis conducts research on blind source separation technology for 4-input observation signals under positive definite convolution mixture and underdetermined instantaneous mixture models.The specific work is as follows:(1)For the positive definite convolution mixture model,the blind source separation algorithm based on the Canonical Decomposition and Parallel Factor Decomposition model is studied,and an amplitude deblurring method is designed to improve the unmixing performance.Then,it is simulated on the MATLAB platform,and the algorithm is verified for speech signals and AIS signals all have a good unmixing effect.(2)For the underdetermined instantaneous mixture model,the blind source separation algorithm based on sparse component analysis was improved,and the fuzzy C-means clustering with low outlier sensitivity was introduced for mixed matrix estimation,and MATLAB was used for simulation.The experimental results show that the algorithm has good unmixing for sufficiently sparse speech signals effect,but its unmixing performance is not good for non-sparse AIS signals.(3)For the underdetermined instantaneous mixture model,the blind source separation algorithm based on the fourth-order cumulant is studied,and the source signal reconstruction is proposed through the pseudo-inverse operation,and the simulation is carried out with MATLAB.The simulation results show that the unmixing performance of the algorithm is slightly weaker than that based on blind source separation algorithm for sparse component analysis,but the algorithm is superior in terms of computational complexity and unmixing speed.(4)Aiming at the above three blind source separation algorithms,a preprocessing system based on Xilinx KC705 development board was proposed and designed.Based on the systolic array structure and Jacobi algorithm,the system realized the centralization and sphericalization of the observed signals,and solved the problem of independent component analysis that the basic assumption of the method for the source signal is difficult to meet in practical engineering applications.The preprocessing system effectively increases the universality of the independent component analysis method.The 4-input preprocessing system designed in this paper has a system clock frequency of 100 MHz,and a total of 51032 clock cycles are required to process AIS test signals at 16384 points.