Research On Independent Component Analysis And Its Application In Array Signal Processing

Independent Component Analysis (ICA), a powerful technique for blind signal processing, is developed well recently. It is able to recover or estimate the sources from observations without knowledge of sources or mixing system. The basic assumption of ICA refers to that the sources are mutually statistically independent, which is not a strict requirement, thus resulting in a wide range of applications of ICA in many fields. The narrowband array model in array signal processing is consistent with the basic model in ICA, so it is a natural choice to deal with the receiving signals of array using ICA technique when it's difficult to get transmission modeling and prior information. Based on the requirement of application in array signal processing system, this paper conductes a study of ICA on direction of arrival (DOA) estimation and adaptive beamforming, and realizes selected algorithm on chip ADSP-TS201. The main content is as following:1. As the basis of follow-up processing, ICA algorithm has been deeply studied in this paper. Classical JADE algorithm and FastICA algorithm are introduced, and a RobustICA algorithm is brought forward in order to get more robust performance, which is able to converge to the global extremum and has good separation effect particularly for few samples. Combining with the performance indexes the performance of above three algorithms are validated in signal separation and array signal processing, and the ICA algorithm shows good separation efficiency even in the existence of array gain-phase errors or frequency-close signals.2. The research is also related to the application of ICA in DOA estimation. Because the beam-space DOA estimation algorithms are based on the acknowledge of signal direction and at present nearly all the beam-space DOA estimation algorithms are designed for linear array, a beam-space DOA estimation based on ICA is proposed, which is fit for any array and better adapt to situation of large array and small source number situation without knowing the approximate directions of signals. To solve the problem that DOA estimation of uniform circular array (UCA) with superior characteristics needs spatial spectrum searching with high computational complexity, a DOA estimation algorithm based on ICA for circular array is put forward, which provides the closed-form solutions of signal direction avoiding the spatial spectrum searching and mainly studies the period revising method through the research of UCA's manifold to eliminate DOA estimation ambiguity, and the new algorithm has comparative performance to MUSIC algorithm while the computational complexity is only one of fortieth compared with MUSIC algorithm.3. Tqhe adaptive beamforming based on ICA is studied. As to the problem of constant modulus algorithm cannot converge to desired signal in strong jamming, a blind beamforming algorithm based on ICA and angular constraint is presented to solve it. This algorithm converts the optimization problem of non-Gaussian objective function to second-order cone programming model, which is solved using interior point method and alternative iterative operation, and it is able to converge to desired signal in strong jamming directly without separating all array receiving signals. Aiming at the performance of almost all the beamforming algorithms decrease when closely spaced signals impinge on array, a high angular resolution blind separation algorithm is brought forward. An array extrapolation way based on auto-regressive model which fits for all arrays is derivated first, and then ICA algorithm is used to separate the signals. This algotithm combines the strong points of array extrapolation with ICA,and is able to separate the signals as close as 1 degree under the precondition of interference rejecting ability.4. The DOA estimation algorithm of uniform circular array based on ICA and high angular resolution beamforming algorithm are implemented on chip ADSP-TS201. According to the system requirements of practical system we carry out a series of program optimization, and realize above algorithms on chip ADSP-TS201 fast and efficiently.