Research Of Fast And Robust Estimation Of Signal Parameter And Beamforming

Array signal processing is an important research field of modern signal processing, which finds widely applications in many fields, such as radar, sonar, navigation, communication, image, seismology and biomedicine engineering, etc. So the study of array signal processing is of great theoretical and practical value. Based on the reasons above, this thesis mainly focuses on the issues of the parameter estimation and beamforming in the array signal processing system, which aims at the development of fast and effective signal subspace estimation method in different noise fields, and building fast and robust beamforming technique.The main contents of the dissertation are described as following:(1) For the problem of signal leakage into noise subspace using Multi-stage wiener filter (MSWF) technique to estimate the signal subspace, this paper presents a novel subspace estimation algorithm. This method does not require the training signal and pre-knowledge of source number. Then, a post detection method is proposed for the source number estimation. And the source number detection and DOA estimation can be simultaneously completed. The whole algorithm does not need computing covariance matrix and eigen-decomposition, so the complexity is greatly decreased. Compared with the classical eigen-value decomposition (EVD) based methods, the procedure have less computation complexity with the approximation performance in the simulation of uniform linear array and uncorrelated signals. The simulation results demonstrate its effectiveness.(2) For fast DOA estimation in spatial correlated noise field, temporal correlation information is introduced into the MSWF technique. A multistage wiener decomposition based on spatio-temporal correlation is proposed to estimate the DOA of signals in spatial correlated noise fields. Moreover, to deal with coherent sources in DOA estimation, a fast multistage wiener subspace estimation algorithm based on front-back spatial smoothing is given.(3) A low complexity ESPRIT algorithm based on power method and QR decomposition is presented for direction finding, which doesn't require the priori knowledge of sources number and the predetermined threshold in separation of the signal and noise eigen-values. Firstly, the estimation of noise subspace is obtained by the power of covariance matrix and a novel source number detection method without eigen-decomposition is proposed based on QR decomposition. Furthermore, the eigen-vectors of signal subspace can be determined according to Q matrix, and then the directions of signals could be computed by the ESPRIT algorithm. In determining the source-number and subspace, the proposed algorithm has a substantial computational saving with the approximation performance compared with the Single-Vector-Decomposition (SVD) based algorithm.(4) A fast and robust GSC beamforming algorithm based on Multistage Wiener filter technique is proposed. Firstly, according to the multistage wiener decomposition, the equation of diagonal loading level is derived, through which the loading level can be accurately determined. In conjunction with the presented match filter loading method, the efficient technique achieves the advantages of robust capabilities. The algorithm does not requiring computing the covariance matrix and its eigen-decomposition, so the computation load is greatly saved.(5) A novel robust beamforming algorithm based on adaptively parameter selection is proposed. It overcomes the drawback of fixed error parameter selection, which affects the performance of robust beamforming algorithm. According to the double constraint worst-case performance optimization problem, the equation of diagonal loading level is derived, through which the loading level can be accurately determined. Analyzed the feature of loading equation, a method of adaptively parameter determination is presented. Simulations demonstrate the effectiveness and robustness of proposed method, compared with the loading based and eigenspace based robust beamforming algorithms, respectively.(6) A practical unsupervised source number detection technique is presented, which is called unsupervised eigenvalue clustering detector (UECD) algorithm. This algorithm is insensitive to the number of snapshots and different noise models, especially to the non-uniform noise. Simulation results demonstrate the effectiveness of the UECD method in both Gaussian white noise and spatially correlated noise. But in spatially correlated noise, especially in strong correlated noise field, the performance of our method is degraded, which is one aspect of the future study. In general, the UECD method can deal with various noise fields. So it is a good algorithm for practical applications of unknown noise field.