Research On Signal Sparsity-based Methods For Underwater High Resolution Direction Finding

To safeguard the national marine rights and interests,improving the capacity of marine re-sources development and building a powerful marine country have become the objectives goal of China’s marine development in the 21 st century.In order to building a powerful marine coun-try,high resolution direction finding technology urgently is needed both in military applications and in civil needs,which can improve the detection ability of underwater targets and the explo-ration ability of underwater resources.Therefore,the research on underwater target direction finding method has important strategic significance for coastal defense security and effective utilization of marine resources.This thesis focuses on the demand of underwater system for high-resolution target DOA estimation,based on sparse representation signal processing method,use different signal wave-forms and linear array structures,a series of theoretical methods of underwater target direction finding based on sparse representation are studied.The effectiveness and feasibility of these theoretical methods are verified by computer simulation and tank experiments,which provides the theoretical foundation for engineering application.Specific innovations can be given as follows:1.To solve the problem of inter-atom interference caused by spatial grid refinement in sparse DOA estimation,a multiple snapshot orthogonal matching pursuit(MOMP)high-resolution DOA method based on inter-atom interference suppression is proposed.This method uses data-driven method to design sensing dictionary,which utilizes the priori information provided by array receiving data,and iteratively designs adaptive sensing dictionary through the idea of minimum variance distortless response,which reduces the inter-atom interference in sensing dictionary and improves the accuracy of MOMP method in selecting correct atoms.Simulation results show that,under high coherence dictionary condition,this method can effectively sup-press the coherence between adjacent atoms,reduce errors in the process of atom selection,and improve the performance of sparse high-resolution target DOA method.2.To improve the performance of DOA estimation under strict non-circular signals,a real-valued high-resolution sparse direction finding algorithm based on subspace weighted mixed norm minimization and Capon spectrum weighted atom nuclear norm minimization is proposed.The algorithm utilizes the non-circular signal characteristic to expand the virtual aperture of the array,which improves the degree of freedom of the array,transforms the array data from com-plex domain to real domain by unitary transformation,and then utilizes singular value decom-position to reduce the matrix dimension,which effectively reduces the number of optimization variables and computational complexity.In the algorithm,subspace weighting strategy and Capon spectral weighted are used to enhance the reliability of sparse target DOA estimation respectively.Simulation results show that,compared with the traditional sparse method,these methods have the better high-resolution DOA estimation performance under different SNR and snapshot numbers,and can effectively estimate the targets under undetermined conditions..3.To solve the grid mismatch problem of static dictionary in sparse target DOA estima-tion,a simultaneous orthogonal matching pursuit(SOMP)sparse high-resolution DOA estima-tion method based on dictionary learning in data domain and a sparse high-resolution DOA estimation method based on Khatri-Rao product dictionary learning in covariance domain are proposed.The first method uses singular value decomposition(SVD)to reduce data dimension and noise,SOMP is used to reconstruct sparse signals,and gradient descent method is used to learn the perturbation parameters of atoms,so as to improve the influence of off-grid atoms on high-resolution DOA estimation performance.The second method combines the covariance ma-trix sparse signal representation model,a forward-backward matching pursuit method is used to estimate the coarse target direction.The gradient descent method is used to learn the per-turbation parameters of Khatri-Rao product dictionary.Two-stage iteration strategy is used to approximate the off-grid error.Simulation results show that,these two methods can effectively improve the performance of high-resolution DOA estimation in off grid condition.4.To solve the problem that the discrete grid can not accurately represent the continuity of target azimuth parameters in sparse DOA estimation,a reduce dimension multi-snapshot gridless high-resolution direction finding method based on atomic norm minimization is proposed.SVD is used to reduce dimension and denoising,estimate the DOA parameters in a continuous param-eter space,and constructs a base dictionary containing infinite atoms for sparse representation of signals.By calculating the Lagrangian form of the primary problem,the dual function of this form is used to transform the primary problem into a semi-definite programming problem.This method improves the performance of high-resolution estimation and reduces the computational complexity by using the joint sparse property of multi snapshot data.Simulation results show that this method is not restricted by the grid size,and can reduce the computational complexity and has excellent high-resolution estimation performance.5.The sparse representation method is further extended to the nonuniform linear array,an iterative weighted nuclear norm minimization covariance matrix reconstruction method and a sparse representation high-resolution DOA estimation based on improved two-dimensional parallel array is proposed.The first method utilizes the low rank,positive semidefinite and vandermonte Teoplitz structure of the array covariance matrix,and reconstructs the noiseless covariance matrix by iteratively weighting the nuclear norm,estimates the target direction by the gridless search method.The second method is based on the improved coprime parallel array,the sparse representation method and the least squares alternating iteration are used to estimate the azimuth and elevation angles of the target.Simulation results show that the two methods can improve the performance of high-resolution DOA estimation by using the high degree of freedom of the array benefited for the nonuniform linear array.6.Pool experiment data is used to validate the sparse high resolution DOA method The sparse high-resolution DOA method based on fine grid under the condition of high coherence dictionary and the sparse high-resolution DOA method based on dictionary learning and atomic norm minimization under the off grid condition are validated and analyzed by the pool experi-ment system.The experimental results show that the proposed method has superior estimation performance in the high coherence condition and sparse high-resolution DOA method based on dictionary learning and atomic norm minimization in the grid mismatch condition.The exper-imental results support the theoretical results of these three methods and lay a foundation for their engineering application.The research results in this thesis have important theoretical and practical significance for improving the performance of underwater multi-target high-resolution DOA estimation,and can be used for reference for other array systems.