Researches On Fast Super-resolution DOA Estimation

Array signal processing allows the electronic surveillance and civil communicationsystems with excellent performance, but there has also been a prominent problem ofhigh algorithm complexity and slow processing speed, especially for multidimensionalparameter estimation. The existing spatial spectrum algorithms can not be applied to theactual engineering environment because of high computational complexity, so it isnecessary to do research on fast direction of arrival (DOA) estimation algorithms. Thispaper devotes to improve the computing speed of the algorithms as much as possibleunder the condition of appropriate precision. We make an in-depth study onone-dimensional (1-D) and two-dimensional (2-D) fast subspace-based approaches, andobtain some valuable results. The main contents are as follows:(1) We make research and analysis on the classical subspace-based algorithmtheory and its computation complexity, and then propose two ideas to carry out fastDOA estimation. Firstly, we can replace the high-dimensional spectrum peak searchingwith low-dimensional search or the closed-form solution of the DOA. Secondly, we canachieve high-precision fast DOA estimation by use of the practical computationalmathematics, getting rid of the original eigenvalue decomposition (EVD) bound fordirection finding speed.(2) We propose a new automatic pairing approach to high resolution estimation of2-D DOA. By making full use of the new defined L-shaped array geometry and thecross-correlation matrix, the effect of additive noise is eliminated. The proposed methodonly needs1-D estimation for azimuth and elevation angle estimation. The azimuth andelevation angles estimated by the proposed method can pair automatically, which canreduce the computational burden. Simulation results show that the proposed methodperforms better than the CCM-ESPRIT, and it is shown that the proposed methodovercomes the failure of angle estimation and the pair-matching problem. In addition,the proposed method can also perform well even in the situation of low SNR andsmall angular separation.(3) We propose a new automatic pairing method for2-D DOA based on EVD. By making full use of the cross-correlation matrix, the effect of the additive noise iseliminated. By combining the advantages of ESPRIT and propagator methods, theproposed algorithm avoids spectral peak searching, and simply needs the EVD of alow-dimension matrix, which can sharply reduce the computational burden. What’smore, the elevation angles are estimated by the eigenvalues, and the correspondingazimuth angles are estimated through the use of the virtual array response matrixconstructed by the corresponding eigenvectors. The azimuth and elevation anglesestimated by the proposed method can pair automatically. The effectiveness of theproposed method is verified through a lot of simulations, and it is shown that theproposed method performs well in the situation of low SNR, small snapshots and littleangular separation. In a word, the proposed method well coordinates the computationalcomplexity and DOA estimation performance, so it has a high practical value.(4) We make an in-depth study on fast high-resolution1-D DOA estimationalgorithms. By analysis of the computational complexity and lots of computersimulations, we draw a practical conclusion of fast1-D DOA estimation algorithms:when using fast1-D DOA estimation in2-D space, if the incident source wave is known,it is recommended to use the MSWF algorithm to estimate DOA quickly; however, ifwe do not know the incident signal waveform, it is recommended to use the PMalgorithm to achieve fast DOA estimation.(5) Finally, we carry out an in-depth research on fast high-resolution2-D DOAestimation algorithms. By analysis of the array geometry, fast2-D DOA estimationalgorithms and pair-matching algorithms respectively, we draw some practicalconclusions of fast2-D DOA estimation algorithms: when using fast2-D DOAestimation in three-dimensional (3-D) space, if it simply requires that the amount ofcomputation is as small as possible but ignores the angular estimation performance, werecommend to use the JADE CCM algorithm; however, if it needs low computationalcomplexity as well as high accuracy, it is recommended to use the second methodproposed in this article, which has a high practical value.