Study Of DOA Estimation Based On ML Algorithm

Spatial spectrum estimation is an important area in array signal processing, it overcome the problem of "Ravleigh limit" in the conventional beam-forming method and have the excellent merits, such as ability to estimate multiple sources at the same time, high resolution ability and so on. As a very important method of Spatial spectrum estimation, The ML direction finding's capability is very excellent, which can work when the incident sources are coherent. And theoretically the ML estimation can get the best capability. Although the Performance of the MLE is the most optimal, it involves the High computational load of the MLE multivariate nonlinear maximization problem. The Computational load of the MLE limits its practical application in the Present chip technology. The main application area of the computational intelligence methods is the hard solved problems in the optimization. Based on this consideration, this thesis is dedicated to the application of the computational intelligence methods such as chaos optimization algorithm, cultural algorithm and their improved algorithms to solve the maximum likelihood localization of multiple sources in the non-cooperative context. Additional this thesis also does some research on how to improve the use efficiency of the array elements.The main work of this thesis is on several DOA estimation methods based on the ML algorithm: (1)ML Direction Finding Method Based on Fourth-Order cumulant (2) ML Direction Finding Method Based on chaos optimization algorithm (3) ML Direction Finding Method Based on cultural algorithm (4) An improved Cultural algorithm used on ML Direction Finding (5) The GML algorithm. We give some simulation results and performance analysis using MATLAB tools. Research show that: Fourth-Order cumulant have the ability of fourth-order array extension which can improve the use the use efficiency of the array elements remarkably ,it can also restrain the Gaussian color noise ,but it can't work when the incident sources are coherent; The GML algorithm, the incident sources may be a mixture of multi-clusters of coherent sources, and the number of the sources can be more than the number of the array elements, but we must have the exact estimation or prior knowledge of the number and structure of the coherent sources, which is a big limit of GML; computational intelligence methods such as chaos optimization algorithm, cultural algorithm and their improved algorithms can reduce the Computational load of the MLE markedly, and the search patterns of them are easy to parallel implementation while it can avoid the local convergence, so it's a very valuable direction in the research.