Source Localization Based On The Sparse Regularization Methodology

A new source localization method is presented in this thesis. It is different from theexisting classical algorithms, such as multiple signal classification algorithm and theestimation signal parameters via rotational invariance technique algorithm based onsubspace decomposition. The new method converts the source localization problem intoan ill-posed inverse problem from the point of signal overcomplete representation andobtains the accurate location estimation by sparse regularization method.The thesis is divided into four chapters. The first chapter is the introduction part,mainly on the research background, significance and the main work completed in thispaper. The second chapter is the theoretical basis of this paper. After a detaileddescription of source localization problem and sparse regularization methodology, thesource localization algorithms based on sparse regularization method and the commonDOA estimation algorithms are compared. The experiment results indicate that theformer has a comparatively better performance especially in small snapshots even asingle snapshot, low signal to noise ratio and sources-correlated cases. A detaileddescription of the sparse regularization source localization algorithms is given in thethird chapter. After the introduction of a single snapshot processing, source localizationwith multiple snapshots is also considered. As for the division of spatial sampling grid,a multiresolution grid refinement method is introduced. It can reduce computationalcomplexity while improving the estimation accuracy. The selection of regularizationparameter in the sparse regularization source localization algorithms is analysed in thefourth chapter. As for this problem, a new regularization parameter estimation method isproposed. The simulation results demonstrate that the proposed method has manyadvantages, including enhancing resolution, effectively suppressing spurious peaks,improving robustness to noise, as well as increasing the number of resolvable sources,etc.