Superresolution source localization through data-adaptive regularization

We address the task of source localization using a novel non-parametric data-adaptive approach based on regularized linear inverse problems with sparsity constraints. The class of penalty functions that we use for regularization favors sparsity of the reconstructions, thus producing superb resolution of the sources. We present a computationally efficient technique to carry out the numerical optimization of the resulting cost function. In comparison to conventional source localization methods, the proposed approach provides numerous improvements, including increased resolution, reduced sidelobes, and better robustness properties to noise, limited snapshots, and coherence of the sources. The method is developed for the general source localization scenario, encompassing nearfield and farfield, narrowband and broadband, and non-linear array geometry cases. Simulation results manifest the capabilities of the approach.