Sparse deconvolution beamforming with non-negative L1-αL2 regularization for acoustic source localization

This letter proposed a sparse deconvolution localization method (FFT-L1ML2) driven by non-convex L1-αL2 regularization that more closely approximates the ideal L0 norm. It is an alternative that explores the sparse structure of sound sources to enhance localization accuracy, while the original sparse deconvolution beamforming lacks a sufficiently accurate sparse description. An optimization solver composed of forward gradient descent and backward proximal operator is then developed for the FFT-L1ML2 model to reconstruct the beamforming map. Both simulation and experimental results show the effectiveness and superiority of the proposed method in localization accuracy, energy concentration, pseudo source reduction, and computational cost.