Non-line-of-sight sound source localization based on block sparse Bayesian learning and second-order edge diffraction

For environments where the sound source is located directly behind an obstacle, this work proposes a non- line-of-sight sound source localization algorithm based on fast marginalized block sparse Bayesian learning (BSBL-FM) and second-order edge diffraction. The second-order edge diffraction transfer function is calculated using the Biot-Tolstoy-Medwin method and used to construct the sensing matrix. By leveraging the spatial sparsity of the sound source signal, a block sparse measurement model is formulated, and BSBL-FM is applied for sparse reconstruction to achieve high-resolution localization. Simulation results demonstrate that the proposed algorithm accurately locates the source position and identifies the source strength, providing higher spatial resolution than beamforming and more precise source strength identification than matched-field processing (MFP). Experimental results validate the simulation findings and further show that the proposed algorithm achieves smaller localization errors than both beamforming and MFP.