Identification of multiple sound sources using GMUSIC

The present paper investigates the spatial array imaging of multiple sound sources. In this paper, MUSIC method is modified to address the problem of imaging resolution for low signal-to-noise ratio (SNR) and small sample. The asymptotic behavior of the subspace including eigenvalue and eigenvector is first analyzed based on the large random matrix theory, and thus the improved method, GMUSIC, is introduced, which is the consistent estimator of MUSIC essentially by adjusting the weight coefficients of the eigenvectors of source subspace. In this paper, the new method to estimate sound source number is proposed to further improve the imaging resolution of GMUSIC by making the two subspaces completely separate. In this paper, we conducted array imaging experiments using multiple signal generators, empirically validating the superior resolution of the proposed method compared to conventional MUSIC in imaging resolution for low SNR and small sample, and the necessity of the proposed source number estimation method is also verified.