Off-grid deconvolution beamforming for acoustic source identification

Abstract

Deconvolution beamforming treats the output of conventional beamforming as the convolution of source strength distribution and point spread function (PSF), and solves it through different algorithms to eliminate the influence of PSF and obtain clear acoustic source identification imaging results. Accurate construction of the PSF requires not only the transfer vectors between the scanning points and the array microphones, but also between the acoustic sources and the array microphones. In practical applications, since the location of sources are unknown, the existing deconvolution beamforming methods assume that the sources lie on the scanning points. They replace the transfer vectors between the sources and the array microphones with those between the scanning points and the array microphones to construct an approximate PSF. When the acoustic sources lie off the scanning points, the source identification performance of existing deconvolution beamforming methods will degrade. To overcome this problem, this paper proposes an off-grid deconvolution beamforming named as Newtonized orthogonal matching pursuit deconvolution approach for the mapping of acoustic sources (NOMP-DAMAS). The method constructs a PSF with the acoustic source coordinates as variables, and constructs the acoustic source identification problem based on deconvolution beamforming as a maximum likelihood estimation with the acoustic source coordinates and averaged sound pressure contribution as variables. It is solved based on NOMP and continuously updates the PSF through iteration to achieve accurate removal of the PSF, resulting in clear imaging and precise acoustic source identification. Simulations and experiments show that the proposed NOMP-DAMAS has good adaptability to sources lie off the scanning points, can accurately locate the sources and accurately quantify their sound pressure contribution, achieving high-efficiency and high-precision acoustic source identification. Besides, its performance is less affected by grid spacing, noise interference, and source coherence, and the acoustic source identification performance is stable.