The Enhancement Of Sound Source Identification Performance Of Beamforming With Spherical Microphone Arrays

Beamforming with microphone arrays,also known as acoustic camera,is a powerful technology that can intuitively visualize sound source distribution based on the sound pressure signals captured by microphones.Among all kinds of microphone arrays,rigid spherical microphone arrays possess good rotation symmetry,high flexibility as well as strong scattering effect and can simultaneously identify sound sources in all directions,therefore they have broad application prospects in the field of sound source identification for the interior noise of vehicle,aircraft and high-speed train.In this paper,a filter and sum（FAS）based CLEAN-SC algorithm system and adaptive reweighting homotopy（ARH）based compressive spherical beamforming（CSB）with spherical microphone arrays are studied.Different from the other deconvolution algorithms,CLEAN-SC achieves sound source identification by iteratively removing sidelobes coherent with mainlobes and does not require point spread function（PSF）,so it can avoid the inconsistency between the actual beam pattern of sound source and the theoretical PSF.Based on CLEAN-SC,high resolution CLEAN-SC（HR-CLEAN-SC）and enhanced high resolution CLEAN-SC（E-HR-CLEAN-SC）select the focus point,at which the beamforming output is mainly contributed by the target source,as source marker,and then reconstruct sound source distribution.The two algorithms can improve spatial resolution.However,CLEAN-SC and its high resolution versions cannot identify coherent sources.CSB throws away the assumption of the source incoherence,hence it is applicable to any sources.The original algorithms used to solve CSB model,₁ norm minimization（₁-CSB）and iterative reweighted ₁ norm minimization（IR-₁-CSB）have to estimate signal-to-noise ratio（SNR）,and their performance is significantly affected by SNR estimation accuracy.Besides,the computational efficiency of the two algorithms is low.To overcome these limitations,ARH algorithm is adapted to CSB,and then ARH-CSB is proposed,which iteratively solves CSB model by determining support set and adaptively searching for weights while estimating the source strength.Simulation and experiments show that:（1）FAS based CLEAN-SC algorithm（FAS-CLEAN-SC）can effectively identify sound sources.Based on FAS-CLEAN-SC,FAS based HR-CLEAN-SC algorithm（FAS-HR-CLEAN-SC）improves spatial resolution,and its performance is prior to the existing spherical harmonics beamforming based HR-CLEAN-SC algorithm.（2）FAS based E-HR-CLEAN-SC algorithm can further improve spatial resolution,but it takes a lot of time to calculate maximum sidelobe level（MSL）and enough space to store MSL.（3）The performance of ₁-CSB and IR-₁-CSB is significantly affected by SNR estimation accuracy:underestimating SNR reduces the estimate of source strength while overestimating SNR leads to an increase in spurious sources.ARH-CSB algorithm does not require the priori SNR knowledge,and it can achieve the same performance as ₁-CSB and IR-₁-CSB when SNR is accurately estimated.（4）Compared with ₁-CSB and IR-₁-CSB,ARH-CSB enjoys better adaptability to low SNR,a higher probability of accurately identifying weak sources,and faster computational speed.