Study On The Hybrid Time-Frequency Domain Approach For Rotating Acoustic Source Identification

Beamforming method based on the microphone array has many advantages such as intuitive visualization,high efficiency,suitable for the medium and high frequency noise,able to identify the moving acoustic sources,and measurement for medium and long distance.It is widely used for the acoustic sources identification in automobiles,aero engines,aircrafts,etc.The beamforming method for identifying rotating acoustic sources is an important branch of beamforming.Among them,the hybrid time-frequency approach which can adapt to different array forms has practicality in identifying rotating acoustic sources.In this study,the hybrid time-frequency approach based on the Deconvolution Approach for the Mapping of Acoustic Sources(DAMAS)and Nonnegative Least Squares(NNLS)is analyzed,and their disadvantages are also demonstrated.This study presents two novel hybrid time-frequency domain approaches to overcome disadvantages of the above method.Firstly,the Doppler Effect caused by the rotating acoustic source is verified.By comparing and analyzing the conventional time domain delay and sum(DAS)and time domain tracking DAS,it is verified that the time domain tracking DAS can effectively identify the rotating acoustic sources.The DAMAS-and NNLS-based hybrid timefrequency approaches are analyzed at different sampling rates.It is proved that DAMASbased approach is sensitive to the sampling rates,while NNLS-based approach is low efficiency.The impact of the sampling rate on hybrid time-frequency approach is illuminated.The higher sampling rate is,the smaller error of the time delay estimation is,the higher accuracy of the time domain DAS is,and the better convergence of the hybrid time-frequency approach is.Then,to decrease the demand of the sampling rates and obtain good acoustic source identification performance with high efficiency at a low sampling rate,the Fast Iterative Shrinkage-Thresholding Algorithm(FISTA)is used for solving the hybrid time-frequency problem in consideration of the spatial sparse distribution characteristic of main sound sources.Compared with DAMAS and NNLS,the results show that FISTA is insensitive to sampling rate and has better comprehensive performance.Furthermore,a split variable is introduced into the deconvolution model of acoustic source identification,and the Split Augmented Lagrangian Shrinkage(SALSA)deconvolution model is established,and then the model is iteratively solved by alternating direction method.The comparison between SALSA and FISTA shows that SALSA method has better performance,higher efficiency and faster convergence.Finally,the test of dual acoustic sources with different intensity is carried out.The results show that the comprehensive performance of SALSA is the best,FISTA is second,DAMAS and NNLS are general,and the conclusion is consistent with the simulation analysis.