Research On Denoising Beamforming Methods Based On Spectral Matrix Reconstruction And Generalized Inverse Problem For Acoustic Source Identification

Beamforming technology based on microphone array system was an advanced spatial filtering technology,which was widely used to identify acoustics sources in the acoustic design of aerospace,automobile and architecture.Because it is suitable for far-field measurement and sensitive to high frequency sources,measurements are often performing outdoors,in open space or in environments with high frequency noise.However,there are often many kinds of interference noise in these complex environments,which affects the recognition of interest sources.Recently,the diagonal removal method and spectral subtraction method were proposed to suppress self-noise and background noise,but these methods’ performances are not good enough and have obvious drawbacks.To overcome the limitations of these methods,a new method called denoising weighting generalized inverse beamforming method for sound source identification is proposed in this thesis.First of all,based on plane-wave hypothesis,the theory of conventional frequency domain beamforming and its related characteristic parameters are given.The drawbacks of this method are analyzed through simulation experiments,including poor spatial resolution at low frequency,high sidelobe level at high frequency and poor recognition ability of coherent sources.After that,according to the principle of diagonal removal method and spectral subtraction method,the realization process of the algorithm is analyzed in detail.The performance and shortcomings of these methods are simulated and analyzed.These two methods not only have limited de-noising performance,but also directly subtract the cross-spectral matrix,which destroys the semi-positive definite property of the matrix and causes unexpected errors in the reconstructed sound field.Therefore,to overcome the shortcomings in diagonal removal method and spectral subtraction method,the diagonal denoising method and prewhitening method are presented in the thesis.The diagonal denoising method uses convex optimization technique to optimize the diagonal elements of cross-spectral matrix under constraints,and eliminates some self-noise.The cross-spectral matrix of target without background noise is reasonably estimated by constructing pre-whitening vectors and using matrix eigenvalue analysis method.Then,the performance of these methods is analyzed based on simulation experiments.Compared with the diagonal removal,the diagonal denoising method can effectively suppress part of the self-noise and ensure the semi-definite property of the cross-spectral matrix.Compared with spectral subtraction,prewhitening method not only guarantees the properties of cross-spectral matrix,but also improves the performance of suppressing background noise.Furthermore,in order to further improve the noise suppression performance of the above methods.In the thesis,a denoising weighted generalized inverse beamforming method for source identification is proposed.This method combined diagonal denoising method and prewhitening method with an iterative generalized inverse method based on beamforming regularization.The principle of the algorithm is deduced in detail.The selection method of regularization parameters and the factors affecting the performance of sound source identification are analyzed.Three different sound sources models are tested through simulation experiments.The simulation results show that the denoising weighted generalized inverse beamforming method not only has better denoising performance than the traditional denoising beamforming method,diagonal denoising method and "pre-whitening" method,but also has a great improvement in spatial resolution and sidelobe suppression ability.Finally,the feasibility and practicability of the weighted denoising generalized inverse beamforming method are verified by experiments.The experimental results further show that the noise suppression ability of this method is better than the existing beamforming noise reduction technologies.