Research On Multi-sound Source Direction Finding Method Based On Wide-pitch Microphone Array

Relying on the National Natural Science Foundation of China(61171167,61401203),this thesis focuses on a study of multiple sound source direction measurements based on widely spaced microphone arrays.By measuring mixed-signal information received from microphone array,the multiple sound source direction measurement technology can obtain direction of each source and is widely used in many military and civil areas.For example,estimating the speaker position in video conference accurately can help enhance the signal and control the camera.Reception and processing of gunfire signal can achieve detection of enemy snipers and battlefield situation monitoring.Otherwise,with the ever-increasing requirement of broader bandwidth and higher resolution,the application of widely spaced microphone arrays is an effective way for lower cost and system complexity.Therefore,the study of multiple sound source direction measurements based on widely spaced microphone arrays is of important significance.In this paper,traditional sound source direction measurements are summarized firstly.Due to the advantages of low calculation quantity and high accuracy of direction of arrival(DOA)estimation technique based on time delay estimation,the one with dual-microphone array is mainly discussed in this paper which is one of the most basic issues in this area.Focused on far-field multiple sound source mode,the angular-spectrum-based method is mainly studied based on analysis and comparison of the characteristics and limitations of traditional generalized cross correlation(GCC)and degenerate unmixing estimation technique(DUET).And the pooled angular spectra based on minimum variance distortionless response(MVDR)and diffuse noise model(DNM)are then explained respectively.This kind of method can obtain directions of multiple sources with reverberation,but when half of wavelength is less than the spacing of the two microphones,it is prone to phase wrap-around aliasing due to spatial aliasing which limits the application of widely spaced microphone arrays.In response to these problems,a new pooled angular spectrum is studied.In this method,only the high-valued elements of spectral accumulation outputs are temporally pooled over frames by using the energy accumulation strategy and constant false-alarm rate detection technique,and the height and number of spurious peaks coming from spatial aliasing are significantly lowered.The principle and implementation flow are described in detail and the results of numerous simulations and outdoor experiments demonstrate that this method can suppress spatial aliasing effectively when applied to widely spaced microphone arrays and improve the performance of multiple sound source detection and direction estimation.