Research On Sound Source Location Based On Beamforming Algorithm

In recent years,human-computer interaction,car whistle capture,audio and video conference,smart home and other emerging fields have developed vigorously.As one of its core technologies,sound source location method has become a research hotspot.The sound source location technology based on beamforming is to scan the beam in the whole space domain.The direction with the largest energy in the scanning area is the direction of the sound source.In practical application,the mismatch of algorithm model,the error of steer vector and the small number of snapshots will seriously affect the performance of beamforming algorithm,and then affect the accuracy of sound source location.To solve the above problems,this thesis carries out the research on sound source location technology based on beamforming algorithm.The main innovations include the following aspects:(1)To address the performance degradation of adaptive beamforming algorithm in the case of model mismatch,a robust adaptive beamforming algorithm based on spatial smooth difference is proposed.The core idea of this method is to remove the expected signal component from the covariance matrix,which can avoid the "selfcancellation" phenomenon of the expected signal as interference when there is an error in the steer vector.At the same time,the spatially smoothed difference algorithm is introduced in the source number estimation method,which divides the source number estimation into two parts.Firstly,the non-coherent source number is estimated,and then the information related to the non-coherent sources in the covariance matrix is removed by the spatial smoothing algorithm to obtain the spatially smoothed difference matrix containing only the coherent source information,and finally,the estimated coherent source number is obtained.Simulation results show that the improved algorithm has better performance even when the steer vector is biased,and also has better robustness under different input signal-to-noise ratios and snap numbers,and further demonstrates experimentally that the improved algorithm has improved sound source localization performance.(2)Aiming at the performance degradation of adaptive beamforming algorithm in low snapshot,a low-rank sparse support vector machine beamforming algorithm is proposed.Support vector machine has excellent generalization ability and performs very well in estimation and prediction learning in the case of low snapshot.Based on the traditional MVDR beamforming algorithm,the support vector machine theory is introduced to make it have better performance in the case of low snapshot.At the same time,combined with the characteristics that the number of sound source signals is generally much less than the number of array elements in sound source location,the sound source signals are preprocessed on the basis of low-rank sparse theory.The simulation results show that when the number of snapshots is low,the positioning effect of support vector machine beamforming algorithm based on low-rank sparse is greatly improved.Finally,the positioning performance of the proposed algorithm is further verified by experiments.(3)A data acquisition system based on FPGA is designed.The microphone array of the system includes 32 digital microphone channels.The data received by the array is transmitted to the computer through Ethernet.