Research On Source Counting Based On Singer Acoustic Vector Sensor

The AVS can pick up the information of sound pressure and vibration velocity at the same time,and a single AVS can realize spatial filtering and target azimuth estimation.Due to the ability of resisting the interference of isotropic noise,AVS is widely used in underwater vehicles,buoys and other small unmanned platforms.Target number estimation is the key to realize autonomous detection of underwater unmanned platform.Therefore,this project studies the multi-target number estimation method based on the single AVS,aiming to clarify and verify the multi-target resolution mechanism of the single vector hydrophone,realize the under-determined multi-target number estimation of the single AVS,and provide technical support for the autonomous detection of small underwater unmanned system.This paper mainly carries out the following research:(1)The introduction of WDO of the signal theoretically explains the phenomenon that the spectrum corresponding to the multi-target appears in the cross-spectral DOA histogram in the AVS broadband multi-target experiment.The WDO of the signal means that at a time-frequency point,only the energy of one target is dominant,while the energy of other targets is secondary.Therefore,the DOA estimation result is close to the real DOA of the target.When a considerable proportion of time-frequency points have this characteristic,these time-frequency points form clusters close to the real DOA of the target,and then form spectral peaks by DOA histogram statistics,so as to achieve multi-target resolution.Thus the mechanism of the multi-target resolution capability of the single AVS's DOA histogram has been clarified.(2)The method of multi-target number estimation based on Gaussian mixture model with single vector hydrophone is studied.The method is conducted by modeling the DOA histogram from AVS as a GMM.The Dirichlet distribution is used to constrain the weight of the Gaussian component to avoid the situation that multiple Gaussian components correspond to the same target spectrum peak.EM algorithm was used to calculate the model parameters.To filter the Gaussian component of the spectral peak of the fitted sound source the threshold is set.Finally,the Gaussian component of the fitted side lobes is eliminated to achieve the target number estimation.It is proved that when the number of targets reaches or exceeds the number of channels of the single vector hydrophone,the single vector hydrophone still has the ability of multi-target resolution and number estimation.(3)Due to the threshold setting in the GMM-based multi-target discrimination method will have a great impact on the results and the low efficiency.This paper presents a multi-target number estimation method for single AVS based on density clustering.The product of the sample density of the DOA estimation result and the minimum distance from the higher density sample was taken as its feature,and the DOA estimation result whose eigenvalue was significantly larger than that of other samples was searched by the ordered sequence difference value as the center of the cluster class,so as to calculate the number of cluster classes and realize the target number estimation.The contribution of samples with different frequency points to clustering is different.The larger the proportion of energy of the dominant target is,the closer the azimuth estimation result is to the true value of target azimuth,the greater the contribution of samples to clustering is,and vice versa.Therefore,this paper further proposes that the local confidence measure,which represents the energy proportion of dominant target,is used to enhance the sample density.The larger the energy proportion of dominant target,the larger the sample density will be,while the smaller the energy proportion of dominant target,the sample density will be diluted,making the cluster more compact.Thus,the accuracy of clustering is improved,and the accuracy of multi-target number estimation is improved.The validity of the proposed method is verified by the analysis of lake test data.