Research On Signal Direction Estimation Algorithm For Acoustic Vector Sensor Array Based On Tensor Data Model

The traditional acoustic vector sensor array signal processing algorithm expands the obtained signals into matrix or long vector forms,which disrupts the multi-dimensional structural characteristics of the information obtained by the acoustic vector sensor array.To solve this problem,this paper selects the representative method in the field of high-dimensional data operation-tensor as the data expression form,and integrates the tensor model with the traditional acoustic vector sensor array signal processing algorithm,Enable the full utilization of multidimensional structural information in obtaining sound source information,thereby achieving the goal of improving the accuracy of direction estimation algorithms.This article first studies the fusion algorithm of tensor data model with conventional multiple signal classification(MUSIC)algorithm and rotation invariant subspace(ESPRIT)algorithm,and obtains MUSIC algorithm and ESPRIT algorithm based on tensor data model.Compared with conventional MUSIC algorithm and ESPRIT algorithm,they have the advantage of higher azimuth estimation accuracy.However,due to the expansion of the calculation process from low dimension to high dimension,it also brings the problem of high computational complexity.As the tensor data model based MUSIC algorithm needs a higher dimension spectral peak search due to the expansion of the data model dimension,which leads to a huge amount of computation,and the tensor data model based ESPRIT algorithm cannot automatically match the two-dimensional bearing estimation results,this paper then studies the DOA estimation algorithm based on autocovariance tensor model parallel factor decomposition(PARAFAC).Through analysis,it can be seen that although this algorithm does not require the highdimensional peak search and parameter pairing process mentioned earlier,it is prone to falling into local optima when performing PARAFAC decomposition under low signal-to-noise ratio,resulting in ineffective direction estimation.Aiming at the problem of estimation failure of DOA estimation algorithm based on autocovariance tensor model PARAFAC decomposition under low signal noise ratio,this paper successively studies two algorithms: DOA estimation algorithm based on tensor data model and direct trilinear PARAFAC decomposition and DOA estimation algorithm based on tensor data model and fast PARAFAC decomposition.The above two algorithms use different methods to initially estimate the factor matrix,thereby avoiding the problem of PARAFAC decomposition results falling into local optima due to random initialization of the factor matrix.Compared to other tensor based data model processing algorithms studied in this article,the latter can significantly reduce computational complexity.Simulation analysis and field experimental data processing confirm the correctness of the theoretical analysis of these two algorithms.