High-precision DOA Estimation For Underwater Acoustic Signals Based On Sparsity Adaptation

In order to make full use of underwater acoustic channel resources,accurately grasp the direction of incoming waves of the source,and ensure the consistency between the receiving azimuth of the array and the direction of incoming waves of the source,the direction of arrival(DOA)estimation of an array signal has attracted much attention.However,the prediction accuracy of traditional DOA estimation algorithm is low due to the complex and varied underwater acoustic environment and serious interference such as noise.For application scenarios with a small number of snapshots and a low signalto-noise ratio(SNR),the DOA estimation algorithm of underwater acoustic signal based on compressed sensing(CS)achieves better estimation performance.Nevertheless,the DOA estimation methods based on CS theory require information on source sparsity.Moreover,the influence of a complex underwater acoustic environment limits the accuracy of estimation algorithms.To make the algorithm better apply to complex underwater environment and improve the prediction accuracy,the study was based on adaptive sparse degree of DOA estimation is crucial.To address this limitation,this study proposes a high-precision DOA estimation model for underwater acoustic signals based on sparsity adaptation.The proposed model improves the prediction accuracy of the algorithm,and can be applied to various application scenarios.This paper summary of main work is as follows:(1)To solve the problem that source sparsity is usually unknown in underwater acoustic environment and the DOA estimation method based on CS is limited by the prior information of source sparsity,this paper designs an adaptive model of source sparsity based on causal convolutional neural network.It can effectively predict source sparsity when the number of sources is unknown.In order to make the model apply to different underwater acoustic arrays,the model uses the underwater acoustic signals received by the single array of uniform linear array as network input to learn the characteristics of underwater acoustic signals received by the array at different times.By learning the local features of underwater acoustic signal and enhancing the overall correlation between the local features of underwater acoustic signal,the adaptive prediction of source sparsity is finally realized.And the network model is still valid when the underwater acoustic receiving array is replaced.So,it can be applied to different array receiving scenarios.(2)In order to solve the problems of invalid atom screening and threshold selection difficulty in DOA estimation based on CS reconstruction algorithm,this paper proposes a differential combination matching pursuit(DCMP)algorithm.This method can be effectively applied to the scene with different SNR,different arrays,different snapshots and multiple sources.By introducing the differentiation path screening strategy,only the atoms with high matching degree and high differentiation are screened in each iteration.It avoids the problem of invalid screening in atomic screening and reduces the complexity of the algorithm.Then,the combined optimization strategy is proposed.And the prediction accuracy of the algorithm is improved by the left and right atom approximation criterion.It provides a new error correction idea for applying CS to DOA estimation problem.Moreover,in the process of iterative approximation,the noise power is used as the principle of atom filtering in the proposed algorithm,that is,the power of the residual difference between the reconstructed signal and the original signal constantly approximates the noise power of the received signal.(3)The underwater acoustic channel model of Bellhop was used for simulation verification.The test set accuracy of the source sparsity adaptive model proposed based on causal convolutional neural network can reach 95.1% under seven SNR and three array types.It is proved that the model can realize the adaptive prediction of source sparsity under different scenarios.Compared with Orthogonal Matching Pursuit(OMP)algorithm,Sparsity Adaptive Matching Pursuit(SAMP)algorithm and Multipath Matching Pursuit(MMP)algorithm in seven different SNR cases,the accuracy of the proposed DCMP algorithm is improved by 9.99% in the lowest case and 19.94% in the highest case on the basis of ensuring low time complexity,and the mean absolute error is reduced by about 0.05° in the lowest case and 14° in the highest case.In conclusion,this paper mainly studies the DOA estimation method of underwater acoustic signal based on CS reconstruction algorithm,and proposes an adaptive model of underwater acoustic source sparsity based on causal convolutional neural network to solve the unknown problem of underwater source sparsity.A DOA estimation method of underwater acoustic array based on DCMP algorithm is also proposed in this paper,which can be used in many different application scenarios.