DOA Estimation For Underwater Sensors Based On Sparse Bayesian Learning

Direction of Arrival（DOA）estimation is one of the important research directions in the domain of array signal processing,which has a wide range of applications in wireless communication,radar,sonar,biomedical and other fields.The estimation performance of traditional underwater acoustic target detection technology is limited by factors such as array aperture,environmental noise,and data length,and the performance is close to the limit.Studies have shown that using the spatial sparsity of the signal and Bayesian learning methods can achieve higher accuracy DOA estimation.However,most of the existing DOA estimation methods based on sparse Bayesian learning are studied under the perfect array model.In the sonar system,there are mutual coupling between hydrophones,which leads to the deterioration of the performance of the DOA estimation methods.Aiming at mutual coupling of array elements and off-grid errors in the system,this paper studies the DOA estimation method based on sparse Bayesian learning.It can improve the estimation performance,but the computational cost is relatively large.This paper studies an efficient multi-snapshot data processing method.The main research contents are as follows:1.In the sonar direction finding system composed of underwater sensors,there are the coupling effect between the array elements,which leads to the DOA estimation performance degradation or even failure.Aiming at this situation,a off-grid Sparse Bayesian learning estimation method is proposed.Firstly,the spatial domain is discretized into a uniform grid,and the off-grid error and the mutual coupling coefficient vector are introduced for modeling the array data.Secondly,the prior distribution of the off-grid error and the mutual coupling coefficient vector is determined,and the Bayesian model is established by using these parameters.Finally,the expectation maximization algorithm is used for parameter learning and parameter update such as unknown mutual coupling and off-grid error.So as to the DOA estimation method can eliminate the influence of unknown mutual coupling,and simulation results show that can still achieve better azimuth estimation performance in the case of unknown array elements and have a strong multi-target resolution capability existing mutual coupling.2.The multi-snapshot signal datas are processed in the sonar system.In sparse Bayesian DOA estimation model,the traditional multi-snapshot data processing method refer the single-snapshot data processing process,and then get the estimated results of each snapshot,and the corresponding parameter estimates are obtained by using l₁-norm.Aiming at low computational efficiency and large computational cost in this process,a DOA estimation method based on sparse representation of covariance matrix is proposed.Firstly,using Khatri-Rao product to convert multi-snapshot data model into a single-snapshot data model,and then block it.Secondly,a Bayesian parametric learning model is established,the variational inference algorithm is used to learn and update parameters such as expectation and mean.Finally,the noise parameters are removed from the parameters,and the DOA estimate of the incident signal is obtained by spectral peak search.The simulation results show that azimuth estimation performance of the proposed method in this paper is better than traditional multi-snapshot data joint processing method,such as OGSBL.3.The effectiveness of the method proposed in this paper is verified by the experiment in Qian Dao Lake.The experiment results show that the performance of the method proposed in this paper is better than the traditional DOA estimation methods,such as MUSIC,CBF,MVDR,OGSBL and l₁-SVD.