基于变分稀疏贝叶斯的off-grid DOA估计  

An off-grid DOA Estimation Method via Variational Sparse Bayesian Learning

作　　者：冯晓敏 沈重[1,2] 张鲲[1,2,3] Feng Xiaomin;Shen Chong;Zhang Kun(College of Information and Communication Engineering,Hainan University,Haikou 570228,China;State Key Laboratory of Marine Resources Utilization in South China Sea,Hainan University,Haikou 570228,China;College of Ocean Information Engineering,Hainan Tropical Ocean University,Sanya 572022,China)

机构地区：[1]海南大学信息与通信工程学院,海南海口570228 [2]海南大学南海海洋资源利用国家重点实验室,海南海口570228 [3]海南热带海洋学院MTA教育中心,海南三亚572022

出　　处：《海南大学学报（自然科学版）》2019年第3期193-202,共10页Natural Science Journal of Hainan University

基　　金：国家自然科学基金(61461017,61861015);海南省高等学校科学研究重点项目(Hnky2019ZD-35);海南省自然科学基金(2017CXTD0004)

摘　　要：针对传统稀疏贝叶斯学习方法在解决低信噪比条件下信号到达角DOA估计中,到达角不完全落在阵元端离散化网格点上的情况,提出了一种基于变分稀疏贝叶斯学习期望最大化的离格DOA处理方法.该方法首先对均匀线阵的实际接收信号off-grid情况建立参数化稀疏模型,利用变分稀疏贝叶斯学习方法,通过最小化KL散度寻求一个与后验概率近似的概率分布,其次期望最大化方法分别在期望阶段和最大化阶段进行参数推断,进一步在离格误差模型下以较高的精度和分辨率实现信号到达角的估计,最后仿真结果表明,该方法不仅在传统SBL方法的基础上提高了运算效率,而且具有更高的空域分辨率和角度估计精度,该方法具有优越的角度估计性能.In the report,a method based on the variation sparse Bayesian learning using expectation maximization was proposed to tackle with the off-grid problem induced by which the true direction-of-arrival (DOA) is not exactly impinged on the discretized grid point,which was inherent in traditional sparse Bayesian learning algorithm in DOA estimation field.Firstly,a parametric sparse model was constructed for the off-grid problem of the actual received signal in a uniform linear array,variational sparse Bayesian learning algorithm was used to obtain a distribution that is approximate to the posterior probability by minimizing the KL divergence.Secondly,the unknown variable and parameter was inferred respectively in the stage of expectation and maximization in EM scheme,which determines the estimation accuracy under the off-grid error framework.The numerical experiment result indicated that the method not only in enhancing the computational efficiency compared to the previous SBL algorithm,but also providing a higher spatial resolution and estimation accuracy,with a superior estimation performance.

关 键 词：DOA估计 OFF-GRID 稀疏贝叶斯学习 变分推断 期望最大化 

分 类 号：TN919.3[电子电信—通信与信息系统]