| 摘要: |
| 本文提出一种基于稀疏贝叶斯学习的改进离网DOA估计算法，以提升非理想测向环境下在低信噪比、低快拍数时的DOA估计性能，称之为MOGSBL算法。本算法将信号源方位区间进行离散化，得到方位离散网格。为阵列接收信号建立稀疏贝叶斯模型，将网格节点修正量设为模型超参数。采用期望最大化算法迭代更新网格节点修正量，使更新后的网格节点更接近真实源信号方位。为了检验MOGSBL算法的性能，本文进行了大量的数值实验，并将MOGSBL算法的DOA估计结果与RSBL算法、OGSBL算法和L1?SVD算法进行对比。在不同信噪比和不同快拍数时，MOGSBL算法均能清晰分辨方位很接近的两个信号源，角度分辨率明显高于RSBL算法、OGSBL算法和L1?SVD算法。随着信噪比和快拍数的增加，4种算法的RMSE均逐渐减小。但MOGSBL算法的RMSE明显低于RSBL算法、OGSBL算法和L1?SVD算法，且RSBL算法、OGSBL算法优于L1?SVD算法。实验还分析了方向测试范围的离散网格节点数对DOA估计的影响，发现细密的离散网格可以提高DOA估计精度，但DOA估计的计算量会增加。且在任意网格节点数时，相比于RSBL算法、OGSBL算法和L1?SVD算法，本文的MOGSBL算法均具有最低的RMSE和最短的计算时间。 |
| 关键词: DOA估计 离网模型 稀疏贝叶斯学习 网格更新 角度分辨率 |
| DOI：DOI:10.3969/j.issn.1672-2337.2024.01.006 |
| 分类号:TN911.7 |
| 基金项目:国家自然科学基金资助项目（No.61890544，91748106） |
|
| An Off⁃Grid DOA Estimation Algorithm Based on Sparse Bayesian Learning |
|
ZHANG Yu, JING Xinlei, JIANG Zhongjin
|
|
State Key Laboratory of Millimeter Waves, Southeast University, Nanjing 210096, China
|
| Abstract: |
| In this paper, an improved off?grid DOA estimation algorithm based on sparse Bayesian learning is proposed to improve the performance of DOA estimation in non?ideal direction finding environment with low SNRs and low snapshots, which is called MOGSBL algorithm. The algorithm discretizes the signal source orientation interval to obtain an orientation discrete grid. A sparse Bayesian model is established for the array received signal, and the grid node corrections are set as the model hyper?parameters. The expectation maximization algorithm is used to iteratively update the grid node corrections, so that the updated grid nodes are closer to the real source signal orientation. In order to test the performance of the MOGSBL algorithm, a large amount of numerical experiments are conducted, and the comparisons are made among the DOA estimation results of the MOGSBL algorithm, the RSBL algorithm, the OGSBL algorithm and the L1?SVD algorithm. At different SNRs and different amounts of snapshots, the MOGSBL algorithm can clearly distinguish the two signal sources at close orientations, and the angular resolution is significantly higher than the RSBL algorithm, the OGSBL algorithm and the L1?SVD algorithm. With the increase of the SNR and the number of snapshots, the RMSE of all four algorithms gradually decreases, while the RMSE of the MOGSBL algorithm is clearly lower than the RSBL algorithm, the OGSBL algorithms and L1?SVD algorithm, and the RSBL algorithm and the OGSBL algorithm are better than the L1?SVD algorithm. The experiments also analyze the effect of the number of discrete grid nodes in the direction test range on DOA estimation. It is found that a fine discrete grid can improve the DOA estimation accuracy, however, the computational effort of DOA estimation will increase. And at any number of grid nodes, the MOGSBL algorithm in this paper has the lowest RMSE and the shortest computation time compared with the RSBL algorithm, the OGSBL algorithm and the L1?SVD algorithm. |
| Key words: DOA estimation off⁃grid model sparse Bayesian learning grid update angular resolution |
