| 摘要: |
| 受恶劣电磁环境和元器件老化等因素影响，多输入多输出(Multiple-Input Multiple-Output,MIMO)雷达的天线阵元发生故障的概率增加，而阵元故障会严重降低目标波达方向(Direction of Arrival,DOA)估计性能。现有的大多数基于深度学习的DOA估计方法未能充分利用阵列模型的先验信息，导致其建立的映射关系极为复杂，从而使得网络拟合难度较大。为此，提出一种基于先验驱动残差注意力网络的阵元故障MIMO雷达DOA估计方法。首先，利用MIMO雷达协方差矩阵的双重Toeplitz先验特性，构建了基于先验驱动的残差注意力网络，并引入残差注意力块对协方差矩阵的特征进行加权处理，旨在学习阵元故障下存在数据缺失的协方差矩阵和完整协方差矩阵生成向量之间的映射关系。然后，根据残差注意力网络输出的生成向量估计值得到完整的协方差矩阵。最后，利用RD-ESPRIT(Reduced Dimension ESPRIT)算法估计目标DOA。仿真结果表明，所提算法在阵元故障下的DOA估计性能优于现有算法，在信噪比为15 dB时，其DOA估计精度比效果最好的现有算法提高了43.26%。 |
| 关键词: MIMO雷达 DOA估计 双重Toeplitz先验 残差网络 注意力机制 |
| DOI：10.20079/j.issn.1001-893x.240521002 |
|
| 基金项目:国家自然科学基金资助项目（62071238）；江苏省自然科学基金（BK20191399） |
|
| DOA Estimation for MIMO Radar under Element Failure Based on Prior-driven Residual Attention Network |
| CHEN Jinli,ZHOU Long,LI Jiaqiang,YAO Changhua |
| (School of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 21044,China) |
| Abstract: |
| Affected by factors such as harsh electromagnetic environment and aging components,the probability of antenna array element failure of multiple-input multiple-output(MIMO) radar increases,and array element failure will severely degrade the performance of Direction of arrival(DOA) estimation.Most existing deep learning-based DOA estimation methods fail to fully exploit the prior information of the array model,resulting in extremely complex mapping relationships,which makes network fitting difficult.To this end,a DOA estimation method for MIMO radar under element failure based on a prior-driven residual attention network is proposed.Firstly,by leveraging the doubly Toeplitz prior properties of the covariance matrix of MIMO radar,a prior-driven residual attention network is constructed.Residual attention blocks are introduced to weight the features of the covariance matrix.The network aims to learn the mapping relationship between the covariance matrix with missing data of the failed elements and the generation vector of the complete covariance matrix.Then,the complete covariance matrix is obtained from the generated vectors output by the network.Finally,the Reduced Dimension ESPRIT(RD-ESPRIT) algorithm is used to estimate the target DOA.Simulation results show that the proposed algorithm outperforms existing methods in DOA estimation under element failure,achieving a 43.26% accuracy improvement at a signal-to-noise ratio of 15 dB compared with the best existing algorithm. |
| Key words: MIMO radar DOA estimation doubly Toeplitz prior residual network attention mechanism |
