低快照数下多目标DOA估计方法

Multi-target DOA estimation method under low number of snapshots

针对低快照数和较低信噪比条件下多信号到达方向(DOA)估计性能下降问题,提出了基于深度学习的离格DOA估计方法。选择具有特殊结构的非均匀阵列以提高阵列自由度,将样本协方差矩阵建模为真实协方差矩阵的噪声版本,利用堆叠降噪自动编码器(SDAE)重构出新协方差矩阵,最后结合超分辨率算法实现DOA估计。仿真结果表明:在低快照数为10及较低信噪比2 dB情况下,数据先采用SDAE进行处理再进行DOA估计,多目标DOA估计准确率能达到92.06%,相对于传统方法及深度神经网络(DNN)分别提高了55.39%,25.025%。

Aiming at the performance degradation problem of multi-signal direction of arrival(DOA)estimation under the conditions of low snapshots and low signal-to-noise ratio,an off-frame DOA estimation method based on deep learning is proposed.Select a non-uniform array with a special structure to increase the degree of freedom of the array,model the sample covariance matrix as a noise version of the real covariance matrix,and use the stacked denoising auto encoder(SDAE)to reconstruct the new covariance matrix.Finally,DOA estimation is achieved by combining with the super-resolution algorithm.The simulation results show that when the number of snapshots is 10 and the low signal-to-noise ratio is 2 dB,the data is firstly processed by SDAE and then DOA estimation is done.The accuracy of multi-target DOA estimation can reach 92.06%,compared with traditional methods and DNN,which is increased by 55.39%and 25.025%,respectively.