基于深度卷积网络的二维波达方向估计方法  

2D Direction of Arrival Estimation Based on Deep Convolutional Network

作　　者：袁野[1] 张伟科 许左宏 YUAN Ye;ZHANG Weike;XU Zuohong(Unit 32806 of PLA,Beijing 100091,China;Unit 96901 of PLA,Beijing 100094,China;Institute of Systems Engineering,Academy of Military Sciences,Beijing 100141,China)

机构地区：[1]中国人民解放军32806部队,北京100091 [2]中国人民解放军96901部队,北京100094 [3]军事科学院系统工程研究院,北京100141

出　　处：《电讯技术》2024年第4期497-503,共7页Telecommunication Engineering

摘　　要：为了提高信号波达方向估计技术的实时性和简便性,设计了一种适用于估计均匀圆阵多信号波达方向的深度卷积网络。由阵列观测数据得到的协方差矩阵被当作是包含实部和虚部两个通道的图像,将其当作是卷积神经网络的输入张量,便可以通过训练网络来提取包含在信号协方差矩阵中的波达方向细微特征,从而实现准确快速地同时对多个入射信号的方向进行估计的目的。仿真结果表明,设计的深度卷积网络能够很好地完成二维信号波达方向估计。相比于现有估计方法,卷积网络给出的结果更加精确,且算法相对稳定。因此,提出的深度卷积网络在多目标方位识别与跟踪领域具有潜在的工程应用价值。To improve the real-time and convenient feature of the direction of arrival(DOA)estimation technology,a deep convolution network(DCN)is proposed.The covariance matrix of received signal obtained from the uniformed circular array(UCA)is regarded as an image which contains the real part channel and imaginary part channel.By using the covariance matrix as the input tensor of the convolutional network,it is possible to extract the subtle feature of the DOA implied in the covariance matrix,therefore,the DOA information of multi-signal can be estimated quickly and precisely.The simulation results show that the proposed DCN can achieve the DOA estimation of two dimensional signals well.Compared with the traditional method based on the sub-space calculation,the proposed network can obtain more accurate result and the algorithm is relatively stable,therefore,the network has some potential applications in engineering.

关 键 词：均匀圆阵(UCA) 波达方向(DOA)估计 深度卷积网络(DCN) 人工智能 图像分类 

分 类 号：TN911[电子电信—通信与信息系统]