基于神经网络的相干信源稳健DOA估计

目前基于神经网络的DOA估计主要是针对理想情况下的均匀线阵，且信源非相干，可估计的信源数较少.针对阵列误差和相干信源同时存在的问题，在理想数据集中引入互耦误差、阵元幅度误差、阵元相位误差以及阵元位置误差，并设计了一个多通道CNN+DNN网络和目标函数生成方法，用于相干信源的稳健DOA估计.引用B-band互耦模型和相关误差模型合成阵列输出信号，通过提取阵列输出信号的协方差矩阵的实部、虚部与相位角，构建网络的输入信号.对理想条件下的MUSIC算法DOA估计结果进行拟合，根据拟合公式生成多信源从不同角度入射时的空间谱，作为网络的目标信号.使用相同的数据集对本文DOA估计网络与其它文献中的DOA估计网络进行训练和测试.结果显示，在不同信噪比、不同误差大小以及不同信源数的情况下，本网络的稳健性和解相干能力都更优.

Robust DOA estimation of coherent sources based on neural network

Most neural network-based DOA estimation methods are designed for the uniform linear array with a few incident signals which are uncoherent under ideal situation. To tackle the case that the array is imperfect and its signals are coherent, this paper designs a multi-channel CNN-DNN network and an objective function generation by introducing the errors of mutual coupling, amplitude, phase, and location with coherent signals. The input signals of the proposed nerual network are constructed by extracting the real part, imaginary part and phase angle from the covariance matrix of the array output. The DOA estimation results of the MUSIC algorithm under ideal conditions are fitted and the fitting result is used to generate the target of the network. The DOA estimation networks proposed in this paper and other literature are trained and tested using the same data set. The results show that the robustness and decoherence ability of the proposed network are better in terms of different signal-to-noise ratios, array errors and numbers of signal sources, compared to the previous neural network-based DOA methods.