一种深度学习稀疏单快拍DOA估计方法
A Deep Learning Approach for Sparse Single Snapshot DOA Estimation
-
摘要: 基于信号的稀疏特性，稀疏恢复（Sparse Recovery， SR）方法可利用单快拍数据进行相关信号源的高分辨波达方向（Direction of Arrival， DOA）估计。然而，现有SR-DOA模型求解方法存在参数设置困难、运算复杂度高或精度有待提高等问题，实际应用受限。针对上述问题，本文提出平滑L0网络（Smoothed L0 Net， SL0-Net）方法，将基于模型驱动SL0算法和基于数据驱动的深度学习方法相结合，用于SR-DOA模型的求解。首先，建立DOA估计的SR模型，并对用于求解该模型的SL0算法进行分析。然后，根据深度学习框架构建SL0-Net，并基于充足完备的数据集对其网络参数进行训练。最后，利用训练得到的SL0-Net对SR-DOA模型进行求解，获得DOA高分辨估计。仿真结果表明，与现有典型算法相比，所提SL0-Net更适于信号源数目未知条件下的快速高分辨DOA估计。Abstract: Based on the sparsity of signal， Sparse Recovery （SR） method can use single snapshot data for high-resolution DOA （Direction of Arrival） estimation of correlated signal sources. However， existing methods for solving the SR-DOA model always suffer from the problems of parameter setting difficulty， high computational complexity， or low recovery accuracy， limiting their practical applications. To solve these problems， this paper proposes Smoothed L0 Net （SL0-Net）， which combines the model-based SL0 algorithm and the data-driven deep learning method to solve the SR-DOA model. At first， the SR model for DOA estimation is established and the SL0 algorithm used to solve this model is analyzed. Then， based on deep learning framework， SL0-Net is constructed， whose parameters are trained with sufficient and complete datasets. At last， the trained SL0-Net is used for solving the SR-DOA model， achieving the high-resolution DOA estimation result. Simulation results show that， compared with existing typical algorithms， the proposed SL0-Net is more suitable for fast and high-resolution DOA estimation under the condition of unknown signal source number.
下载: