Two-Stage Deep Convolutional Neural Networks for DOA Estimation in Impulsive Noise
Ruiyan Cai, Quan Tian
Abstract
Direction-of-arrival (DOA) estimation methods have been widely and deeply studied in Gaussian noise environments. However, if there is impulsive channel noise, the performance of the method will significantly decline, and reasonable results may not be obtained. Considering that the high performance of model-driven DOA estimation algorithms requires large arrays and more sample data, this communication proposes a two-stage deep convolutional neural network (TSDCN) algorithm for DOA estimation. The first stage suppresses alpha-stable distributed impulsive noise through an adversarial learning network, and the second stage realizes DOA estimation through a deep convolutional neural network. Simulation and real-world experiments show that the TSDCN outperforms most DOA estimation algorithms in terms of robustness and estimation accuracy in impulsive noise environments.