DOA Estimation Method Based on Cascaded Neural Network for Two Closely Spaced Sources
Yu Guo, Zhi Zhang, Yuzhen Huang, Ping Zhang
Abstract
In this letter, we explore the problem of DOA estimation using neural networks for two closely spaced sources. Since the traditional high-resolution techniques based on classical algorithms cannot achieve high-accuracy DOA estimation in the presence of two closely spaced sources, especially at low signal-to-noise ratios (SNR), we propose a novel DOA estimation method based on a cascaded neural network to address this problem. Specifically, this network comprises two parts: the SNR classification network and the DOA estimation network. The latter network contains two estimation subnetworks, which are appropriate for different SNRs by training with noisy data and activated by the output of the SNR classification network. Simulation results demonstrate that the estimation performance of our proposed method achieves much better than that of the existing algorithms under various conditions, especially for the scenes with low SNRs or small snapshot number.