Denoised Attention Neural Network for Direction-of-Arrival Estimation in Low SNR

Direction-of-arrival (DOA) estimation methods are divided into model-driven and data-driven algorithms according to the driving method. However, traditional model-driven algorithms suffer from poor accuracy at low signal-to-noise ratios (SNR), while data-driven models fail to consider the significance between data, leading to a high number of algorithm parameters, inaccurate DOA estimation. To address these issues, we propose the Denoised Attention Neural Network (DANN), which utilizes the channel attention module, spatial attention module, and multi-head attention module to assign weights to features and learn their importance adaptively. In addition, to reduce the interference of noise on the estimation results, a threshold denoising module is introduced to achieve noise abatement. Simulation results demonstrate that our method achieves higher estimation accuracy at low SNR with smaller network parameters compared to existing methods.