DOA Estimation using Multi-head Self-Attention with Relative Positional Encoding and Hybrid Multi-Objective Grey Wolf Optimization

Abstract

Direction of Arrival (DOA) estimation is a process of estimating the incident angle of waves from a radiating or reflecting source using an antenna array. DOA estimation is of high importance in several areas, like navigation, remote sensing, radar, sonar and wireless communication. Nevertheless, noise, interference and multipath effects complicate the accurate estimation of DOA. These challenges can be overcome by advancing deep learning (DL) techniques and modeling the intricate relationships in the signal data. In the presented work, a new DOA estimation method utilizing a hybrid LSTM-CNN feature extraction pipeline, multi-head attention-GWO-mRMR-based feature selection, and ANN is presented. The method includes improvements such as positional encoding in multi-head attention and adaptive learning rate in ANN. It is trained and evaluated on synthetic signal data generated by a Uniform Linear Array (ULA) model. In evaluation, it achieved exceptional results with MAE (2.15), MSE (15.67), RMSE (3.95), and R² score (0.97). The accuracy of the DOA estimations can be attributed to the robust feature extraction pipeline that effectively captured spatial and temporal features and the model’s focus on the most relevant and informative features. Additionally, the performance was superior to that of conventional existing techniques of MUSIC, SVR, and ESPRIT, establishing the efficacy of the suggested method for DOA estimation in realistic scenarios.