Research on DOA estimation based on Transformer model

Abstract

This study primarily aims to validate the applicability of Transformer in DOA estimation and to explore data processing procedures suitable for this field. Since the experiments are based on simulation data, the study first examines the size of the dataset and its parameters. Given that training Transformer networks typically requires larger datasets and is more challenging compared to convolutional neural networks, a convolutional neural network was chosen for testing. After selecting the dataset, the study presents a DOA estimation model implemented using Transformer and analyzes its performance. Finally, it provides methods for DOA estimation under conditions of unknown source numbers and wideband conditions.