DOA Estimation With Deep Learning: A Limited Training Data Framework

doi:10.1109/TCOMM.2025.3535873

DOA Estimation With Deep Learning: A Limited Training Data Framework

2025-01-01

0

PRE

AI

DOI：10.1109/TCOMM.2025.3535873

Original

Original request for help

Save

Abstract

En 中文

Deep Learning (DL) achieves significant performance in estimating the direction of arrival (DOA) in array signal processing. However, many existing DL methods require a large amount of data to train a specialized DL network. To reduce data requirements for training, this paper presents a novel DL-based DOA estimation algorithm for limited training data(LTDDOA-net). The proposed algorithm utilizes the properties of second-order derivatives of the loss function and the ‘learn to learn’ approach to construct a framework that can achieve good performance with minimal data training. Initially, we developed a neural network designed for DOA estimation. This network was subsequently trained using proposed method and loss function on a limited dataset. Ultimately, we validated the practicality and benefits of our approach through simulations and hardware experiment. The results of simulations and hardware experiment have verified the superiority of the proposed approach.

Keywords:

Direction-of-arrival estimation

array signal processing

deep learning

limited training data

uniform linear array

neural networks

DOA Estimation With Deep Learning: A Limited Training Data Framework

Abstract

Journal

Organization