Article: Deep learning approach for high-resolution DOA estimation Journal: International Journal of Ad Hoc and Ubiquitous Computing (IJAHUC) 2024 Vol.46 No.2 pp.90 - 103 Abstract: The direction-of-arrival (DOA) estimation is a significant problem in array signal processing and their application in different fields ranging from radars to wireless communications. This problem poses significant challenges when the separation of closely spaced sources becomes difficult. The state-of-art methods fail in providing sufficient resolution to separate the DOA's of closely spaced sources unless large number of large number of antennas is being used. In this paper, the problem of resolving direction-of-arrivals (DOAs) of two closely spaced sources using deep learning technology is considered. We propose a novel network named as DAE-CNN-MUSIC network. The problem is framed as a multi-label classification task. The proposed network is trained to make predictions of DOAs across a wide range of signal-to-noise ratios (SNRs), demonstrates enhanced durability in the presence of noise, and resilience even with smaller number of snapshots. The efficiency and superiority of the proposed network is tested under various scenarios. The experimental results demonstrate vast improvement in performance gains at low-SNR regime, and ability to resolve angles with more precision for SNR ≥ 15 dB when compared to state-of-the-art methods. Inderscience Publishers - linking academia, business and industry through research

Title: Deep learning approach for high-resolution DOA estimation

Authors: Ch Lokesh Dharma Theja; Kishore Kumar Puli

Addresses: Electronics and Communication Engineering, National Institute of Technology, Kondruprolu, Tadepalligudem, 534101, Andhra Pradesh, India ' Electronics and Communication Engineering, National Institute of Technology, Kondruprolu, Tadepalligudem, 534101, Andhra Pradesh, India

Abstract: The direction-of-arrival (DOA) estimation is a significant problem in array signal processing and their application in different fields ranging from radars to wireless communications. This problem poses significant challenges when the separation of closely spaced sources becomes difficult. The state-of-art methods fail in providing sufficient resolution to separate the DOA's of closely spaced sources unless large number of large number of antennas is being used. In this paper, the problem of resolving direction-of-arrivals (DOAs) of two closely spaced sources using deep learning technology is considered. We propose a novel network named as DAE-CNN-MUSIC network. The problem is framed as a multi-label classification task. The proposed network is trained to make predictions of DOAs across a wide range of signal-to-noise ratios (SNRs), demonstrates enhanced durability in the presence of noise, and resilience even with smaller number of snapshots. The efficiency and superiority of the proposed network is tested under various scenarios. The experimental results demonstrate vast improvement in performance gains at low-SNR regime, and ability to resolve angles with more precision for SNR ≥ 15 dB when compared to state-of-the-art methods.

Keywords: direction of arrival; DOA; DOA estimation; array signal processing; denoising autoencoder-convolutional neural network-multiple signal classification; DAE-CNN-MUSIC.

DOI: 10.1504/IJAHUC.2024.138742

International Journal of Ad Hoc and Ubiquitous Computing, 2024 Vol.46 No.2, pp.90 - 103

Received: 08 Feb 2024
Accepted: 19 Mar 2024
Published online: 29 May 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Deep learning approach for high-resolution DOA estimation

Keep up-to-date