Abstract

A deep-learning-based time–frequency domain signal recovery method is proposed to deal with the signal distortion in fiber-connected radar networks. In this method, the deteriorated signal is converted to the time–frequency domain, and a two-dimensional convolutional neural network is used to conduct signal recovery before inverse conversion to the time domain. This method can achieve high-accuracy signal recovery by learning the complete features in both time and frequency domains. In the experiment, distorted linear frequency modulated radar signals with a bandwidth of 2 GHz after 8-km fiber transmission are recovered with the noise effectively suppressed. The proposed signal recovery method works well under different input signal-to-noise ratios. Specially, the average peak to floor ratio after radar pulse compression is improved by 25.5 dB. In addition, the method is proved to be able to recover radar signals of multiple targets.

© 2021 Optica Publishing Group

Photonics-assisted compressed sensing radar receiver for frequency domain non-sparse signal sampling based on dictionary learning

Shiyang Liu, Shi Wang, Taixia Shi, and Yang Chen
Opt. Lett. 48(3) 767-770 (2023)

Adaptive deep-learning algorithm for signal recovery of broadband microwave photonic receiving systems based on supervised training

Shaofu Xu, Rui Wang, Xiuting Zou, and Weiwen Zou
J. Opt. Soc. Am. B 38(3) 834-841 (2021)

Multi-band low-noise microwave-signal-receiving system with a photonic frequency down-conversion and transfer-learning network

Sicheng Yi, Shaofu Xu, and Weiwen Zou
Opt. Lett. 46(23) 5982-5985 (2021)

Deep learning of image- and time-domain data enhances the visibility of structures in optoacoustic tomography

Neda Davoudi, Berkan Lafci, Ali Özbek, Xosé Luís Deán-Ben, and Daniel Razansky
Opt. Lett. 46(13) 3029-3032 (2021)

Temporal and spectral unmixing of photoacoustic signals by deep learning

Yifeng Zhou, Fenghe Zhong, and Song Hu
Opt. Lett. 46(11) 2690-2693 (2021)