Temporal waveform denoising using deep learning for injection laser systems of inertial confinement fusion high-power laser facilities  

作　　者：Wei Chen Xinghua Lu Wei Fan Xiaochao Wang 

机构地区：[1]Key Laboratory of High Power Laser and Physics,Shanghai Institute of Optics and Fine Mechanics,Chinese Academy of Sciences,Shanghai,China [2]Center of Materials Science and Optoelectronics Engineering,University of Chinese Academy of Sciences,Beijing,China

出　　处：《High Power Laser Science and Engineering》2024年第6期207-220,共14页高功率激光科学与工程(英文版)

基　　金：supported by the Strategic Priority Research Program of the Chinese Academy of Sciences(No.XDA25020303);operation of the SG-Ⅱfacility

摘　　要：For the pulse shaping system of the SG-Ⅱ-up facility, we propose a U-shaped convolutional neural network that integrates multi-scale feature extraction capabilities, an attention mechanism and long short-term memory units, which effectively facilitates real-time denoising of diverse shaping pulses. We train the model using simulated datasets and evaluate it on both the simulated and experimental temporal waveforms. During the evaluation of simulated waveforms, we achieve high-precision denoising, resulting in great performance for temporal waveforms with frequency modulationto-amplitude modulation conversion(FM-to-AM) exceeding 50%, exceedingly high contrast of over 300:1 and multistep structures. The errors are less than 1% for both root mean square error and contrast, and there is a remarkable improvement in the signal-to-noise ratio by over 50%. During the evaluation of experimental waveforms, the model can obtain different denoised waveforms with contrast greater than 200:1. The stability of the model is verified using temporal waveforms with identical pulse widths and contrast, ensuring that while achieving smooth temporal profiles,the intricate details of the signals are preserved. The results demonstrate that the denoising model, trained utilizing the simulation dataset, is capable of efficiently processing complex temporal waveforms in real-time for experiments and mitigating the influence of electronic noise and FM-to-AM on the time–power curve.

关 键 词：deep learning frequency modulation-to-amplitude modulation conversion inertial confinement fusion SG-II facility temporal waveform denoising 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TN24[自动化与计算机技术—控制科学与工程]