中国物理B ›› 2026, Vol. 35 ›› Issue (5): 54204-054204.doi: 10.1088/1674-1056/ae3474

• • 上一篇    下一篇

Autocorrelation-based convolutional neural network for reconstruction of noisy attosecond streaking traces

Xiaoxian Zhu(朱孝先)1,2, Yitan Gao(高亦谈)1,2, Yiming Wang(王一鸣)2, Kun Zhao(赵昆)1,2,†, and Zhiyi Wei(魏志义)1,2   

  1. 1 Songshan Lake Materials Laboratory, Dongguan 523808, China;
    2 Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China
  • 收稿日期:2025-12-03 修回日期:2025-12-24 接受日期:2026-01-07 发布日期:2026-05-11
  • 通讯作者: Kun Zhao E-mail:zhaokun@iphy.ac.cn
  • 基金资助:
    Project supported by the CAS project for Young Scientists in Basic Research (Grant Nos. YSBR-059 and YSBR-115) and the National Natural Science Foundation of China (Grant No. 92150103).

Autocorrelation-based convolutional neural network for reconstruction of noisy attosecond streaking traces

Xiaoxian Zhu(朱孝先)1,2, Yitan Gao(高亦谈)1,2, Yiming Wang(王一鸣)2, Kun Zhao(赵昆)1,2,†, and Zhiyi Wei(魏志义)1,2   

  1. 1 Songshan Lake Materials Laboratory, Dongguan 523808, China;
    2 Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China
  • Received:2025-12-03 Revised:2025-12-24 Accepted:2026-01-07 Published:2026-05-11
  • Contact: Kun Zhao E-mail:zhaokun@iphy.ac.cn
  • Supported by:
    Project supported by the CAS project for Young Scientists in Basic Research (Grant Nos. YSBR-059 and YSBR-115) and the National Natural Science Foundation of China (Grant No. 92150103).

PDF (PC)

1

赞

摘要： Attosecond light sources serve as crucial tools for investigating the ultrafast electronic dynamics in matter with remarkable temporal resolution. Traditional methods face difficulties in accurately measuring attosecond pulses, and the prevailing approach involves utilizing attosecond streak cameras coupled with inversion algorithms to reconstruct phase information. However, these algorithms often require multiple iterations and extensive computational time. This study investigates the utilization of autocorrelation graphs as inputs for a convolutional neural network (CNN) to invert streaking traces obtained by attosecond streak camera. We explore the noise resistance capability of autocorrelation within the CNN inversion and aim to provide a physical explanation for its effectiveness. The objective of this research is to enhance the accuracy and reliability of CNN inversion for attosecond streaking traces, enabling improved resilience against experimental noises.

关键词: attosecond, convolutional neural networks, autocorrelation, attosecond streaking camera, phase retrieval, noise robustness

Abstract: Attosecond light sources serve as crucial tools for investigating the ultrafast electronic dynamics in matter with remarkable temporal resolution. Traditional methods face difficulties in accurately measuring attosecond pulses, and the prevailing approach involves utilizing attosecond streak cameras coupled with inversion algorithms to reconstruct phase information. However, these algorithms often require multiple iterations and extensive computational time. This study investigates the utilization of autocorrelation graphs as inputs for a convolutional neural network (CNN) to invert streaking traces obtained by attosecond streak camera. We explore the noise resistance capability of autocorrelation within the CNN inversion and aim to provide a physical explanation for its effectiveness. The objective of this research is to enhance the accuracy and reliability of CNN inversion for attosecond streaking traces, enabling improved resilience against experimental noises.

Key words: attosecond, convolutional neural networks, autocorrelation, attosecond streaking camera, phase retrieval, noise robustness

中图分类号:  (Ultrafast processes; optical pulse generation and pulse compression)

  • 42.65.Re
42.30.-d (Imaging and optical processing) 42.50.Hz (Strong-field excitation of optical transitions in quantum systems; multiphoton processes; dynamic Stark shift) 07.05.Mh (Neural networks, fuzzy logic, artificial intelligence)