Deep learning reconstruction algorithm for frequency-resolved optical gating.

Area of Science:

• Ultrafast Optics and Photonics
• Machine Learning in Physics
• Laser Science and Technology

Background:

• Frequency-resolved optical gating (FROG) is crucial for characterizing ultrashort laser pulses.
• The delay operation in traditional FROG is a significant bottleneck for high-speed measurements.
• Existing FROG algorithms struggle with rapid pulse reconstruction.

Purpose of the Study:

• To develop a deep learning framework for accelerated ultrashort pulse reconstruction.
• To enable high-speed measurements by overcoming FROG's time-consuming delay stage.
• To demonstrate accurate pulse reconstruction using partial spectrogram data.

Main Methods:

• Implementation of a sequence-to-sequence (Seq2Seq) model with attention mechanism.
• Training and validation of the deep learning model on FROG spectrograms.
• Comparison of the deep learning approach with traditional 2D phase retrieval algorithms.

Main Results:

• Achieved root mean squared errors (RMSE) of 9.5 × 10^-4 for amplitude and 0.20 for phase reconstruction.
• Reduced spectral measurement time by at least 8 times compared to classic FROG.
• Demonstrated pulse reconstruction in approximately 0.2 seconds, surpassing existing iterative methods.

Conclusions:

• The proposed deep learning framework enables accurate and significantly faster ultrashort pulse reconstruction.
• This advancement has the potential to revolutionize high-speed measurements in ultrafast science.
• The model's ability to use partial spectrograms opens new possibilities for real-time pulse characterization.