Frequency Resolved Optical Gating Analysis with a Deep Neural Network for Ultrafast Pulse Characterization

Frequency Resolved Optical Gating (FROG) is a widely used technique for characterizing ultrafast laser pulses, providing both the amplitude and phase of the electric field. Accurate and rapid analysis of FROG data is crucial for advancing applications in ultrafast science. In this work, we introduce a novel approach for analyzing FROG data using a deep neural network. Our algorithm can operate in two modes: untrained and trained.  In the untrained mode, the neural network does not require pre-training on a large dataset. With just a single measurement, it can reconstruct the pulse as efficiently as traditional phase retrieval algorithms. In the trained mode, the neural network is pre-trained on a large set of synthetic or experimental data. Notably, our network architecture does not require ground truth data for training. It can be directly trained with FROG measurements without prior knowledge of the actual pulse shape. The trained neural network significantly reduces computational time and offers a powerful tool for real-time pulse characterization, which is particularly beneficial for high-repetition rate applications. We will demonstrate with data taken from both the ALEPH laser facility and the Scarlet laser facility.

 

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.