Field-programmable gate array and deep neural network-accelerated spatial-spectral interferometry for rapid optical dispersion analysis

The paper is in F-DASI OPTICA

Abstract:

Spatial-spectral interferometry (SSI) is a technique used to reconstruct the electrical field of an ultrafast laser. By analyzing the spectral phase distribution, SSI provides valuable information about the optical dispersion affecting the spectral phase, which is related to the energy distribution of the laser pulses. SSI is a single-shot measurement process and has a low laser power requirement. However, the reconstruction algorithm involves numerous Fourier transform and filtering operations, which limits the applicability of SSI for real-time dispersion analysis. To address this issue, this Letter proposes a field-programmable gate array (FPGA)-based deep neural network to accelerate the spectral phase reconstruction and dispersion estimation process. The results show that the analysis time is improved from 124 to 9.27 ms, which represents a 13.4-fold improvement on the standard Fourier transform-based reconstruction algorithm.

Getting Started

Installation

• Python >= 3.10.12
• Tensorflow-gpu >= 2.12.0
• CUDA = 11.8
• cuDNN = 8.6.0