Deep-Learning-Enabled High-Fidelity Absorbance Spectra from Distorted Dual-Comb Absorption Spectroscopy for Gas...

Area of Science:

• Laser Spectroscopy
• Computational Spectroscopy
• Chemical Analysis

Background:

• Dual-comb absorption spectroscopy offers broad spectral coverage, high resolution, speed, and frequency accuracy.
• Etalon effects and complex baseline extraction hinder accurate spectral recovery and quantification in this technique.

Purpose of the Study:

• To develop a deep learning framework for processing distorted dual-comb absorption spectroscopy data.
• To overcome challenges in etalon removal and baseline extraction for improved absorbance spectra recovery.

Main Methods:

• A U-net model for etalon removal.
• A modified U-net with physical constraints and an iterative penalized least squares method for baseline extraction.
• Training datasets combining experimental baselines and simulated gas absorption spectra from the HITRAN database.

Main Results:

• The framework successfully obtained accurate CO2 absorbance spectra with high consistency to HITRAN database.
• Mean absolute error of the absorbance spectrum was less than 1% of the maximum absorbance value.
• Retrieved CO2 concentration showed a relative error under 2%, outperforming traditional methods.

Conclusions:

• The proposed deep learning framework significantly enhances the accuracy of dual-comb absorption spectroscopy data processing.
• The method demonstrates practical potential for precise quantification in atmospheric measurement and industrial monitoring.
• Further development with larger networks and datasets could extend its applicability.