A wide dynamic range gas analysis model with deep learning based on cavity ring-down spectroscopy

Abstract

The cavity ring-down spectroscopy (CRDS) plays an important role in the detection of sensitive gas detection. However, the dynamic range of CRDS-based gas sensors is limited due to the reduced accuracy of the ring-down time and the occurrence of low signal intensity under high gas concentration. To overcome the limitation on the dynamic range, a CNN-assisted CRDS algorithm was proposed for gas sensing, designed to handle low and high-concentration absorption spectra simultaneously. A cavity ring-down spectroscopy gas sensing system was implemented, and CO₂ was selected as the target analyte for evaluating the performance of the constructed CNN-assisted algorithm. It is trained on a dataset including absorption spectra of low and high concentrations. The experimental results indicate the feasibility of using a neural network to assist in processing cavity ring-down spectral signals. The dynamic range of the CNN-assisted CRDS technique exceeds that of the traditional CRDS technique by an order of magnitude, improving from 4000 ppm to 40000 ppm. This method provides a way to expand the application of CRDS, especially for applications that require measuring gas concentrations with significant fluctuations, such as industrial emissions, gas leak detection, and geological hazard monitoring.