Zero-absorption-assisted multitask learning for simultaneous measurement of acetylene concentration and gas pressure from overlap-deformed spectra

Spectral deformation from the overlap of nonconstant impurities, as well as system noises, has become one of the critical factors that limit the wide applications of tunable diode laser absorption spectroscopy, particularly when multiparameter sensing highly relies on the accuracy of retrieved spectroscopy. Herein, we demonstrate a joint retrieval model that can simultaneously measure analyte concentration and gas pressure. The model is composed of expert module, gating module and tower module, which are responsible for feature extraction, expert selection and concentration/pressure output, respectively. An alternative method of introducing zero-absorption system noise into spectral simulation prepares datasets for model training, achieving measurement of acetylene concentration (50–500 ppm) and gas pressure (0.1–1 atm) with low relative deviations maintained, even with concentration-varying ammonia. This method, in a simple manner, enables accurate multiparameter sensing in many applications, where potential spectral deformation from impurities exists.