Mid-infrared laser-based sensors are commonly used to detect and quantify many chemical species for environmental, industrial, defense, and security applications. Data-driven approaches, including machine learning and information theory, can be applied to photonics-based sensors to quantify drifts and improve precision. These methods are used to classify signals from rotational-vibrational absorption spectra of Nitrous oxide (N2O) in the 4.3 m region of the spectrum. The detection method utilizes the structural complexity of wavelength modulation spectroscopy signals and information encoded in the spectra. We create our basic training models by simulating temperature, pressure, density fluctuations effects, and molecular transition line broadening of a Voigt lineshape profile. Instrument (laser and detector) noise optical fringing effects can be incorporated in the models. The paper shows that signal variations due to Trace gas density fluctuations and molecular collision dynamics can be discriminated from instrument drifts. The proposed methodology can be used to accurately predict, detect, and evaluate short-term and long-term drifts in sensing systems which can be integrated with the conventional Allan variance methods. We demonstrate this methodology by high-precision sensing of rotational-vibrational transitions of Nitrous oxide and carbon monoxide using an interband cascade laser operating at a 4.3 m spectral region.
|