Simultaneous multi-gas classification and concentration inversion using a subsecond-tunable mid-infrared filter

Simultaneous analysis of multiple gases in the 3-5 µm wavelength band is important for safety monitoring, industrial process control, and environmental sensing. Conventional non-dispersive infrared instruments generally obtain wavelength selectivity with multiple fixed filters, detector arrays, or mechanically switched filter wheels, increasing optical complexity and limiting rapid measurements of gas mixtures. This work presents an infrared multi-gas sensing method based on a dynamically tunable microelectromechanical-system (MEMS) Fabry-Pérot filter. The filter covered 3-5 µm and exhibited a measured tuning time of approximately 180 ms, enabling multichannel spectral sampling with one optical path and one detector. Single-gas measurements of CO, CO₂, and N₂O achieved detection limits of 24, 30, and 67 ppm, respectively. For mixed-gas recognition, a multi-channel voltage scan generated sparse spectral fingerprints that were mapped into a joint phasor-response feature space for three pure gases and all three binary mixtures. Leave-one-file-out cross-validation yielded a six-class classification accuracy of 80.56%. Classification results then guided effective-channel selection for a dual-domain concentration inversion model combining response-amplitude and phasor-domain spectral-shape constraints. For the spectrally overlapping CO-N₂O system, the 400/600 ppm CO/N₂O sample yielded an average relative concentration error of 2.08% and an average signal-domain relative error of 0.01% after correct class assignment. Within the calibrated concentration range, the proposed framework enables interpretable multi-gas analysis with a compact infrared sensing architecture. Further optimization and validation under variable environmental conditions are required for practical measurement scenarios.