Active Concentration-Independent Chemical Identification With a Tunable Infrared

Abstract — This paper presents an active-sensing framework for concentration-independent identification of volatile chemicals using a tunable infrared interferometer. The framework operates in real time to generate a sequence of absorption lines that can best discriminate among a given set of chemicals. The active-sensing algorithm was previously developed to optimize temperature programs for metal-oxide chemosensors. Here, we adapt it to tune a nondispersive infrared spectroscope on the basis of a Fabry–Pérot interferometer (FPI). We also extend this framework to allow the identification of chemical samples irrespective of their concentrations. Therefore, we use nonneg-ative matrix factorization to create concentration-independent absorption profiles of different chemicals, and then employ linear least squares to fit sensor observations to the response profiles. We tested the framework on a simulated classification problem with 27 chemicals and compared against a passive sensing approach; the active-sensing consistently outperformed the passive sensing in terms of classification performance for various sensing budgets and at various levels of sensor noise. We also validated the approach experimentally using a commercial FPI sensor and a database of eight household chemicals. Our results show that the method can predict the sample identity irrespective of concentration. Index Terms — Active sensing, concentration normalization, Fabry–Perot interferometer, tunable sensors