High-speed odor sensing using miniaturized electronic nose

Abstract

Animals have evolved to rapidly detect and recognize brief and intermittent encounters with odor packages, exhibiting recognition capabilities within milliseconds. Artificial olfaction has faced challenges in achieving comparable results-existing solutions are either slow; or bulky, expensive, and power-intensive-limiting applicability in real-world scenarios for mobile robotics. Here, we introduce a miniaturized high-speed electronic nose, characterized by high-bandwidth sensor readouts, tightly controlled sensing parameters, and powerful algorithms. The system is evaluated on a high-fidelity odor delivery benchmark. We showcase successful classification of tens-of-millisecond odor pulses and demonstrate temporal pattern encoding of stimuli switching with up to 60 hertz. Those timescales are unprecedented in miniaturized low-power settings and demonstrably exceed the performance observed in mice. It is now possible to match the temporal resolution of animal olfaction in robotic systems. This will allow for addressing challenges in environmental and industrial monitoring, security, neuroscience, and beyond.

Fig. 1.. Electronic nose and odor delivery system.

(A) Decoding temporal information of odor plumes requires fast sensing. Top: Two sequential TiCl₄ smoke plume photographies, shifted and superimposed, provided by P. Szyszka. Bottom: Dual-PID recordings of source-separated odor plumes, from Ackels et al. (24). Plume and sensor location (red) for illustrative purposes only. (B) Experimental setup with odor delivery device and electronic nose. (C) Electronic nose circuitry. (D) Microscopy image of the MiCS-6814 NH3 sensor with its housing removed. (E) Heater modulation cycle in ambient air. (F) PID and flow meter traces for a 20-Hz stimulus. Solid/faded (occluded) traces for mean/SD. of five trials. (G) Resulting olfactometer temporal fidelity, for various frequencies. Odorants abbreviations: IA, isoamyl acetate; EB, ethyl butyrate; Eu, cineol; 2H, 2-heptanone; blank, odorless control.

Fig. 2.. Rapid heater modulation enables robust data features.

(A) Sensor resistance of four MOx sensors with 20-Hz hotplate temperature modulation, responding to a 1-s odor pulse of IA (green background). (B) Fifty-millisecond data feature for different gases, selected between 500 and 550 ms after odor pulse onset. Raw sensor response (upper) and normalized sensor response (lower; see Materials and Methods for normalization procedure). Time shifted by cycle phase ρ w.r.t. odor onset, for visual guidance only. (C) Principal components analysis, explained variance (most left) and projections, and (D) t-SNE visualization, for the set of normalized data features extracted between 500 and 1000 ms after odor onset. (E) Accuracy scores for a k-NN classifier trained on 50-ms data features from 1000-ms odor pulses at full concentration, and tested on 50-ms features from 1000-ms odor pulses at different concentration levels (tuned by adjusting the duty cycle of the microvalves).

Fig. 3.. Electronic nose can classify short odor pulses on the basis of 50-ms data features.

(A) Feature labels for the training set were phase aligned in relation to odor on- and offset. Features that overlapped with transition periods were not considered for training (“rejected”; see Materials and Methods for parameters). (B) Odor stimulus classification over time for odor pulses of various lengths (10 to 1000 ms), as predicted by a RBF-kernel SVM classifier trained on 50-ms features from 1000-ms second odor pulses. Shown here are 1000-ms pulses. For visual clarity only, the trials are sorted by odor, and within each odor are sorted by phase w.r.t. stimulus onset. (C) Classification correctness over time (evaluated via the true odor presence), for different pulse durations. (D) Test accuracy, onset time and offset time for the prediction over time described in (B) and (C). Onset and offset were extracted using time-to-first nonblank and blank prediction, respectively, and shown here with respect to theoretical odor onset and offset.

Fig. 4.. Decoding temporal structure of rapidly switching odors.

(A) Odor valve commands. (B) PID response. (C) Electronic nose response. (D) Frequency, magnitude, and phase of the dominant spectral peaks. Thick lines, means of corresponding trials; thinner lines, single trials. (E) Accuracies for modulation frequency classification. (F) Accuracies for binary modulation frequency classification. (G) Accuracies for binary modulation mode classification (corr. versus anticorr.). (H) Subset of (F) for IA-EB, for mouse performance comparison [described in detail by Ackels et al. (24)]. (I) Subset of (G) for IA-EB, for mouse performance comparison. Panels (A) to (E) show representative trials only. For (E) to (I), electronic nose accuracy mean and SD (clipped at 1.0) arise from repeated training and testing with different random seeds.