Skip to Main Content
Two fundamental problems in olfactory signal processing is the large time constant and the large variance in the odor receptor code. Depending on the sensing technology and the analyte under investigation, obtaining a steady-state pattern from a sensor array may take minutes, yet still be unreliable. Therefore, odors are encoded in a spatio-temporal fashion in the nature, a task that fits very well in Cellular Neural Network (CNN) paradigm. Inspired by the generic insect olfactory system, we propose a CNN-based signal conditioning system that can be directly applicable on raw sensor data in real time. We interface the system with a Support Vector Machine (SVM) classifier, which maps the dynamically-encoded odor to an identity, and demonstrate the recognition system on a dataset recorded from a metal-oxide odor sensor array.