A scalable and adaptable probabilistic model embedded in an electronic nose for intelligent sensor fusion

Abstract

The electronic nose (eNose) mimicking the mammalian olfactory system has each sensor respond to different odours to different extents. However, the real environment contains multiple odours interacting with each other, such that the sensory signals of an eNose exhibit high variability and the responses to different odours usually overlap with each other. This paper presents the development of a scalable and adaptable probabilistic model for recognising eNose signals and realising the model as customer-designed integrated circuits. The scalability refers to interconnecting multiple chips to form a large network of probabilistic model for processing high-dimensional sensory signals, or to form a multi-expert system that is able to estimate the likelihood of belonging to different types of odours. The latter case could provide extra information for clinical diagnosis. Furthermore, on-chip adaptability is incorporated to learn new data distribution or variability in real time. By integrating the adaptable probabilistic model with eNose sensor arrays, analog-to-digital converters, and a digital processor on a single chip, an intelligent eNose microsystem will be formed for a variety of biomedical applications.