Design of a low-power, portable sensor system using embedded neural networks and hardware preprocessing

This paper addresses the issue of data analysis for real-time identification of volatile chemicals in a portable sensor unit. One targeted application is breath alcohol detection, but there are quite a few other potential applications in areas as diverse as food safety, military reconnaissance, and consumer products manufacturing. It is shown that correct odour recognition rates as high as 96% can be achieved using mixed analog and digital VLSI circuitry for on-board data preprocessing, together with pulse coupled neural network and/or principal components analysis for feature extraction. Finally, multilayer perceptron and radial basis function neural networks are used for pattern recognition