An On-Chip Multi-Class Support Vector Machine Applied to Portable Electronic Nose Data Classification

In this paper, a multiple-class support vector machine chip applied to a portable electronic nose system is presented. The multiple-class method of the classifier was implemented with a "one-versus-one" method, and the kernel function for SVM is Gaussian kernel, which is highly common and usually demonstrates high quality performance. The power consumption of the chip is 118 μW; therefore, this low power design is highly suitable for portable applications. The feasibility of this work was verified by classifying fruit gas data, which were collected in a gas experiment. From the post-simulation results, all of the parameters could be successfully trained, and all testing data were dispensed to correct categories.