Cancellation of chemical backgrounds with generalized Fisher's linear discriminants

This article presents a signal-processing technique capable of canceling the effect of background chemicals from the multivariate response of a sensor array. We propose a generalization of the Fisher's eigenvalue solution that minimizes the discrimination between undesirable chemicals and a neutral reference. The proposed technique is a generalization of an earlier model that was limited to the removal of single volatiles. A reformulation of class memberships allows the new model to cancel the effect of both single and mixture backgrounds. The model is validated on experimental data from an array of temperature-modulated metal-oxide sensors exposed to binary and ternary mixtures.