Sensor array optimization using variable selection and a Scanning Light Pulse Technique

Abstract

In the design of a chemical sensor, the constructor has several degrees of freedom setting parameters that influence the final characteristics of the component. For applications where several sensors are required, the number of possible parameter configurations increases dramatically. The work of configuring a sensor array is therefore tedious and many test sensors may need to be processed before a final configuration is found. The Scanning Light Pulse Technique (SLPT) is a technique for investigating insulator–semiconductor interfaces and can be used to scan surfaces with non-uniform properties. Thereby a ‘virtual’ pool of test components can be evaluated simultaneously eliminating the need for processing individual test sensors. We report here on a method combining SLPT with algorithmic sensor selection techniques. This is a powerful combination providing the user with a candidate array configuration containing combinations of sensors optimal for the current application and data analysis algorithms. The need to process many individual test sensors is eliminated and the only sensor components that must be produced are those included in the final array. The selection techniques evaluated here are based on forward selection and Asymmetric Class Projection (ACP), Canonical Correlation Analysis (CCA), Linear Discriminant Analysis (LDA), and Mutual Information (MI). The suggested method is successfully evaluated using an experiment in which the purpose was to find means to detect small amounts of hydrogen in a background dominated by an interfering gas, in this case ammonia. In this particular study, the selection techniques based on ACP and CCA showed the most promising result.