SIGNAL PROCESSING HIERARCHIES FORPORTABLE, LOW-POWER SAW-BASED CHEMICAL SENSING SYSTEMS

摘要

This paper presents a comparison of multi-stage linear andrnsingle-stage nonlinear processing techniques for accomplishingrnchemical discrimination. Data variance, computational overhead,rnand memory storage requirements are compared between (linear)rnmultistage principal components analysis, non-linear VERIrn(visually empirical region of influence) and non-linear artificialrnneural network techniques. For the linear techniques, data variancernis reduced by 88%, compared to that of raw data during datarnpreprocessing. Computational overhead is reduced up to 82.5% andrn77% for non-linear clustering and artificial neural networkrntechniques respectively. These improvements offer clear promisernfor significant reduction in power and space consumption forrnportable chemical sensing systems design.