Linear predictive spectral coding and independent component analysis in identifying gasoline constituents using infrared spectroscopy

An advanced spectral encoding method used in combination with independent component analysis (ICA) yields promising results in identifying refinery fractions contained in commercial gasoline mixtures based on infrared (IR) spectroscopy data. Previous work has shown how the signatures of the gasoline constituents can be recovered by solely relying on the IR spectra of their mixtures using ICA as a blind separation procedure. The present methodology encodes peak information from the spectra in linear predictive (LP) coefficients which are subsequently transformed into line spectrum frequencies (LSF). Such encoded spectra have a drastically reduced size (to 1/20 of the original size) while preserving the crucial peak information that characterizes each constituent. Source identification is then established by simply computing a Euclidean distance measure between the corresponding LSF of the gasoline constituents predicted by ICA and the LSF available from the spectral library of candidate matches. High correlation scores are associated with successful identification of source spectra, and this indicates that the present methodology can be employed as an effective tool in fingerprinting applications.