Effects of Bandwidth and Overlap on Multivariate Calibration Models Based on Simulated Fourier Transform Infrared Interferogram Data

Abstract

Simulated Fourier transform infrared (FT-IR) interferograms and spectra of mixture samples are used to investigate the effects of bandwidth and overlap on a quantitative analysis in which multivariate calibration models are constructed from short segments of digitally filtered interferogram data. By a comparison of calibration models constructed with interferogram data to similar models built with spectral data of varying instrumental resolution, the concepts of spectral resolution are explored in the context of an interferogram-based analysis. In this work, nominal spectral point spacings of 1, 2, 4, 8, and 16 cm⁻¹ are explored. Five data sets are constructed in which the width and overlap of an analyte and interference band are varied. Calibration models based on partial least-squares (PLS) regression are optimized for both interferogram and spectral data. Even in cases of extreme spectral overlap between the analyte and interference bands, interferogram-based calibration models constructed from segments of 151 points or less are observed to perform as well as or better than all models computed with absorbance and single-beam spectral data, regardless of instrumental resolution. These results suggest that an interferogram-based analysis is a viable option for applications in which an FT-IR spectrometer is used in a dedicated monitoring application.