| Over the past two decades two approaches have emerged for processing multivariate measurements contaminated by noise and interferences: multivariate regression and digital filtering. These two different approaches use different technologies and mathematical notations but possess the identical goal of optimally deriving state variables from analytical measurements. In this work, the merits of multivariate regression-based calibration and prediction, and the applicability of digital filtering techniques for processing multivariate measurements are analyzed. As an improvement over existing techniques, a method called 'Optical Regression', which improves the analytical precision of estimation in quantitative analysis is presented. In this method, the regression vector obtained by multivariate calibration methods is employed as a template to optimize the data collection time at each wavelength of an unknown spectra. This in turn performs the dot product of the spectrum and regression vector on the detector, prior to digitization, thereby achieving improved precision of quantitative estimation in spectroscopic analysis. Thus, the entire digitized spectra need not be collected or post processed for prediction. Instead, all mathematical operations for regression are performed on the detector itself.; The theory of 'Optical Regression' is developed and discussed in the case of scanning and filter wheel based process analyzers. The results are supported by Monte Carlo simulations and experimental studies on the quantitative analysis of a mixture of three fluorescent dyes. A fiber optic, acousto-optic tunable filter based single channel spectrometer was used as a test bed for these studies. |
