A study has been made of the use of polynomial curve fitting for removal of nonlinear background and high-spatial-frequency noise components from Raman spectra. Two variations on polynomial curve fitting through a least-squares calculation are used. One, involving fitting data x values to corresponding y values, was used to approximate background functions, which are subtracted from the original data. For smoothing, a reference matrix of six vectors that contains a unity d.c. level, a ramp made up of x values, a quadratic made up of x2 values, etc., is fitted to a section of data. The reference vectors are scaled by the fit values and added to give the smoothed estimate of a spectral peak. It is demonstrated, with factor analysis as a test procedure, that the background removal procedure does remove nonlinearities that were present in the original data. The smoothing procedure rejects high-spatial-frequency noise without introducing detectable nonlinearities.

We have examined the effect of intensity-axis correction on noise in white light spectra recorded with charge-coupled device (CCD) detectors. Measurements were made with five detectors in Raman spectrometers. Detectors were both liquid-nitrogen and thermoelectrically cooled devices and one room temperature device. Both random and pattern noise have been considered. We used cross-correlation of noise sets to provide an indicator of a fixed pattern in the spectra and an assessment of the efficacy of the correction procedure in removing this pattern. For four of the five detectors intensity-axis correction provided a significant improvement in signal-to-noise ratio. Correction was particularly important for measurements made with lower-cost CCD detectors of the kind proposed for process control instruments.