Smoothing methods applied to dealing with heteroscedastic noise in GUMS

In order to improve detection ability and quality of resolution of overlapping peaks with low signal-to-noise (SNR) ratio data obtained from GC/MS, the effect of heteroscedastic noise is investigated in the present paper. A new index named smoothing distortion (SD) is first developed for evaluating the smoothing efficiency. Roughness penalty smoothing method recently appearing in chemometrics is then compared with wavelet denoising technique and convolution smoothing approach under condition of heteroscedastic noise. The performance of the methods is assessed using both simulated and experimental GUMS data. The results obtained show that the roughness penalty method cannot only enhance the detection ability but also improve quality of resolved chromatographic profiles and spectra significantly for the noisy GUMS data. (C) 2002 Elsevier Science B.V. All rights reserved.