Designing Ideal Denoisers for Chromatographic and Spectroscopic Signals
- 1Department of Chemistry and Biochemistry
- 2University of Texas at Arlington
Abstract
Modern instrumental analyses have different types of noise superimposed on the true signal, from low frequencies (baselines and colored noise) to very high-frequency components like sharp spikes. Convolution based denoisers or frequency domain filters inadequately handle colored noise and induce baseline oscillations while broadening narrow peaks. Herein, two new dataadaptive denoising methods are developed for chromatographic and spectroscopic data. The goal is to suppress noise without appreciably distorting peak shape or quantitative information (area, location, and variance). The proposed methods are posed as optimization problems inspired by the Bohlmann-Whittaker smoother. The objective function consists of a data fidelity term and adaptive second difference and bi-scale log-cosh penalties. The latter term suppresses pink or colored noise effectively, while adaptive functions protect narrow and weak peaks. The second method is designed for intensity-dependent additive and proportional noise and combines an inverse hyperbolic sine variance-stabilizing transformation with adaptive second-difference regularization. Non-uniform sampling rates are tolerated by this second smoother. In simulated chromatograms containing white and pink noise, the proposed adaptive method reduced the reconstruction error by approximately 40-50% relative to Gaussian weighted moving average and penalized least squares smoothing. The resolution of closely spaced peaks was maintained, and a weak peak with a weak peak with a signal-to-noise ratio near 5 was recovered. The proposed denoisers were evaluated on real data from GC-IMS, LC-MS/MS, Raman spectra relevant to planetary studies, X-ray photoelectron spectroscopy of graphitic surface, and overloaded LC-MS instruments as test cases with and without baseline correction. Most denoiser parameters are derived from noise estimated from a single chromatogram or spectrum; no training or learning algorithms are required. The denoisers are fast and can be extended to 2D data.
Related
- Extraction of some pesticides from grape juice samples using an amide-based polymer as a sorbent followed by gas chromatographic determination
- Guiding chromatographic purification through data-driven condition recommendations
- Application of Hansen Solubility Parameters for Green Solvent Extraction of Flavonoids from Oroxylum indicum (L.) Kurz and Chromatographic Profiling Across Plant Organs
- Chromatographic Peak Shape beyond the High-Péclet Approximation: Analytic Time-Domain Expression and Derivatives
- Multi-Dimensional Benchmarking of Liquid Chromatographic Methods for Parabens Determination in Cosmetic Products: Toward Sustainable Analysis
BibTeX
@misc{wahab2026designing,
title = {{Designing Ideal Denoisers for Chromatographic and Spectroscopic Signals}},
author = {M. Farooq Wahab},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15009782/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009782/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Designing Ideal Denoisers for Chromatographic and Spectroscopic Signals
AU - M. Farooq Wahab
PY - 2026
DA - 2026/10/02
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009782/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009782/v1
AB - Modern instrumental analyses have different types of noise superimposed on the true signal, from low frequencies (baselines and colored noise) to very high-frequency components like sharp spikes. Convolution based denoisers or frequency domain filters inadequately handle colored noise and induce baseline oscillations while broadening narrow peaks. Herein, two new dataadaptive denoising methods are developed for chromatographic and spectroscopic data. The goal is to suppress noise without appreciably distorting peak shape or quantitative information (area, location, and variance). The proposed methods are posed as optimization problems inspired by the Bohlmann-Whittaker smoother. The objective function consists of a data fidelity term and adaptive second difference and bi-scale log-cosh penalties. The latter term suppresses pink or colored noise effectively, while adaptive functions protect narrow and weak peaks. The second method is designed for intensity-dependent additive and proportional noise and combines an inverse hyperbolic sine variance-stabilizing transformation with adaptive second-difference regularization. Non-uniform sampling rates are tolerated by this second smoother. In simulated chromatograms containing white and pink noise, the proposed adaptive method reduced the reconstruction error by approximately 40-50% relative to Gaussian weighted moving average and penalized least squares smoothing. The resolution of closely spaced peaks was maintained, and a weak peak with a weak peak with a signal-to-noise ratio near 5 was recovered. The proposed denoisers were evaluated on real data from GC-IMS, LC-MS/MS, Raman spectra relevant to planetary studies, X-ray photoelectron spectroscopy of graphitic surface, and overloaded LC-MS instruments as test cases with and without baseline correction. Most denoiser parameters are derived from noise estimated from a single chromatogram or spectrum; no training or learning algorithms are required. The denoisers are fast and can be extended to 2D data.
ER -