Practical Algorithm for Reducing Convex Spike Noises on a Spectrum

Quantitative data analysis and high-sensitivity measurements by use of Raman and infrared spectroscopy often suffer from noisy spikes such as those due to cosmic rays and water vapor. Since these spikes are unidirectional and isolated, the conventional smoothing techniques do a poor job of removing them. However, a small modification can improve these smoothing techniques significantly. In this paper, we present a simple denoising technique for a single-scan spectrum that is corrupted by convex spikes. In general, the noisy spikes arising from cosmic rays or water vapor have a much narrower bandwidth compared with target “informative” bands in Raman and infrared spectra. This means that these noisy spikes can be separated from the target bands by means of the difference in the bandwidths. The proposed method employs the moving window averaging procedure to distinguish and separate the convex spike. The proposed algorithm allows us to take away the convex spikes from measured spectra without preparing multiply recorded spectra and (much) biasing the measured spectrum.