Practical Methods for Noise Removal: Applications to Spikes, Nonstationary Quasi-Periodic Noise, and Baseline Drift

Abstract

A new approach to signal processing of analytical time-domain data is presented. It consists in identifying the types of noise, characterizing them, and subsequently subtracting them from the otherwise unprocessed data set. The algorithms have been successfully applied to three classes of noise commonly found in analytical signals: spikes, ripples, and baseline drift. Traditional filters have been used as an intermediary step to detect and remove spikes in the signal with 96.8% success. Adaptive ensemble average subtraction has been developed to remove nonstationary ripples that have similar time scales as the signal of interest. This method increased the signal-to-noise ratio by up to 250% and led to minimal distortion of the signal, unlike conventional Fourier filters. Finally the removal of baseline drift has been achieved by subtraction of a mathematical model for the baseline. These three methods are generic, computationally fast, and applicable to a wide range of analytical techniques. Full Matlab codes and examples are included as Supporting Information.