Exploring the role of signal pollution rate on the performance of despiking velocity time-series algorithms

Abstract

Despite the existence of numerous studies on signal velocity time-series filtering methods, the influence of the quantity of invalid data on the performance of the filtering algorithm remains uncertain. Our current study aims to address this gap by introducing a novel concept called the "Pollution Rate" which defines as the level of signal contamination irrespective of the filtering algorithms. Through this concept, we explore how the number of invalid data affects the performance of the filtering algorithm. For this purpose, a notable number of ADV measured data set in various flow types with different Pollution Rate (PR) have been employed. The findings reveal that the PR significantly influences the choice of an appropriate filtering method. Indeed, for the low polluted signals (less than 4–5 % of the data are spikes) and the polluted signals (more than 4–5 % and less than 10–15 % of the data are spikes), the Phase-Space thresholding algorithm together with the 12 points polynomial interpolation leads to the best results. While, for highly polluted signals (over 15 % of the data are spikes), the best results have been obtained using Kernel density estimation with 12 points polynomial replacement. Employing the sensitivity analysis of turbulent characteristics with respect to the filtering techniques show that although for low polluted signals, these parameters are insensitive, most of the polluted and highly polluted signals parameters especially the flow momentums and fluxes are drastically changed through the filtering process.