Unleashing the power of three-dimensional kernel density estimation for Doppler Velocimeter data despiking

Abstract

Spikes in data compromise the accuracy of Acoustic Doppler Velocimeter (ADV) measurements, and existing despiking algorithms are influenced by noise levels. However, the lack of standardized definitions for noise levels makes it challenging to compare algorithm performance. This study proposes a novel signal despiking algorithm, 3d-KDE, based on three-dimensional kernel density estimation that effectively removes spikes across a broad range of noise levels. We evaluated the performance of the 3d-KDE algorithm using datasets from a highly turbulent free surface flow and compared it with other despiking algorithms. Results show that the 3d-KDE algorithm effectively detects and removes spikes from ADV signals, regardless of contamination levels, and is adaptable to varying flow conditions. Incorporating cross-correlations between velocity components further improves the algorithm's performance. Our study highlights the potential of the 3d-KDE algorithm to improve the accuracy of ADV measurements and advance the field of fluid dynamics research.