Wave Dynamic Time Warping Algorithm for Periodic Signal Similarity Estimation

Similarity estimation between periodic biological signals is crucial for vital sign-sensing applications. In this paper, we propose a novel approach for the accurate alignment and similarity estimation of such signals, named wave Dynamic Time Warping (wave DTW), and its variant, wave derivative DTW. Proposed methods take advantage of the inherent structure of periodic signals and align signal segments rather than entire sequences. Wave DTW employs a two-dimensional feature vector, derived from the signal's amplitude envelope and phase, to perform segment point-by-point alignment using DTW. Validation of the proposed algorithms based on Beth Israel Deaconess Medical Center Photoplethysmogram and Respiration Dataset (BIDMC PPG and Respiration Dataset) demonstrates that wave DTW and wave derivative DTW significantly outperform state-of-the-art algorithms DTW and derivative DTW in terms of path misalignment and path warpings and provide a more intuitive signal alignment.