Repository logo
 

Motion Artifact Reduction From Non-Contact ECG Recording Systems

Files

Thesis(77.49 MB)

Date

2025

Authors

Khalili, Matin

Supervisor

GholamHosseini, Hamid
Kuo, Matthew
Lowe, Andrew

Item type

Thesis

Degree name

Doctor of Philosophy

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

With the growing demand for real-time, long-term, and remote electrocardiogram (ECG) monitoring, non-contact capacitive sensors have emerged as a compelling solution due to their safety, comfort, and suitability for continuous use. However, a significant challenge remains; motion artifacts (MAs) caused by body movements and physiological vibrations. Such artifacts can significantly compromise diagnostic accuracy and may even lead to misinterpretations of cardiovascular conditions. To address this, this PhD research begins by demonstrating the feasibility of capacitive ECG acquisition from the head area to broaden its application scope, then transitions to a torso-worn chest strap system for practical implementation and MA reduction. It also includes a structured review of MAs in capacitive ECG systems, analyzing their mechanisms and mitigation strategies, with focused attention on electrode-tissue impedance (ETI)-based reference signal techniques. To further this development, the study tackles this challenge by developing novel techniques to reduce MAs and improve the quality of capacitive ECG signals. The Adaptive Noise Cancellation (ANC) technique is selected for MA mitigation, using a motion-correlated Reference Input Signal (RIS) and an Adaptive Filter (AF) to estimate and subtract MAs. A full-stack capacitive single-lead system is designed, spanning hardware, analog front-end design, and advanced signal processing. In this system, ECG and a highly correlated ETI-based RIS are simultaneously acquired through an efficient voltage injection network, offering improved performance compared to previous studies. In the first phase, the RIS is fed to a Recursive Least Squares adaptive filter to suppress MAs. Signal enhancement is evaluated using root mean square (RMS) reduction, standard deviation (STD) reduction, and signal-to-noise ratio (SNR) per peak. The system demonstrates strong performance, achieving RMS, STD, and per-peak SNR improvements of 17.9 dB (650.2%), 4.24 dB (36.8%), and 2.8 dB (15%), respectively. In the subsequent phase, a Variable Step-Size Least Mean Squares with Correlation Power Boosting (VSS-LMS-CPB) adaptive filter is proposed. Benchmarking against seven established AFs confirms its superior performance in dynamically adapting step size, maintaining stability, preserving ECG morphology, and robustly handling both intra- and inter-experiment MA variability, fluctuating RIS–MA correlations, and abrupt high-amplitude artifacts, all without instability or over-adaptation. Although ETI-based RIS methods have not been widely assessed across diverse participants and conditions, the system is validated through both controlled and realistic motion experiments involving three participants. A total of nine motion scenarios, including stationary and dynamic conditions, are tested to assess generalizability and robustness. Performance is evaluated using correlation coefficients and RMS improvement, both showing marked enhancements. Collectively, the results confirm the effectiveness of the RIS-based framework and VSS-LMS-CPB algorithm in significantly reducing MAs while preserving ECG integrity. The system effectively suppresses MAs, achieving an RMS improvement of up to 11.12 dB and reducing ECG-RIS correlation from 0.866 to 0.107. By preserving full ECG morphology, including P, QRS, and T waves under motion conditions, this work advances the feasibility of wearable, non-contact ECG monitoring for telemedicine and continuous health tracking. The research contributes to biomedical signal processing by introducing practical, high-performance solutions for artifact mitigation in challenging non-contact recording environments.

Description

Keywords

Source

DOI

Publisher's version

Rights statement

Permanent link

http://hdl.handle.net/10292/20107

Collections

Doctoral Theses