Blind Source Separation on Non-Contact Heartbeat Detection by Non-Negative Matrix Factorization Algorithms.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Radar Technology

Background:

• Non-contact heart rate (HR) monitoring using Doppler radar is challenged by noise from respiration and body motion.
• Accurate HR estimation is crucial for various health and wellness applications.

Purpose of the Study:

• To propose a blind source separation (BSS) approach for mitigating noise in Doppler radar signals for HR monitoring.
• To enhance the resolution of heartbeat spectrum estimation using sparse spectrum reconstruction.

Main Methods:

• Utilized non-negative matrix factorization (NMF) algorithms for BSS to decompose mixed radar signals into distinct sources, including heartbeat.
• Employed hierarchical clustering to learn basis spectra (BS) and applied sparse NMF variants (sparse NMF, weighted sparse NMF) to exploit temporal sparsity of heartbeat signals.
• Developed a BSS approach with advantages including unsupervised learning via clustering, a compact demixing architecture, and single-channel input requirement.

Main Results:

• The proposed BSS method effectively mitigates noise interference in Doppler radar signals.
• Achieved high-resolution heartbeat spectrum estimation through sparse spectrum reconstruction.
• Demonstrated superior precision and robustness in HR estimation compared to existing methods, even during activities like typing.

Conclusions:

• The proposed BSS approach offers a significant advancement in non-contact HR monitoring accuracy and reliability.
• This method provides a robust solution for HR estimation in the presence of physiological and motion artifacts.
• The technique shows promise for practical applications requiring precise, unobtrusive heart rate tracking.