Heart Rate Variability Monitoring Based on Doppler Radar Using Deep Learning.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Radar Technology

Background:

• Non-contact vital sign detection using microwave Doppler radar shows promise.
• Current radar-based heart rate (HR) and heart rate variability (HRV) methods require over 10 seconds of data, causing latency and inaccurate HRV estimates.

Purpose of the Study:

• To develop a novel network for precise HR estimation and HRV monitoring from concise radar time series.
• To reduce detection latency and improve HRV accuracy in non-contact vital sign monitoring.

Main Methods:

• A new network integrating a frequency representation module and a residual in residual module was developed.
• The network transforms time-domain radar signals to the frequency domain for high-resolution spectrum representation.
• The model was validated using simulation, Frequency-Modulated Continuous-Wave (FMCW) radar, and Continuous-Wave (CW) radar data with a 4-second data length.

Main Results:

• The algorithm achieved an average heart rate measurement accuracy above 95% with shortened data length.
• Estimation accuracy for HRV was maintained despite the reduced data duration.
• The developed network significantly reduced latency compared to existing methods.

Conclusions:

• The study introduces an efficient algorithm for HR and HRV estimation using concise radar data.
• This method offers significant practical value for non-contact vital sign detection systems.
• The approach enhances the feasibility of real-time, accurate vital sign monitoring.