WiFi-Based Real-Time Breathing and Heart Rate Monitoring during Sleep

Yu Gu, Xiang Zhang, Zhi Liu, Fuji Ren

Abstract

This paper introduces a non-invasive, cost-effective system for real-time monitoring of vital signs during sleep using standard WiFi signals. By applying Fresnel diffraction theory to optimize antenna placement, the system captures minute chest displacements caused by respiration and heartbeats. Evaluation with five participants across multiple sleeping postures shows high accuracy, exceeding 94% for heart rate and 96% for breathing rate, without requiring wearable sensors or specialized hardware.

Introduction

Sleep-related health issues, such as sleep apnea and sudden cardiac arrest, necessitate continuous monitoring. However, traditional clinical methods like Polysomnography (PSG) and Electrocardiograms (ECG) are intrusive, requiring multiple wired sensors that disrupt natural sleep. While camera-based solutions offer a contactless alternative, they raise significant privacy concerns and are sensitive to lighting conditions.

WiFi-based sensing has emerged as a promising alternative, leveraging Channel State Information (CSI) from existing home networks. Previous research in this area often struggled with sensitivity to different sleeping postures or required expensive directional antennas. This work addresses these challenges by using Fresnel zone theory to guide a low-cost, real-time monitoring system capable of maintaining accuracy regardless of the sleeper's orientation.

Methodology

The system utilizes two mini-PCs equipped with Intel 5300 WiFi cards to extract CSI data. The core innovation lies in the application of the Fresnel diffraction model to optimize the physical setup.

Fresnel Zone Optimization

A Fresnel zone is an elliptical region between a transmitter (Tx) and receiver (Rx). Movements within these zones cause constructive or destructive interference in the radio signal. By positioning the antennas such that the subject's chest is located within a highly sensitive region of the first Fresnel zone, the system maximizes the signal-to-noise ratio (SNR) for tiny movements.

Fresnel zone model for antenna placement optimization.

The researchers found that by adjusting antenna height and using lead sheets to block certain paths, they could "force" the system into a state where even the small displacements of a heartbeat become detectable. This theory-driven approach avoids the "black box" nature of many machine learning methods.

Signal Processing Pipeline

The system processes the raw CSI through several stages:

  1. Subcarrier Selection: The system identifies and selects subcarriers with the highest variance while discarding "deaf" subcarriers that are insensitive to motion.
  2. Denoising: A Hampel filter removes outliers, followed by Butterworth bandpass filters to isolate breathing (0.25–0.5 Hz) and heart rate (1–2 Hz) frequencies.
  3. Frequency Estimation: A Fast Fourier Transform (FFT) is applied to the filtered signals to estimate the rates in real-time via Matlab.

Experimental Evaluation

The system was tested with five university students (ages 21–26) over 30-minute sleep sessions. Four standard postures were evaluated: prone (stomach), supine (back), facing left, and facing right. Ground truth was established using a physical accelerometer for breathing and a fingertip pulse oximeter for heart rate.

Antenna layouts optimized for various sleeping postures.

To handle the reduced chest displacement visible when a person lies on their side, the researchers optimized antenna configurations to capture mediolateral (side-to-side) movements more effectively, as shown in the different layouts (a, b, c) in the figure above.

Results

The system achieved high performance across all tested conditions:

  • Breathing Rate: 96.6% accuracy with an average error of 0.575 bpm.
  • Heart Rate: 94.2% accuracy with an average error of 3.9 bpm.
  • Posture Resilience: While the supine position yielded the highest accuracy, the system remained robust across other postures. The "facing left recumbent" position was the most challenging for heart rate detection but still provided usable data.
  • Real-time Capability: The system requires approximately 40 seconds of data to stabilize, after which it provides continuous real-time updates.

Conclusion

This work demonstrates that high-fidelity vital sign monitoring is possible using commodity WiFi hardware and a theory-driven physical setup. By leveraging Fresnel zone optimization, the system overcomes the limitations of previous contactless methods regarding posture and hardware cost. Future work will focus on developing algorithms to classify macro-movements (e.g., tossing and turning) to provide a more comprehensive sleep quality assessment and improve system robustness in complex multipath environments.

Summary updated

BibTeX
@inproceedings{gu2019wifi,
  author = {Yu Gu and Xiang Zhang and Zhi Liu and Fuji Ren},
  title = {{WiFi-Based Real-Time Breathing and Heart Rate Monitoring during Sleep}},
  booktitle = {GLOBECOM},
  year = {2019},
  doi = {10.1109/globecom38437.2019.9014297},
}
Figure