Source-linked AI summary
UbiBreathe: A Ubiquitous non-Invasive WiFi-based Breathing Estimator
Heba Abdelnasser, Khaled A. Harras, Moustafa Youssef
TL;DR
Respiratory monitoring is often invasive or confined to medical facilities, motivating more ubiquitous approaches. UbiBreathe uses standard WiFi equipment to estimate breathing and detect apnea, achieving less than 1 bpm error and more than 96% apnea-detection accuracy in reported deployments.
Problem
Existing respiratory-monitoring solutions are usually invasive and/or limited to medical facilities, while breathing rates and patterns support diagnosis and potential avoidance of health problems.
Method
UbiBreathe analyzes breathing-related changes in WiFi received signal strength using off-the-shelf devices and processing modules for noisy signals, movement, interference, and apnea detection.
Results
UbiBreathe achieved less than 1 bpm breathing-rate error and more than 96% apnea-detection accuracy across device-on-chest and hands-free scenarios.
Takeaways & Limitations
The reported results support UbiBreathe as a system for ubiquitous, non-intrusive respiratory monitoring using standard WiFi equipment.
Takeaways & Limitations
The authors are expanding UbiBreathe to detect exact apnea types and conduct clinical trials with real patients.
Abstract
from arXiv · showhide
Monitoring breathing rates and patterns helps in the diagnosis and potential avoidance of various health problems. Current solutions for respiratory monitoring, however, are usually invasive and/or limited to medical facilities. In this paper, we propose a novel respiratory monitoring system, UbiBreathe, based on ubiquitous off-the-shelf WiFi-enabled devices. Our experiments show that the received signal strength (RSS) at a WiFi-enabled device held on a person's chest is affected by the breathing process. This effect extends to scenarios when the person is situated on the line-of-sight (LOS) between the access point and the device, even without holding it. UbiBreathe leverages these changes in the WiFi RSS patterns to enable ubiquitous non-invasive respiratory rate estimation, as well as apnea detection. We propose the full architecture and design for UbiBreathe, incorporating various modules that help reliably extract the hidden breathing signal from a noisy WiFi RSS. The system handles various challenges such as noise elimination, interfering humans, sudden user movements, as well as detecting abnormal breathing situations. Our implementation of UbiBreathe using off-the-shelf devices in a wide range of environmental conditions shows that it can estimate different breathing rates with less than 1 breaths per minute (bpm) error. In addition, UbiBreathe can detect apnea with more than 96% accuracy in both the device-on-chest and hands-free scenarios. This highlights its suitability for a new class of anywhere respiratory monitoring.
1. INTRODUCTION
UbiBreathe addresses limitations of conventional respiratory monitoring with a software-only WiFi system that estimates breathing and detects apnea in device-on-chest and hands-free settings. Its evaluation reports less than 1 bpm breathing-rate error and apnea detection accuracy reaching 96%.
- Motivation: Existing hospital monitors require attached devices that can annoy patients, restrict movement, and hinder remote monitoring.Phone-camera respiration monitoring also requires sufficient light, limiting use during sleep.
- Approach: UbiBreathe uses breathing-induced changes in WiFi RSS to provide affordable, pervasive, non-intrusive respiration monitoring without special hardware.The system can monitor multiple persons, detect anomalies, and display the breathing signal in realtime.
- Hands-free operation: The breathing effect remains observable when a person lies in the AP-device line of sight without carrying a device.The paper identifies sleeping-apnea diagnosis and monitoring a sleeping baby as example applications.
- System design: UbiBreathe includes processing for interfering humans, sudden movement, outliers, and realtime apnea alarms derived from loss of the extracted breathing signal.These modules are designed to extract respiratory information from noisy RSS measurements.
- Evaluation: Less than 1 bpm error was achieved for different breathing rates, while apnea detection reached 96% accuracy using RSS from five overheard APs.The system was implemented on standard APs, laptops, and cell phones and evaluated in an apartment and engineering-building testbeds.
2. RELATED WORK
Prior respiratory monitors often depend on specialized contact or RF hardware, limiting comfort, mobility, range, cost, or remote use. UbiBreathe instead applies ubiquitous off-the-shelf WiFi devices to respiration monitoring and apnea detection.
- Respiration monitoring systems: Hospital respiration monitors commonly use contact-based masks, nasal cannulas, or sensors attached to the body.These devices may annoy users, limit movement, and be unsuitable for remote home monitoring.
- Respiration monitoring systems: Phone-camera respiration monitoring analyzes chest motion but requires light, restricting operation in conditions such as sleep.Camera-based and optical approaches process light-related signals to estimate physiological activity.
- Respiration monitoring systems: Contact-free RF systems use microwave Doppler, UWB, or ISM technologies but can have limited range, high cost, or specialized hardware requirements.Some systems achieve high accuracy through special frequency bands or dense deployments.
- UbiBreathe: UbiBreathe uses software-only processing with off-the-shelf WiFi devices and can estimate respiration accurately with one AP-device pair.It must nevertheless address WiFi signal characteristics and low device density.
- UbiBreathe: UbiBreathe extends ubiquitous WiFi event detection to healthcare by monitoring multiple users and detecting apnea.Each monitored user needs to carry a device for simultaneous monitoring.
3. SYSTEM OVERVIEW
UbiBreathe estimates respiration from breathing-induced fluctuations in standard WiFi signals, supporting both chest-device and hands-free line-of-sight operation. Its architecture combines extraction, robust rate estimation, apnea detection, and visualization modules.
- Signal basis: Breathing-induced chest and lung volume changes introduce a dominant periodic component into received WiFi signals.UbiBreathe analyzes these RSS fluctuations to extract breathing-pattern information.
- Architecture: The architecture contains Breathing Signal Extractor, Robust Breathing Rate Extractor, Apnea Detector, and Realtime Visualizer modules.The modules may run on the user device, in the cloud, or on a nearby local device.
- Operating modes: The same sensing principle supports hands-free operation when a person is between a transmitter and receiver in line of sight.The paper evaluates both device-on-chest and hands-free configurations.
- Processing pipeline: RSS processing first extracts breathing signals and instantaneous rates, then stabilizes readings before visualization and apnea processing.The pipeline uses sliding-window processing and a robust rate-extraction stage for more stable readings.
- Frequency filtering: FFT-based processing restricts frequencies to 0.1–0.5 Hz, corresponding to 6–30 bpm, before estimating the breathing rate.A further energy threshold removes low-amplitude frequencies from the band-limited spectrum.
4. THE UBIBREATHE SYSTEM
This section introduces the detailed descriptions of the four UbiBreathe modules presented in the system architecture. It transitions from the overview into the system-component discussion.
- The section discusses the details of the four UbiBreathe modules shown in Figure 2.
- Figure 2 provides the architecture reference for the module descriptions in this section.
- The section focuses on individual system components rather than introducing a separate system-level function.
4.1 Breathing Signal Extractor
The Breathing Signal Extractor identifies breathing from WiFi RSS by isolating dominant frequency components, reconstructing the time-domain signal, and mitigating sudden RSS changes.
- Basic Operation: The extractor applies FFT to sliding RSS windows and restricts candidate frequencies to 0.1–0.5 Hz, corresponding to 6–30 bpm.The breathing rate is selected from the frequency with maximum magnitude in this range.
- Basic Operation: The breathing rate is estimated as the frequency with maximum magnitude within the human breathing-rate range.
- Basic Operation: Low-energy frequencies below 25% of the dominant breathing-frequency amplitude are removed before inverse FFT reconstructs the breathing signal.
- Handling Sudden Changes: Within-window local mean removal reduces sudden-change effects that full-window mean removal leaves across overlapping FFT windows.Sudden RSS changes can result from user movement or interfering humans in the line of sight.
- Handling Sudden Changes: Multiple RSS streams from different heard access points can further reduce interference caused by an external user blocking one access point’s line of sight.The paper notes this approach but does not discuss it further because of space constraints.
4.2 Robust Breathing Rate Extractor
The Robust Breathing Rate Extractor smooths instantaneous estimates, suppresses outliers, and waits for consistent recent estimates before producing a stable breathing-rate reading.
- Robust Breathing Rate Extraction: The extractor fuses estimates from overlapping consecutive windows to improve robustness when a single stable reading is needed.Its submodules are an Outlier Detector and a Robustness Enhancement module.
- Outlier Detector: An α-trimmed mean filter sorts sliding-window estimates, removes the highest and lowest α values, and averages the remainder.
- Outlier Detector: For α = 0 the filter becomes a moving average, whereas α = 0.5 produces a median filter; the system sets α to 0.25.The filter is described as handling both impulse and Gaussian noise.
- Robustness Enhancement: Wavelet denoising recursively decomposes the signal through five levels, thresholds detail coefficients dynamically, and applies the inverse transform.
- Robustness Enhancement: The robustness enhancer checks whether all estimates from the previous 10 seconds agree within 0.75 bpm before accepting a saturated reading.This consistency check introduces a slight delay for the initial robust reading.
4.3 Apnea Detector
The Apnea Detector denoises the extracted breathing signal, tests ten-second windows for absent breathing, and combines decisions across access points. It uses adaptive thresholding to support apnea detection in dynamic environments.
- Apnea is defined as the cessation of oro-nasal airflow for at least 10 seconds.
- Denoising: Wavelet denoising decomposes the raw signal, thresholds noisy detail coefficients, and reconstructs the denoised signal through an inverse transform.The decomposition recurses through up to five levels, using dynamic thresholding on detail coefficients.
- Apnea detection: A ten-second sliding window raises an apnea alarm when its denoised signal range falls below threshold θ.The detector compares the window’s maximum and minimum values.
- Multi-AP fusion: Multiple access-point streams are fused by majority vote over apnea and normal-breathing states to improve accuracy and noise resilience.
4.4 Realtime Visualization Module
The Realtime Visualization Module streams UbiBreathe’s processed outputs to a user device. It presents the breathing waveform and rate while providing audible and visual apnea alarms, although specific apnea types remain future work.
- The module streams the breathing signal and breathing rate to a predefined user device such as a laptop.
- It raises audible and visual alarms when apnea is detected.
- Displaying the processed waveform can help practitioners identify abnormalities in breathing rate, rhythm, and effort.The paper states that detecting the specific apnea type is left for future work.
4.5 Discussion
UbiBreathe addresses noisy wireless conditions with filtering and supports multi-AP, parallel-user operation. Because breathing is low frequency, normal WiFi beaconing provides sufficient sampling without unusual energy demands.
- Band-pass filtering, local mean removal, and α-trimmed averaging address spectral interference, sudden signal changes, and outliers.
- Multiple access points can improve breathing-rate and apnea-detection accuracy while reducing interference from nearby humans.
- Measurements at each user’s device allow multiple users to estimate breathing rates in parallel without cross-interference.The design leverages the broadcast nature of the wireless channel.
- Normal WiFi beacon transmission is well above the Nyquist frequency for low-frequency human breathing signals.The paper therefore reports energy consumption within the normal range required for WiFi operation.
5. EVALUATION
UbiBreathe was evaluated across varied environments, sampling parameters, distances, orientations, hands-free conditions, multiple users, and apnea-detection settings. It maintained generally low breathing-rate error while showing expected trade-offs from weaker signals, longer windows, and detection thresholds.
- Sampling parameters: A 30-second FFT window at the typical 10 Hz WiFi beaconing rate produced a 0.2 bpm estimation error.Larger FFT windows improved accuracy but increased initial Robustness Enhancer latency.
- Distance: 1.6 bpm error remained achievable up to 11 m without walls and 1.7 bpm up to 8 m through walls.Accuracy decreased with distance because weaker RSS reduced signal-to-noise ratio and sensitivity to chest movement.
- User orientation: 0.24 bpm estimation error was achieved in the front orientation, while side and back orientations retained less than 2.6 bpm error.Front orientation provided stronger received signal because the body did not block the line of sight.
- Hands-free evaluation: In the hands-free scenario, increasing AP-to-device distance reduced performance, and the worst orientation accuracy was 2.6 bpm error.The hands-free evaluation varied user position, AP-to-device distance, and user orientation.
- Multiple persons: 0.9 bpm error was achieved for three concurrent users breathing at 12, 18, and 24 bpm.Standard WiFi broadcast allowed users to be monitored without interfering with one another, although higher rates were slightly harder to capture.
- Apnea detection: 92% apnea-detection accuracy was obtained for threshold ratios between 50% and 67% in both device-on-chest and hands-free scenarios.Increasing the threshold ratio reduced false negatives while increasing false positives.
- Multiple APs: With five RSS streams, apnea detection reached at least 96% accuracy with 5% false-negative and false-positive rates.A majority vote across streams selected the apnea or normal-breathing state.
- Latency: A robust single breathing-rate reading was available within 50 seconds in all scenarios, while other modules operated in realtime.The initial latency was associated with the Robustness Enhancer sub-module.
6. CONCLUSION AND FUTURE WORK
UbiBreathe provides non-intrusive breathing-rate monitoring and apnea detection using standard WiFi equipment across device-on-chest and hands-free scenarios. The authors report low breathing-rate error and high apnea-detection accuracy, while identifying exact apnea-type detection and clinical trials as future work.
- Conclusion: UbiBreathe achieved less than 1 bpm error for breathing-rate monitoring under realistic device-on-chest and hands-free deployments.The system was implemented on different off-the-shelf WiFi devices.
- Conclusion: More than 96% apnea-detection accuracy with less than 5% false-positive and false-negative rates was reported.The reported capabilities include remote monitoring, early health-issue detection, fitness monitoring, and infant sleep monitoring.
- Future work: Future work includes detecting exact apnea types and conducting clinical trials with real patients.