TL;DR
This paper introduces a novel WiFi-based breathing monitoring system that leverages the trajectories of Channel State Information (CSI) ratios on the I/Q plane. By analyzing these trajectories and filtering out motion interference, the system achieves highly accurate, non-contact breathing monitoring, offering a more robust and less intrusive alternative to traditional methods.
What Problem Does This Paper Address?
Accurate breathing monitoring is vital for health assessment, particularly for conditions like sleep apnea. Traditional methods often rely on wearable devices, which can be uncomfortable, expensive, and inconvenient for long-term use. While WiFi-based sensing offers a promising non-contact alternative, existing approaches are highly sensitive to body movements, which can mask subtle breathing signals and lead to inaccurate breathing cycle counts. This paper aims to overcome this limitation by developing a more robust WiFi sensing method that can effectively distinguish breathing signals from motion artifacts.
Methods
The core of this research lies in analyzing the trajectories of CSI ratios on the I/Q plane. Instead of relying on individual CSI amplitude or phase values, the authors track how the ratio of CSI between different WiFi links evolves over time.
Key methodological contributions include:
- CSI Ratio Calculation: To mitigate random phase shifts inherent in WiFi devices, the paper calculates the ratio of CSI between different links. This ratio is designed to cancel out common phase errors.
- Dispersion Model for Motion Interference Filtering: A novel dispersion formula is proposed to quantify the spread of CSI ratio sample points on the I/Q plane within a time window. This allows for the identification and filtering of points heavily influenced by significant body movements, thus cleaning the data for breathing analysis.
- CSI Ratio Trajectory Curve Fitting: A spline curve fitting method is employed to smooth the trajectories of the filtered CSI ratio points. This process helps to clearly identify inflection points that correspond to breathing transitions (exhalation to inhalation and vice versa).
- Breathing Segment Extraction Algorithm: Based on the identified inflection points from the fitted curves, a dedicated algorithm is developed to extract breathing segments and reconstruct the breathing waveform.
Datasets
Not discussed in the paper. The experiments were conducted using custom setups with WiFi routers and wearable respiratory sensors for ground truth.
Key Insights and Ideas
- CSI Ratio Trajectories as a Robust Breathing Signature: The paper highlights that the trajectory of CSI ratios on the I/Q plane, rather than individual signal components, provides a more stable and distinctive signature for breathing, even in the presence of some motion.
- Turning Points in Trajectories are Key: Significant angular turns in the CSI ratio trajectories on the I/Q plane are identified as crucial indicators of breathing transitions (exhalation/inhalation).
- Dispersion as a Motion Artifact Detector: Quantifying the "dispersion" of CSI ratio points on the I/Q plane offers an effective way to detect and filter out data corrupted by substantial body movements.
- Complementarity of Amplitude and Phase is Implicit: While not explicitly stated as a primary focus, the use of CSI ratios inherently combines amplitude and phase information from different links, potentially offering a more balanced approach than methods relying on single-link amplitude or phase alone.
Key Results
The proposed method demonstrated significantly enhanced accuracy and reduced error in breathing monitoring compared to existing WiFi-based methods.
- High Accuracy: Experimental results showed an Average Accuracy (ACC) above 93% across various chest orientations and monitoring distances.
- Low Error Rates: The Mean Absolute Error (MAE) was consistently low, and the Absolute Error Rate (AER) was kept below 0.35 bpm in test orientations.
- Robustness to Motion: The dispersion-based filtering effectively reduced interference from large movements, leading to more reliable breathing signal extraction.
- Effectiveness Across Orientations and Distances: The system performed well even when the human chest orientation varied (30, 60, 90 degrees) and at different monitoring distances (1m, 2m, 3m), indicating practical applicability.
Limitations and Future Work
- Sensitivity to Significant Movement: The paper acknowledges that while the method filters out large motion interference, its performance is optimal when the subject remains still. Continuous, significant movement by the subject can still reduce monitoring accuracy for all methods, including theirs.
- Future Work: The authors suggest exploring novel approaches to address the limitation of continuous subject activity. This includes investigating ray tracing models and utilizing fusion technology with multiple antennas from different angles to improve robustness in dynamic scenarios.
Summary updated