TL;DR
This paper introduces a novel, non-contact method for separating respiratory and heartbeat signals from multiple individuals using ultra-wideband (UWB) radar. By employing Empirical Wavelet Transform (EWT) with adaptive frequency boundary estimation, the system achieves high accuracy in vital sign monitoring, even with subject movement, outperforming traditional signal processing techniques.
What Problem Does This Paper Address?
Monitoring vital signs like respiration and heartbeat is crucial for healthcare, but traditional contact-based methods can be inconvenient or even risky in certain clinical settings (e.g., infectious disease wards, neonatal care). Non-contact methods using radar offer a promising alternative, but accurately separating the distinct respiratory and heartbeat signals, especially when multiple people are present, has been a significant challenge. Existing methods often struggle with accuracy, interference from subtle movements, and the complexity of distinguishing signals from multiple individuals simultaneously. This paper aims to bridge this gap by developing an accurate, automated, and robust method for separating these vital signs from UWB radar data, even in the presence of unconscious body movements.
Methods
The core of this work lies in the application of Empirical Wavelet Transform (EWT) for signal separation. Unlike traditional wavelet transforms that use predefined bases, EWT adaptively decomposes a signal based on its frequency content. Here's a breakdown of their approach:
- SFCW-UWB Radar: The system utilizes a Stepped-Frequency Continuous Wave (SFCW) Ultra-Wideband (UWB) radar to capture signals. SFCW-UWB radar offers a good balance of range resolution and system complexity, making it suitable for vital sign monitoring.
- Signal Acquisition and Preprocessing: The radar receives signals, which are then processed to obtain a range-time matrix. Moving Target Indication (MTI) is applied to suppress static clutter (like furniture) and isolate signals from moving targets (the individuals).
- Mixed Vital Sign Signal Extraction: For each detected target, the mixed respiratory and heartbeat signal is extracted from the radar data.
- Fundamental Respiration Frequency Estimation: An initial estimate of the fundamental respiration frequency is made by analyzing the high-pass filtered signal. This provides a starting point for EWT.
- Empirical Wavelet Transform (EWT): This is the key innovation.
- Adaptive Boundary Construction: The EWT requires defining frequency boundaries for decomposition. The paper uses prior knowledge of typical respiration and heartbeat frequencies to set initial boundaries. These boundaries are then adaptively refined based on the actual mixed signal's frequency characteristics.
- EWT Decomposition: The mixed vital sign signal is decomposed into several components using EWT. Each component corresponds to a specific frequency band.
- Signal Separation: Based on the estimated fundamental respiration frequency and known frequency ranges for respiration and heartbeat, the appropriate EWT components are identified and selected as the separated respiratory and heartbeat signals. This process is unsupervised and doesn't require external reference signals.
- Comparison with Existing Methods: The proposed EWT method is compared against band-pass filtering, Empirical Mode Decomposition (EMD), and standard Wavelet Transform (WT) methods.
Datasets
The study was conducted using data collected from three healthy male volunteers. The experiment involved simultaneously measuring their vital signs using the developed SFCW-UWB radar system and traditional contact physiological sensors (breathing belts and finger-clip pulse oximeters). The data collection involved ten continuous measurements for each volunteer, totaling 30 measurements. No pre-existing public datasets were used; the data was generated specifically for this research.
Key Insights and Ideas
- EWT's Adaptability is Key: The paper highlights that EWT's ability to adapt its frequency decomposition based on the input signal is crucial for accurately separating complex vital signs, especially when compared to fixed-basis methods like standard WT.
- Leveraging Prior Knowledge for Initialization: While EWT is adaptive, using prior knowledge about typical vital sign frequencies to initialize the boundary estimation significantly improves its effectiveness and reduces the need for extensive parameter tuning.
- Robustness to Movement: A critical finding is the method's resilience to unconscious body movements. This is a significant advantage over many existing non-contact sensing techniques that can be easily disrupted by motion artifacts.
- Multi-Target Capability: The method is designed and validated for simultaneously monitoring multiple individuals, which is a substantial step towards practical, real-world applications in shared living or clinical spaces.
- Unsupervised Operation: The fact that the method is unsupervised and doesn't require external reference signals simplifies its implementation and broadens its applicability.
Key Results
The proposed EWT-based radar system demonstrated high accuracy in extracting respiratory and heartbeat rates:
- Respiratory Rate (RR) Error: The largest error for RR was within ±0.3 bpm.
- Heartbeat Rate (HR) Error: The largest error for HR was within ±2 bpm.
- Comparison Performance: These errors are significantly smaller than those achieved by band-pass filtering, EMD, and WT methods. For instance, the EMD method showed a much larger bias and standard deviation for both RR and HR.
- Bland-Altman Analysis: Bland-Altman plots confirmed the strong agreement between the radar-extracted vital signs and those from contact sensors, with the proposed method showing the smallest mean difference (MD) and standard deviation (SD) for both RR and HR. The proposed method's results met clinical agreement limits (±2 bpm for RR, ±5 bpm for HR).
Limitations and Future Work
- Limited Subject Pool: The study was conducted with only three healthy male volunteers. The authors acknowledge this limitation and suggest that future work should expand the dataset to include a wider range of subjects (different ages, genders, health conditions) and more extensive measurements to further validate the method's generalizability.
- Unconscious Movement Definition: While the paper mentions robustness to "unconscious body movement," the exact nature and extent of these movements in the experiments could be further detailed.
- Generalizability to Other UWB Radars: The authors propose that the method could be applied to other types of UWB radars (e.g., impulse, FMCW) for multi-human target detection and CW radar for monitoring. Future work could explore this cross-compatibility.
- Clinical Validation: While the results meet clinical requirements, further large-scale clinical trials are needed to fully establish its efficacy in diverse healthcare settings.
Summary updated