TL;DR
This paper presents a novel approach for accurately detecting human vital signs (respiration and heart rate) using a single Doppler radar sensor, even during significant and rapid body movements. By combining adaptive noise cancellation (ANC) with a new discrete cosine transform (N-DCT) method, the system effectively filters out disruptive motion noise, achieving high detection accuracy.
What Problem Does This Paper Address?
Non-contact vital sign monitoring using Doppler radar offers significant advantages for healthcare, particularly in intensive care and home settings. However, a major hurdle has been the interference caused by large-scale and fast random body movements (RBMs). These movements can displace the body by tens of centimeters, dwarfing the subtle movements associated with respiration (millimeters) and heartbeat (sub-millimeter). This overwhelming noise makes it extremely difficult for traditional Doppler radar systems to reliably extract vital sign information. Existing solutions often require multiple radar sensors or additional hardware like cameras, which can be cumbersome and less practical. This paper aims to overcome this limitation by enabling accurate vital sign detection with a single, readily available Doppler radar sensor, even under challenging RBM conditions.
Methods
The core of the proposed solution lies in a two-pronged approach:
- Adaptive Noise Cancellation (ANC) with Polynomial Fitting: To tackle the RBM noise, the system first models the entire received signal as a combination of vital signs and body motion. A polynomial fitting technique is employed to create a waveform that closely approximates the body movement component. This approximated movement signal then serves as the input for an ANC algorithm. The ANC iteratively adjusts its internal filter to cancel out the approximated body motion from the original signal, leaving behind the vital sign components.
- New-Type Discrete Cosine Transform (N-DCT): After the RBM is suppressed by the ANC, the remaining vital sign signal is analyzed using a novel spectral analysis technique called the "new-type discrete cosine transform" (N-DCT). The authors highlight that N-DCT offers improved spectral peak highlighting compared to traditional methods like the Fast Fourier Transform (FFT), leading to more accurate extraction of respiration rate (RR) and heart rate (HR).
The radar sensor itself is built using standard, general-purpose RF and communication instruments, avoiding the need for specialized components or custom hardware.
Datasets
The paper describes experiments conducted using a 10-GHz Doppler radar sensor. The data was collected from human subjects performing large-scale and fast body movements. The paper does not mention the use of pre-existing datasets for training or evaluation. Instead, it relies on numerical simulations and experimental data collected by the researchers.
Key Insights and Ideas
- Single-Sensor Power: The most significant insight is the demonstration that accurate vital sign detection during substantial RBM is achievable with a single, standard Doppler radar sensor, challenging the notion that multiple sensors or complex setups are always necessary.
- ANC as a Powerful Denoiser: The paper effectively showcases the power of ANC, when combined with a well-chosen approximation of the noise (here, polynomial fitting), to isolate weak physiological signals from overwhelming motion artifacts.
- N-DCT's Spectral Advantage: The N-DCT is presented as a superior spectral analysis tool for vital sign extraction compared to FFT, particularly in noisy environments, due to its ability to sharpen spectral peaks.
- Practical Implementation: The use of readily available RF instruments for building the radar sensor makes the proposed system more accessible and cost-effective for research and potential deployment.
Key Results
- Significant Error Reduction: The N-DCT method, when used with the ANC algorithm, dramatically reduced measurement errors compared to the traditional FFT method.
- For a 5-second time window, the average error for RR was reduced from 25.53% (FFT) to 2.14% (N-DCT), and for HR, from 3.84% to 0.87%.
- Even with a shorter 3-second time window, the average error for RR was reduced from 24.77% (FFT) to 4.86% (N-DCT), and for HR, from 7.71% to 1.25%.
- Effective RBM Cancellation: Experimental results demonstrated that the ANC algorithm successfully removed the majority of the body movement signals, allowing the vital sign signals to be extracted.
- Accurate Detection During Movement: The system was able to extract both RR and HR during simulated and actual large-scale and fast body movements, with the N-DCT method consistently outperforming FFT in terms of accuracy.
Limitations and Future Work
- Arm Movement: The current method is primarily effective for upper body movements. The authors acknowledge that arm movements introduce new Doppler phase modulations that make vital sign extraction more challenging.
- Constant Speed Assumption: The ANC approach assumes a constant speed of RBM within a short time window. While the paper proposes segmenting signals into shorter windows to address this, it notes that highly irregular movements with varying speeds remain a challenge.
- Future Directions: The authors suggest developing more flexible fitting methods to better approximate complex RBM curves and improve the ANC's performance in worst-case scenarios where body movement speed constantly changes. They also propose further investigation into segmenting body displacement signals to handle more dynamic movement patterns.
Summary updated