Human Vital Sign Detection Algorithm Based on Energy Focusing and Improved Variational Mode Decomposition

0
0
0

Human Vital Sign Detection Algorithm Based on Energy Focusing and Improved Variational Mode Decomposition

Introduction

Non-contact monitoring of human vital signs, such as respiratory rate (RR) and heart rate (HR), has gained significant attention due to its applications in healthcare, elderly care, and fitness tracking. Among various non-contact sensing technologies, frequency-modulated continuous wave (FMCW) radar stands out for its high sensitivity, long detection range, and robustness against environmental interference. However, a major challenge in FMCW radar-based vital sign detection is the presence of large-scale random body movements (RBM), which can severely degrade measurement accuracy.

Traditional methods require subjects to remain still to ensure stable distance window selection for phase extraction. However, in real-world scenarios, such as home environments, people frequently exhibit spontaneous movements, including head tilting, arm swinging, or torso shifting. These movements introduce significant noise and motion artifacts, making it difficult to accurately extract subtle chest vibrations caused by respiration and heartbeat.

This paper presents a novel algorithm that combines energy focusing and polynomial fitting for optimal distance window selection, along with an improved variational mode decomposition (VMD) technique to eliminate RBM interference. The proposed method achieves high-precision vital sign estimation even when the subject exhibits large-scale random motions with displacements up to 20 cm.

Challenges in FMCW Radar-Based Vital Sign Detection

Impact of Random Body Movements

In a static scenario, vital signs can be modeled as sinusoidal signals representing chest displacement due to respiration and heartbeat. However, when random body movements are introduced, the signal becomes highly nonlinear. The amplitude of RBM is often tens to hundreds of times greater than the chest vibrations caused by respiration and heartbeat, making it difficult to isolate the weak vital sign components.

Additionally, RBM introduces motion artifacts in the radar’s range-time map, leading to incorrect distance window selection. Traditional peak power-based window selection fails because large movements cause abrupt jumps between distance bins, disrupting phase continuity. Moreover, the frequency components of RBM often overlap with those of respiration and heartbeat, complicating spectral separation.

Limitations of Existing Methods

Previous approaches, such as bandpass filtering and wavelet transforms, struggle to separate vital signs from RBM due to overlapping frequency bands and non-stationary signal characteristics. Variational mode decomposition (VMD) has been used to decompose signals into intrinsic mode functions (IMFs), but conventional VMD suffers from mode mixing and endpoint effects, reducing its effectiveness in RBM suppression.

Some studies have attempted to mitigate RBM by using multiple radar antennas for target tracking or adaptive parameter tuning in VMD. However, these methods either fail to handle large-scale movements or require manual intervention, limiting their practicality in real-world applications.

Proposed Methodology

The proposed algorithm consists of three main stages:

  1. Optimal Distance Window Selection using energy focusing and polynomial fitting.
  2. Phase Extraction and Unwrapping to obtain displacement-related phase information.
  3. Improved VMD-Based RBM Suppression using autocorrelation power spectral density and weighted permutation entropy.

Energy Focusing and Polynomial Fitting for Distance Window Selection

Instead of relying solely on the peak power method, the algorithm employs energy focusing to enhance robustness against motion artifacts. The chest’s thickness causes radar reflections to spread across multiple adjacent distance bins. By aggregating energy over a window of bins corresponding to the chest’s physical dimensions, the algorithm reduces noise interference.

After initial window selection, a polynomial fitting step smooths abrupt transitions caused by RBM. This ensures that the distance window sequence follows natural, continuous motion patterns rather than erratic jumps. The resulting smoothed sequence improves phase extraction accuracy.

Phase Extraction and Unwrapping

The phase signal is extracted from the selected distance bins using arctangent demodulation. Since the phase is wrapped within the range of -π to π, a phase unwrapping step is applied to reconstruct the true displacement signal. The unwrapped phase is linearly proportional to chest displacement, enabling further processing for vital sign extraction.

Improved VMD for RBM Suppression

The core innovation lies in the enhanced VMD technique, which incorporates autocorrelation power spectral density and weighted permutation entropy for adaptive signal decomposition.

  1. First-Stage VMD: The unwrapped phase signal is decomposed into IMFs. Autocorrelation power spectral density analysis identifies IMFs dominated by RBM (low-frequency, high-power components), which are discarded.
  2. Second-Stage VMD: The remaining signal is decomposed again with a higher number of IMFs to finely separate respiration and heartbeat components.
  3. Weighted Permutation Entropy: This metric evaluates the randomness of each IMF. Since vital signs are periodic, their IMFs exhibit lower entropy compared to RBM-dominated components. The algorithm automatically selects the most relevant IMFs for reconstruction.

The final reconstructed signal provides clean respiration and heartbeat waveforms, enabling accurate frequency estimation.

Experimental Validation

Setup and Parameters

Experiments were conducted using a 77 GHz FMCW radar (IWR1843BOOST) with a distance resolution of 4.5 cm. Eight subjects participated, simulating real-life scenarios with controlled RBM (e.g., torso swaying up to 20 cm). Reference measurements were obtained using finger-clip pulse oximeters and smartwatches.

Results

  1. Distance Window Tracking: The proposed method outperformed traditional peak power selection, reducing erroneous window jumps by 85%.

  2. Vital Sign Estimation:
    • Respiration Rate: Average accuracy of 97.7%, with a maximum error of 2.4%.

    • Heart Rate: Average accuracy of 96.9%, with a maximum error of 3.3%.

  3. Comparison with Existing Methods:
    • The algorithm improved RR accuracy by 5.2% and HR accuracy by 2.7% over RETF-TVF-EMD.

    • It outperformed IAP-VMD by 14.3% (RR) and 7.9% (HR).

Conclusion

The proposed algorithm addresses the critical challenge of large-scale random body movements in FMCW radar-based vital sign detection. By integrating energy focusing, polynomial fitting, and an improved VMD technique, it achieves robust performance in real-world scenarios. Future work will focus on handling limb movements and extending the method to multi-subject environments.

doi.org/10.19734/j.issn.1001-3695.2024.08.0315

Was this helpful?

0 / 0