Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Interference Motion Removal for Doppler Radar Vital Sign Detection Using Variational Encoder-Decoder Neural Network

Mikolaj Czerkawski, Christos Ilioudis, Carmine Clemente, Craig Michie, Ivan Andonovic, Christos Tachtatzis

TL;DR

This work tackles the challenge of extracting radar-based vital signs when respiration and heart-beat Doppler components are contaminated by random body motion. It introduces a variational encoder–decoder CNN that operates on single-channel complex STFTs to learn a probabilistic reconstruction of vital-sign content from mixtures, using a $128$-dimensional latent space and a loss combining reconstruction with a small KL term on semi-experimental data. The model is trained on a dataset built from real vital-sign radar signals (processed to $f_s=100$ Hz and $128\times128$ STFTs) with simulated interference from a walking model, enabling controlled SIR variations. Results show robust respiration-rate extraction despite interference and noise, demonstrating practical potential for non-contact monitoring with a single radar receiver and enabling extension to broader interference scenarios.

Abstract

The treatment of interfering motion contributions remains one of the key challenges in the domain of radar-based vital sign monitoring. Removal of the interference to extract the vital sign contributions is demanding due to overlapping Doppler bands, the complex structure of the interference motions and significant variations in the power levels of their contributions. A novel approach to the removal of interference through the use of a probabilistic deep learning model is presented. Results show that a convolutional encoder-decoder neural network with a variational objective is capable of learning a meaningful representation space of vital sign Doppler-time distribution facilitating their extraction from a mixture signal. The approach is tested on semi-experimental data containing real vital sign signatures and simulated returns from interfering body motions. The application of the proposed network enhances the extraction of the micro-Doppler frequency corresponding to the respiration rate is demonstrated.

Interference Motion Removal for Doppler Radar Vital Sign Detection Using Variational Encoder-Decoder Neural Network

TL;DR

This work tackles the challenge of extracting radar-based vital signs when respiration and heart-beat Doppler components are contaminated by random body motion. It introduces a variational encoder–decoder CNN that operates on single-channel complex STFTs to learn a probabilistic reconstruction of vital-sign content from mixtures, using a -dimensional latent space and a loss combining reconstruction with a small KL term on semi-experimental data. The model is trained on a dataset built from real vital-sign radar signals (processed to Hz and STFTs) with simulated interference from a walking model, enabling controlled SIR variations. Results show robust respiration-rate extraction despite interference and noise, demonstrating practical potential for non-contact monitoring with a single radar receiver and enabling extension to broader interference scenarios.

Abstract

The treatment of interfering motion contributions remains one of the key challenges in the domain of radar-based vital sign monitoring. Removal of the interference to extract the vital sign contributions is demanding due to overlapping Doppler bands, the complex structure of the interference motions and significant variations in the power levels of their contributions. A novel approach to the removal of interference through the use of a probabilistic deep learning model is presented. Results show that a convolutional encoder-decoder neural network with a variational objective is capable of learning a meaningful representation space of vital sign Doppler-time distribution facilitating their extraction from a mixture signal. The approach is tested on semi-experimental data containing real vital sign signatures and simulated returns from interfering body motions. The application of the proposed network enhances the extraction of the micro-Doppler frequency corresponding to the respiration rate is demonstrated.
Paper Structure (12 sections, 6 equations, 5 figures, 2 tables)

This paper contains 12 sections, 6 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Proposed Solution
  4. Vital Signs Dataset
  5. Human Walking Simulator
  6. Vital Signs Combined with Interference Motion
  7. Architecture
  8. Training
  9. Results
  10. Network Operation
  11. Respiration Rate Extraction
  12. Conclusion

Figures (5)

  • Figure 1: Sample synthesis pipeline
  • Figure 2: Architecture diagram of the utilised network
  • Figure 3: (a) Unseen sample containing vital signs mixed with motion interference (b) Corresponding network output. (c) The ground truth reference. The brightness corresponds to the magnitude of stft.
  • Figure 4: Reconstruction loss influence comparison
  • Figure 5: Bin error influence comparison. (a) Clean vital signs sample. (b) Mixture sample. (c) Mixture sample processed by the network. Logarithmic scaling is applied to the displayed error values.