UWB Radar-based Heart Rate Monitoring: A Transfer Learning Approach

Abstract: Radar technology presents untapped potential for continuous, contactless, and passive heart rate monitoring via consumer electronics like mobile phones. However the variety of available radar systems and lack of standardization means that a large new paired dataset collection is required for each radar system. This study demonstrates transfer learning between frequency-modulated continuous wave (FMCW) and impulse-radio ultra-wideband (IR-UWB) radar systems, both increasingly integrated into consumer devices. FMCW radar utilizes a continuous chirp, while IR-UWB radar employs short pulses. Our mm-wave FMCW radar operated at 60 GHz with a 5.5 GHz bandwidth (2.7 cm resolution, 3 receiving antennas [Rx]), and our IR-UWB radar at 8 GHz with a 500 MHz bandwidth (30 cm resolution, 2 Rx). Using a novel 2D+1D ResNet architecture we achieved a mean absolute error (MAE) of 0.85 bpm and a mean absolute percentage error (MAPE) of 1.42% for heart rate monitoring with FMCW radar (N=119 participants, an average of 8 hours per participant). This model maintained performance (under 5 MAE/10% MAPE) across various body positions and heart rate ranges, with a 98.9% recall. We then fine-tuned a variant of this model, trained on single-antenna and single-range bin FMCW data, using a small (N=376, avg 6 minutes per participant) IR-UWB dataset. This transfer learning approach yielded a model with MAE 4.1 bpm and MAPE 6.3% (97.5% recall), a 25% MAE reduction over the IR-UWB baseline. This demonstration of transfer learning between radar systems for heart rate monitoring has the potential to accelerate its introduction into existing consumer devices.

• The paper demonstrates transfer learning's viability by transferring learned features between FMCW and IR-UWB radar systems to improve heart rate monitoring.
• It employs a novel 2D+1D ResNet architecture to capture spatial and temporal features, enhancing accuracy with reduced dataset requirements.
• The methodology achieves a 25% MAE improvement for IR-UWB data, reducing errors from 5.4 bpm to 4.1 bpm across varied conditions.

UWB Radar-based Heart Rate Monitoring: A Transfer Learning Approach

Introduction

This paper presents a study on radar technology's potential for continuous and contactless heart rate (HR) monitoring using consumer electronics. The study focuses on leveraging transfer learning between frequency-modulated continuous wave (FMCW) and impulse-radio ultra-wideband (IR-UWB) radar systems. The goal is to enable heart rate monitoring capabilities in devices like smartphones by demonstrating efficient transfer learning methods to overcome the requirement for large datasets for each radar system.

Radar Systems and Dataset Description

FMCW radar systems typically use a continuous chirp waveform operating at 60 GHz, providing a high range resolution with frequent use in consumer devices such as Google's Nest Hub. Conversely, IR-UWB radar systems, which transmit short pulses at 8 GHz, are increasingly integrated into mobile devices but face challenges such as lower range resolution.

The study utilized two datasets:

1. FMCW Dataset: Collected using a Google Nest Hub device during overnight sleep sessions. The dataset included 119 valid sessions with continuous ECG ground-truth HR monitoring.
2. IR-UWB Dataset: Collected from various sites using an NXP SR160 UWB radar device, including participants with and without cardiovascular conditions.

Methodology

The methodology involves several stages, including data preprocessing, presence detection, feature extraction, model architecture design, and transfer learning.

• Data Preprocessing: Data from radar signals were processed to reduce noise and enhance signal clarity. Techniques like FFT and CFAR were used for dimensional reduction.
• Model Architecture: A novel 2D~+~1D ResNet architecture was deployed to effectively learn spatio-temporal features from the radar data. Highlights include initial 2D convolutional layers to capture spatial relations, followed by 1D layers for temporal correlations.
• Transfer Learning Strategy: The model, initially trained on FMCW radar data, fine-tuned using IR-UWB data to leverage shared features across datasets, thus improving HR estimation accuracy even with smaller datasets.

  Figure 1: The high-level architecture of the system, showing the signal processing, presence detection, and feature extraction stages, followed by the ML model.

  

  Figure 2: Novel model architecture which combines a 2D and 1D ResNet network.

Results

The proposed method achieved substantial improvements in HR monitoring accuracy:

• FMCW Radar: Achieved an MAE of 0.85 bpm and a MAPE of 1.42%, halving previous SOTA errors with high recall (98.9%).
• Transfer Learning to IR-UWB: Fine-tuning reduced MAE from 5.4 bpm to 4.1 bpm, representing a 25% improvement over baseline models trained from scratch.

The study emphasized model robustness across participant demographics, radar positioning, and heart rate ranges.

Figure 3: a) Representative example of overnight session performance on the test set. b) An example of a session with significant MAE drop due to misestimation of user distance.

Discussion

The study effectively demonstrates leveraging transfer learning between disparate radar systems to enhance HR monitoring capabilities using smaller datasets. The proposed model's high accuracy and robust performance across varied conditions suggest potential for commercial deployment enabling wide-scale adoption in consumer electronics.

The presented solutions primarily target the challenges of HR monitoring through radar, such as differentiating between respiratory and cardiac motions and the necessity of signal processing innovations tailored to the radar modality.

Conclusion

The paper offers a comprehensive framework utilizing a 2D~+~1D ResNet architecture, enabling the transfer of learned features between FMCW and IR-UWB radar systems for accurate heart rate monitoring. Future research could explore enhancements in radar signal processing and transfer learning techniques across different radar technologies.

These advancements hold promise for contactless health monitoring, significantly benefiting telemedicine and general wellness applications. The outlined methodologies offer a scalable approach for integrating health monitoring seamlessly into ubiquitous consumer devices like smartphones.