Comparison of Radar Technologies for In-Bedroom Activity Monitoring

What changed

This paper presents a controlled comparison of three radio frequency (RF) sensing technologies: Frequency-Modulated Continuous Wave (FMCW) radar, Impulse Radio Ultra-Wideband (IR-UWB), and Wi-Fi sensing. The study specifically focuses on ceiling-mounted configurations, which offer practical deployment and cost advantages for healthcare applications but have been underexplored. The comparison was conducted under identical deployment conditions, using synchronized recordings from 20 participants across six room layouts. All technologies were evaluated using the same Convolutional Neural Network (CNN) for both a fine-grained 10-class human activity recognition (HAR) task and a coarse 4-class sleep monitoring task.

Why it matters for builders

For AI builders developing RF-based healthcare monitoring, this study offers a direct comparison of different radar modalities. It moves beyond isolated studies by standardizing hardware, datasets, and evaluation methodologies, providing a clearer understanding of performance trade-offs. This allows for more informed decisions when selecting the most suitable radar technology for specific applications, such as sleep tracking or general activity monitoring.

Practical impact

The findings reveal distinct performance characteristics for each technology. IR-UWB demonstrated the highest cross-subject activity recognition performance with an 89.0% macro F1 score. FMCW radar showed the best generalization to unseen room layouts, achieving an 83.8% macro F1 score. For sleep monitoring, all three technologies surpassed a 92% macro F1 score in unseen environments. The study attributes these differences to factors like range resolution, antenna diversity, Doppler resolution, and spatial information retention, indicating a fundamental trade-off between recognition accuracy and environmental robustness.

Caveats and source limits

The study focuses on in-bedroom monitoring and uses a specific CNN architecture for evaluation. The results are based on a dataset of 20 participants and six room layouts, and further validation across a wider range of environments and participant demographics may be beneficial. The paper is currently under review for publication in IEEE Access Journal.