A comparison between ceiling-mounted FMCW, IR-UWB and Wi-Fi radar for in-bedroom human activity monitoring and sleep interruption detection

This paper investigates the performance and environmental robustness of three prominent radio frequency technologies—frequency-modulated continuous wave (FMCW) radar, impulse radio ultra-wideband (IR-UWB), and Wi-Fi sensing—for contactless in-bedroom human activity monitoring and sleep interruption detection. Although RF sensing is increasingly used as a privacy-preserving alternative to cameras in healthcare, existing studies rarely compare different radio modalities under identical deployment conditions. To address this gap, the authors conducted a controlled study using synchronized recordings from 20 participants across six distinct room layouts, keeping the ceiling-mounted sensing hardware fixed. A uniform convolutional neural network architecture was applied to process all modality-specific representations, ensuring that observed performance differences reflect underlying signal characteristics rather than model design variations.

The measurement campaign utilized a ceiling-mounted hardware setup positioned above a bed in a residential test environment. The study evaluated three specific technologies: a 60-64 GHz Texas Instruments IWR6843AOP FMCW radar module, a Qorvo DW3000-based IR-UWB development kit operating on channel 5, and a ZedBoard-based openwifi platform acting as a monostatic Wi-Fi radar. Measurements were captured concurrently across 20 participants who performed natural activity sequences guided by text-to-speech instructions. The evaluation framework comprised two main tasks: a fine-grained 10-class human activity recognition (HAR) task and a coarse 4-class sleep monitoring task. Robustness was further evaluated through cross-subject and cross-layout cross-validation protocols.

The results reveal a fundamental trade-off between recognition performance and environmental robustness across the evaluated RF technologies. IR-UWB demonstrated superior cross-subject activity recognition, achieving an 89.0% macro F1 score due to its high temporal and spatial resolution. Conversely, FMCW generalized best to unseen room layouts, reaching an 83.8% macro F1 score because its multi-antenna configuration and Doppler resolution provide stable spatial mapping across spatial shifts. For the coarser sleep monitoring task, all three modalities exceeded 92% macro F1 in unseen environments, indicating that high-level sleep interruption detection is less sensitive to modality-specific resolution differences. The authors link these performance traits directly to physical signal properties such as range resolution, antenna diversity, and Doppler retention.