Frequency Matters: Comparative Analysis of Low-Power FMCW Radars for Vital Sign Monitoring.

Vital sign monitoring is a critical step in health assessment in clinical settings. A rising interest in contactless options is pushing for innovative solutions for Heart Rate (HR) and Respiration Rate (RR) estimation. Low-power millimetrewave radars are emerging as a solution thanks to their invariance to lightning conditions, subject phenotype, and privacy guarantees. On the side, the robustness of radar-based systems is still a concern due to incomplete characterization of the systems, evaluation of different frequencies, and limited evaluation on human subjects. Moreover, low-power options have not been rigorously explored, despite their potential in ubiquitous deployment. This paper evaluates three low-power Frequency Modulated Continuous Wave (FMCW) radars with frequencies of 24 GHz, 60 GHz, and 120 GHz, investigating their performance and the influence of the carrier frequency in vital sign estimation. An initial characterization of the displacement noise is conducted using a phantom device. Various techniques to enhance the Signal to Noise Ratio (SNR) of the radar signal and the extraction of the chest displacement signal are combined. Finally, a lightweight and accurate algorithm based on the second-order derivative of the displacement is proposed, developed, and evaluated to assess the HR and RR. The evaluation is conducted for all three radar systems on a dataset with 24 subjects. We demonstrate with the experimental evaluation that the three systems accurately estimate the RR with a Mean Absolute Error (MAE) of less than 2 brpm (±3.05). The 60 GHz and 120GHz system estimate the HR accurately with a MAE of 1.8±3.1 bpm and 3.2±5.3 bpm respectively, while the 24 GHz system is less effective with a MAE of 9.0 bpm, mainly due to its high noise profile. This evaluation demonstrates the feasibility of HR and RR with low-power FMCW radars, underlying the importance of the operating frequency, and the necessity of an appropriate characterization when designing the algorithms.