Millimeter-Wave Radar Non-Contact Heart Rate Monitoring: A Cross-Platform Study Under Steady and Unsteady Conditions

Abstract

As the demand for isolation care increases, clinical practice requires non-contact, low-interference, long-term vital sign monitoring. Remote photoplethysmography (rPPG), infrared thermography (IRT), and pyroelectric infrared sensors (PIR) have been used for vital sign measurement but are susceptible to occlusion and motion interference. In contrast, millimeter-wave (mmWave) radar is not limited by lighting and can detect chest micro-displacement through thin clothing, enabling non-contact heart rate monitoring. However, studies have examined single radar platforms under steady conditions, with limited cross-platform and post-exercise evaluation. Differences in carrier frequency and antenna configuration may affect phase noise, SNR, range resolution, affecting heart rate estimation. Under identical conditions and processing, IWR6843AOPEVM and AWR2944EVM are evaluated under steady and post-exercise unsteady conditions using electrocardiography (ECG) as reference. Because hardware and acquisition architectures differ, results reflect system-level rather than single-hardware effects. Measurements were performed with seated subjects at a 40 cm radar-to-chest distance, with synchronous mmWave and ECG acquisition. Both platforms extract echo time series from Range-Azimuth heatmaps using background estimation and range masking. Adaptive Band-Limited Filtering (ABP), Empirical Mode Decomposition (EMD), Robust Outlier Suppression (ROS), and Local Energy Normalization (LEN) are evaluated, with ROS $\to $ EMD used. MAEs were 6.89 and 21.22 bpm for IWR6843AOPEVM and 2.16 and 4.19 bpm for AWR2944EVM under steady and unsteady conditions, respectively. Absolute Percentage Error (APE), Bland-Altman, and Cumulative Distribution Function (CDF) analyses showed a more concentrated AWR2944EVM error distribution. Overall, under controlled conditions and a fixed workflow, AWR2944EVM provides more stable heart rate estimation and may inform subsequent non-contact monitoring.