DailyBeat: Reliable Cardiac Sensing Under Self-Induced Dynamic Interference Using mmWave Radar

Abstract

Reliable cardiac monitoring in daily work environments could support applications such as stress assessment and mental workload tracking. Radar sensing provides a promising contactless solution by capturing subtle chest motion without requiring wearable devices. However, many existing methods assume quasi-static conditions and experience substantial performance degradation in office environments because of self-induced interference, including body motion and the often-overlooked effect of irregular respiration. To address these challenges, we analyze the temporal structure of cardiac mechanical activity and identify two key properties: short-duration impulsiveness and short-term quasi-periodicity. Guided by these properties, we propose a signal-processing paradigm for reliable cardiac sensing under dynamic interference. Specifically, we design a bidirectional wavelet transform to extract pulse-like cardiac events from complex radar signals, a periodicity-guided multi-scale fusion strategy to retain rhythmically consistent components, and an adaptive spatial selection method to identify reliable chest reflections. We implement this paradigm as a real-time prototype system, DailyBeat, using a commercial mmWave radar. Experiments across multiple participants and office activities demonstrate that DailyBeat recovers detailed cardiac mechanical waveforms and accurately estimates heart rate, inter-beat intervals, and a QT-related mechanical surrogate, enabling fine-grained and unobtrusive cardiac monitoring in daily office environments.