Advantages of MIMO Radar for Contactless Pulse Waveform Monitoring using End-to-End Learning

Radar-based methods for contactless pulse monitoring typically use heart rate estimation accuracy as the primary performance evaluation metric. Recently, more in-depth metrics such as heart rate variability (HRV) have been of interest, requiring precise pulse waveform reconstructions for reliable estimations. Traditional phase variation-based vital signs signal processing techniques have proven to be susceptible to sources of noise, impacting the shape of the extracted pulse waveform. In this work, the advantages of using raw multi-channel radar data and deep learning in an end-to-end manner for precise pulse waveform reconstruction are explored. Preliminary results show electrocardiogram (ECG) waveforms can be reconstructed from raw in-phase and quadrature 60 GHz multiple-input-multiple-output (MIMO) frequency modulated continuous wave (FMCW) radar data using an end-to-end neural network. Additionally, an ablation study is performed, revealing that waveform reconstruction similarity and estimation accuracy of several HRV metrics generally improve as the number of utilized MIMO channels increases.