Robust Heartbeat Detection with Doppler Radar Based on Stochastic Gradient Approach

Heart-rate variability (HRV) is closely related with physical or mental conditions. It is rather necessary to develop remote monitoring technologies to reduce subjects' pressure, e.g., heart-rate (HR) monitoring to drivers. Recently, the contactless heartbeat detection with Doppler radar has drawn extensive attention, due to less burden on subjects. However, the received signals of Doppler radar are easily contaminated by respiration or body motion, resulting in performance degradation. In this paper, to realize robust heartbeat detection with Doppler radar, a stochastic gradient approach is first proposed to reconstruct heartbeat spectrum. By correcting the gradient of cost function constructed by recursion error, and utilizing the sparse characteristics of heartbeat spectrum, the reconstructed spectrum can be obtained with minimum deviation. Furthermore, the zero-attracting sign least mean square (ZA- SLMS) algorithm based on stochastic gradient descent (SGD) is proposed, to accomplish more stable sparse spectrum reconstruction (SSR), by quantizing the updating of recursion error. Finally, HR is estimated by spectral peak tracking consisting of peak selection and verification. Experimental results validate that the proposed method can significantly improve detection accuracy over the conventional methods, based on the measurements from 5 subjects during sitting still or typing with a laptop.