Contactless Heart Rate Estimation using a 61 GHz Continuous-Wave Radar

I. INTRODUCTION

According to estimations by the World Health Organization, more than 8.5 million people are diagnosed with Parkinson’s disease, making it the second most prevalent neurodegenerative disorder worldwide [1]. While the most commonly-known symptoms are motor symptoms, the disease is also characterized by sleep-related non-motor symptoms such as rapid- eye-movement sleep behavior disorder, insomnia, or daytime sleepiness. Therefore, reliable sleep monitoring is crucial for the diagnosis and treatment. As traditional sleep monitoring is obtrusive and hard to perform longitudinally, the contactless measurement of biosignals at home is a promising alternative for meaningful analysis of sleep [2].

Previous research has shown encouraging results in the extraction of relevant biosignals such as heart sounds or respi- ration waves from radar data [3]. However, there is insufficient research on assessing the robustness against different sleeping positions, as well as the combination of multiple radar sensors.

For that reason, this work aims to present a radar sens- ing system for nocturnal heart sound extraction using a bi- directional Long short-term memory network. In particular, we combine the information collected by four radar sensors placed horizontally at thorax height under the bed mattress. Furthermore, we compared the extracted heart rate with respect to different sleeping positions and upright sitting.

sleep study of 17 participants (age: 27.6 ± 8.4; 52.9% female), resulting in a total of 99.5 hours of sleep data. The dataset includes polysomnography recordings and simultaneous sig- nals captured by four 61 GHz continuous wave doppler radars. For each radar sensor, heart sounds were extracted from the raw signals using a pre-trained bi-directional LSTM model. On the model output, we performed a threshold-based peak detection to identify individual heart beats. To fuse the output of the four sensors, we selected the sensor with the highest number of detected heart beats for every 30-second window. Subsequently, the selected beats were utilized to calculate the mean heart rate, which was then compared with the heart rate extracted from the electrocardiogram data for each respective 30-second window.

III. RESULTS & DISCUSSION

Our findings underline the potential of contactless vital sign monitoring for unobtrusive sleep analysis. This allows for longitudinal and more realistic measurements of sleep patterns. Thus, it is a promising step towards transferring sleep labora- tories into a home monitoring environment.