Contactless FMCW Radar-Based Health Monitoring Using Continuous Wavelet Transform and Machine Learning

The rapid growth of the world's elderly population coupled with health issues has led to a high demand for robust health monitoring solutions. Frequency-modulated continuous waveform (FMCW) radars have attracted attention due to their capabilities for contactless monitoring of relevant health parameters.Nonetheless, to date, there are not many methods for extracting respiratory and electrocardiogram (ECG) signals from single-input-single-output (SISO) FMCW radars. This paper proposes a method based on continuous wavelet transform (CWT) and machine learning (ML) to extract respiratory and ECG signals from a SISO-FMCW radar located under a hospital bed. Respiratory and ECG signals are separated by using the wavelet thresholding method. Once both signals are clearly identified, supervised models are trained to provide fine grained ECG and respiratory information. We achieve an average error of less than 2 beats per minute (BPM) in heart rate monitoring when combining biorthogonal CWT with long short-term memory (LSTM) networks.