An Approach for Sleep Apnea Detection based on Radar Spectrogram Envelopes

This research aims to develop a contactless, radar-based sleep apnea detection method. A novel identification approach for this is proposed, based on the envelope of UWB radar spectrograms and machine learning. The envelope of the spectrogram is extracted by an image-based method, followed by signal smoothing via variational mode decomposition (VMD). The method is validated via simulations, and experimental data collected on 14 volunteers in controlled conditions, including supine, side and prone positions and the presence of a blanket. Initial results show that the proposed approach provides over 90% accuracy, precision and recall.