Abstract
Automatically determining when a person falls asleep from easily available vital signals is important, not just for medical applications but also for practical ones, such as traffic safety or smart homes. Heart dynamics and respiration cycle couple differently during sleep and awake. Specifically, respiratory modulation of heart rhythm or respiratory sinus arrhythmia (RSA) is more prominent during sleep, as both sleep and RSA are connected to strong vagal activity. The onset of sleep can be recognized or even predicted as the increase of cardio-respiratory coupling. Here, we employ this empirical fact to design a method for detecting the change of consciousness status (sleep/awake) based only on heart rate variability (HRV) data. Our method relies on quantifying the (self)similarity among shapelets - short chunks of HRV time series - whose "shapes" are related to the respiration cycle. To test our method, we examine the HRV data of 75 healthy individuals recorded with microsecond precision. We find distinctive patterns stable across age and sex, that are not only indicative of sleep and awake, but allow to pinpoint the change from awake to sleep almost immediately. More systematic analysis along these lines could lead to a reliable prediction of sleep.