Sleep is essential for human health. Bad sleep and sleep disorders have been increasingly prevalent and are gradually becoming a social problem that cannot be ignored. The current gold standard in sleep monitoring is polysomnography (PSG) allowing nearly complete approach. Unfortunately, this wealth of information is obtained at the cost of invasive system, only usable in hospital environment under the control of sleep experts. Therefore, the development of a wireless body network for long-term home sleep monitoring is a good way to achieve this in a less-intrusive, portable and autonomous way. In this paper, an overall architecture from the sensors to the user's display is presented with a focus on the main functions and hardware.

The hardware architecture is composed of simple miniaturized wearable devices. Then, we introduce the chosen indicators for sleep monitoring and the algorithms developed for sleep stages classification. Finally we show the evaluation of our approach compared to the PSG. We illustrate the sleep stage classification during one night in the sleep unit of Toulouse University Hospital and highlight correlation between body temperature on extremities and Periodic Limb Movement during Sleep.

Based on the confusion matrix analysis, the results show that the T1 method appears to be effective for the detection of awake and deep sleep in particular. For PLMS detection, we define the detection rules based on the foot movement data. The results show that the total number of PLMS and the number of PLMS distributed in each sleep stage detected by our foot module are both very close to the PSG. Furthermore, we have found correlations between body temperature and hypnogram and between body temperature on extremities and PLMS.

A wearable sensor system could be an alternative to PSG for long-term monitoring. Validation of the two proposed threshold-based algorithmic methods for sleep stage classification compared to the PSG gold standard shows good agreement, while the k-means based approach shows poor agreement with PSG. Furthermore, this method could be a good candidate for predicting periodic leg movements in sleep.