Detecting sleep phase from motion sensors

There are are number of devices that claim to be able to detect which phase of sleep you’re in, and then trigger lucidity by playing some recorded audio.

I was thinking about making one, and as part of tbe literature search I came across this paper:

Actigraphy detection of REM sleep based on respiratory rate estimation, by Ken Kawamoto and Hiroyuki Kuriyama.

According to this article, detecting whether you’re asleep or not works fairly well, but detecting whether you’re in REM sleep is harder. What they did was to use the motion sensors to measure breathing in and out, and then use the change in breathing rate to detect the onset of REM sleep.

I thought this paper was interesting because it gives a reason why some of the devices that are on the market won’t work very well.

I have found success with using Heart rate monitors to judge sleep patterns. I recorded my HR for a week while I slept, along with a movement sensor (my fitness watch). I found that my HR consistently changes while I sleep. My HR dropped down as low as 30bpm during deep sleep, ( normal for me, big hearts run in my family, you should see my brothers waking bpm, it hangs around 32-34!) , and during light sleep, it would raise to about 60-70 bpm. I also found that whenever I had an exciting dream, my pulse would raise to the 80-90 range. I often considered programing something that would signal me when my HR got over a certain level, probebly 70bpm, but I never got around to it.

A couple years ago I did find some sleep tracking apps that monitored movement and would then play a sound during the right sleep phase to trigger a LD, but when I read the instructions, it said “Place phone on mattress during sleep”, I slept on the floor at the time, so I forgot about that idea. ( I now sleep in a hammock, so I guess I could try those apps again…)