An electrical engineering friend who did some hardware development work on these a few years back suggested that noise in the observations was a fairly significant challenge. Seems the devices use a lot of data smoothing to counteract this.