Estimating noise for clock-synchronizing Kalman filters

We implemented methods to determine the main uncertainty parameters for a Kalman-based clock servo whilst running that servo. This allowed us to provide a clock servo that can provide strong synchronization performance and accurate synchronization error estimates without manual tuning for any specific hardware. Evaluating the servo on real hardware we find that it provides synchronization accurate to about twice the resolution of the clock. Furthermore, the parameter estimation seemed to introduce only a modest amount of additional startup time for the filter at around 10 minutes.