[time-nuts] GNSS Disciplined Clock
Tom Van Baak
tvb at LeapSecond.com
Thu May 25 14:33:43 EDT 2017
> The long term stability of GPS is very good. Some one here will point
> out exactly how one measures it. But roughly when speaking of
> accuracy you always need to specify a time interval. For example
> if the 1PPS is "off" by 15ns that is not bad and yes there are much
> better systems if you need to measure time intervals on the order of
> one second. But if the signal is "off" by 15 ns over 100,000 seconds
> that is well, 100,000 time better.
It's not 100,000 times better; it's not better at all.
If you are looking for *timing accuracy* then 15 ns is 15 ns. Doesn't matter if it right now, an hour from now, or tomorrow. It's an error, plain and simple. It doesn't get significantly better or worse over time. It is often quoted as an rms statistic on GPS receiver 1PPS specs. You can measure over a few minutes, or a few hours, or a few days -- you'll get approximately the same rms timing error.
What you're probably thinking of is *long-term average frequency accuracy* -- and then, yes, a bounded error like 15 ns rms looks less and less like a problem as you average longer and longer. But this does not mean the "GPSDO is getting better over time". All it means is the parameter you chose to measure (average frequency) happens to have elapsed time in the denominator, so of course the number gets lower as elapsed time goes on. But nothing tangible is getting "better" over time.
More information about the time-nuts