[time-nuts] Z3801A performance plots
Tom Van Baak
tvb at LeapSecond.com
Fri Jan 25 20:06:51 EST 2008
>> Here I put the two runs in color on the same plot:
> That's an interesting graph.
> I find it strange that the free-running system is better than the locked
> system in the range from 50 to 5000 seconds.
All GPSDO, by definition, will have a range of tau where the
free-running LO is better than the locked system. That is the
compromise you accept for a hybrid system. And even when
perfectly tuned, any GPSDO is worse than sqrt(2) of GPS and
LO performance at or near the cross-over point.
If you use a GPSDO as your house reference a case could
be made, depending on your needs, to let the GPSDO free-run
during the day, and then catch-up overnight.
You pay a medium performance penalty to track GPS short-term.
Of course, it all depends on the quality of the LO; but it's not
uncommon for a good GPSDO to use a loop filter on the order
of many hundreds or a few thousand seconds.
So that's why those free vs. locked plots look they way they do.
Note also for long-term, high -accuracy UTC measurements
(against Cs or maser) I don't use a GPSDO. Instead I use
a plain, non-disciplined OEM GPS board or something like a
CNS II (http://www.cnssys.com/cnsclock/CNSClockII.html).
High-end, short-term, or low phase noise, you always use
quartz; long-term you always used GPS. The cool thing about
a GPSDO is that you get (almost) the best of both in one box.
> Does that mean the filter needs a longer time constant? Or is there
> something fundamental about systems like this that makes graphs with that
> sort of pattern?
Yes, in that example, with that particular S/N of Z3801A, in my
particular hands-off, temperature controlled lab, and my well-placed
GPS antenna, a longer time constant would have been more ideal.
But that's a lot of if's. When HP ships a product they have to make
sure it works to spec; into the wild cell-tower near Fairbanks or in
my home lab.
In addition, some surplus GPSDO randomly have better LO than
others so the disparity between free-run and locked will appear
to be greater.
> How much did they understand about things like this graph when the Z3801A was
Based on conversations with some of the SmartClock engineers
years ago, and from reading the published papers, I would judge
it was all perfectly calculated, in the usual HP way.
More information about the time-nuts