[time-nuts] GPSDO Sigma Tau at large observation times
Tom Van Baak
tvb at LeapSecond.com
Thu Oct 6 16:05:50 EDT 2016
>> See also the examples here: http://leapsecond.com/pages/gpsdo/
> Hi Tom, quick question: I've seen these plots before and they are very
> useful to know what to aim at for GPSDO performance. Am I right in
> thinking these were measured against a master - the page says " a very
> stable 10 MHz reference. " without more details (unless I missed it)
Yes, I think so. I can check, but if I say something like "a very stable reference" it usually means I picked something sufficiently accurate for the job. As a rule-of-thumb you want both your time/frequency comparator instrument and your time/frequency reference to be 10x better than the device you are testing. So to measure a typical GPSDO you want at least a rubidium, if not cesium or H-maser. But the key point is to always be aware of the limitations of your own test setup as they may affect the look of the ADEV plot. Using multiple comparators and references with N-corner hat techniques is often a useful trick.
If you want to get picky there are other factors hidden behind every ADEV plot. Often unstated is the bandwidth of the measurement and also the type of Allan statistic being used (ADEV, MDEV, HDEV, etc.). Temperature and airflow too: an ADEV plot may look different if it was created from data in a +/-0.1C lab vs. a +/-10 C lab. And with GPSDO, an ADEV plot can also be affected by the quality of your antenna, the quality of your sky-view, and especially, your latitude. So all these factors go into what the ADEV plot looks like.
BTW, one of my greatest surprises was the effect of still air on a plain TCXO. We call it "still", but it is nothing but stationary! The plots and two photos shows the massive difference a single sheet of TP makes: http://leapsecond.com/pages/LTE-Lite/
But back to your question. There's a good chance I used a TSC 5120 analyzer and a CH1-76 passive H-maser for all those GPSDO plots. This means all the data beyond about tau 1 s is fully trustable. The data nearer tau 0.1 s is likely distorted by a known short-term noise issue in that particular surplus maser.
And this brings up another point. Both Allan deviation (ADEV) and phase noise (PN) plots tend to cover a wide dynamic range. So in general no one reference is perfect both short-term and long-term, both close-in and far-out. For example, the typical cesium standard is good for long-term but not so good for short-term (a free-running high-quality quartz can be better). And my best ULN (ultra low noise) quartz standards have superb phase noise, but lousy ADEV. A good quartz will blow away GPS short- and mid-term; but GPS wins in the long-term. For short-term a free-running GPSDO is always better than a locked GPSDO. And so on.
All this is just a reminder that you constantly have to double check your assumptions. An ADEV or PN plot is merely the sum of all the noise in the "system": the DUT, the instrument, the REF noise, but also the room, weather, cables, wife/kids/pets, power lines, FM stations, etc. And that sum changes as you slide from the left to the right of the plots.
More information about the time-nuts