[time-nuts] What is the correct Allan value?
magnus at rubidium.dyndns.org
Fri Oct 29 20:40:18 UTC 2010
On 10/29/2010 09:57 PM, mike cook wrote:
> looks like this might run...
> I would say that all the values are valid , but agree that the overall
> confidence in the performance couldn't be better than the worst case.
Which trace is worst? Check the plot again. Look at the green trace. At
0,1 - 20 s it is the worst, but over 100 s is it among the better. The
down-dipping form at higher taus is not what we expect, and the bulb
upwards is not expected either. They hint that you have either a
modulation or too little data, but you need more data to be able to
tell. Thus, get a longer trace. The confidence interval is poor there,
and the actual value can be spread anywhere for that trace for most of
the upper half so the other traces in neat groups is more credible...
So the worst one isn't a good pick. It's just the least overoptimistic,
which isn't the same thing as near the real value.
> I have noticed that I never get the same ADEV values on successive runs
> but put it down to an undisciplined thermal environment. I suppose that
> if you were running an experiment you would ideally measure the ADEV
> over the periods of interest and use those values as opposed to relying
> on some longer baseline.
The thermal environment is certainly a problem. A TCXO I measured
recently had nice properties... unless I changed anything in the
neighborhood of the boxed testboard. Laying a hand on the side of the
cardboardbox made the frequency jump quickly. Even simple OCXOs isn't
that sensitive... so yes. You can get quite varied results... with the
stable environment noise as the floor level.
More information about the time-nuts