[time-nuts] Allen deviation? Yes indeed that subject again

Magnus Danielson magnus at rubidium.dyndns.org
Sat Jan 9 00:11:56 EST 2016


Paul,

On 01/09/2016 03:05 AM, paul swed wrote:
> I have been looking at the subject of Allen deviation and this has been
> discussed numbers of time on Time-nuts.
> But my question is this.
>
> If an oscillator is stable in frequency but shifts phase 90 degrees and
> then comes back in a short time. From my reading I don't think that will
> show up in a typical Allen deviation plot that runs 1000s of readings at 1
> second intervals. Typical HP5370 test setup. I think this capture approach
> will miss the issue.
>
> How often? Correct phase .5-3 hours, shift 5-10 minutes it seems random
> actually and the duration varies.
> Am I looking at this correctly?

No.

OK, I might be a bit more specific. :)

We assume that it does it's phase-shift dance in perfect symmetry, then 
it is not as obvious that it will be observable, as if it didn't get 
exactly back we would naturally be able to observe that difference.

The condition for you not to observe it is really that the occurrence of 
these evens is synchronous to your measuring rate, and that these evens 
occur phase-wise in-between the two phase samples (as then we don't have 
to assume much more about the signal). For this case, it is obvious that 
you will always miss it. However, here we have the synchronization 
condition, which is obscure.

If we don't have the sampling rate and occurrence being synchronous,
it will be visible... but hard to notice. The power averaging would 
shift only so slightly that it would be hard to detect and it would 
disappear in the noise. If random as mechanism, it would only appear as 
the increase of the noise level. This is the case where ADEV isn't is 
necessarily your preferred tool, and is not intended to be your tool. It 
*might* appear as random-walk frequency noise, but it takes a little 
more analysis to conclude that.

Story-time: At NIST they saw how one of their cesium standards started 
to show unexpected random-walk frequency noise. This is exceptional as 
there is no real random-walk frequency noise source in clocks. They 
discovered this because they actually looked at their data. Turned out 
that the cleaning lady had to move the standard over the floor whenever 
she came in to clean up, and then she moved it back. The vibration from 
dragging it across the floor caused the modulation. Some re-arrangement 
in the lab she didn't have to move things around and the random-walk 
noise got back to normal.

So, look at the random walk noise, but do look at the phase-plot 
instead, especially the linear or quadratic fit residue plot of the 
phase. The normal frequency offset and even slow drift might obscure 
these deviations from being visible in the plot, only due to dynamic 
range. Remove the systematic shifts and you can see the fine-grain 
details. For TimeLab, press r for viewing the linear residue while 
viewing phase or frequency plot.

So, this is why I say you are not looking at it the right way.

Cheers,
Magnus


More information about the time-nuts mailing list