[time-nuts] Question on Tom's ADEV plots

Tom Van Baak tvb at LeapSecond.com
Fri Oct 17 14:40:58 UTC 2008


> Yes, I am wondering what sort of reference would be ten times better than a
> good OCXO at tau=0.1 second and stay good to tau=10^4 seconds. I am looking
> at the very first plot of http://www.leapsecond.com/pages/fury/. Also, what
> do you mean by " to avoid having to subtract background noise from the ADEV
> plot"? Is there a standard recipe for this type of operation?
> Thanks again,
> Javier

One short answer is that what is better than a *good* OCXO is
a *better* OCXO. Many FTS 1000- or 1200- series oscillators,
old Sulzers, an HP106, etc. can have performance well down
into the -13's. BVA if you can afford, or borrow one. Once in
a while you'll find a 10811 that's in that range too. When you
do you hang onto it and it becomes your reference, until you
run across one even better, and so on. Or, you use a maser
which gets you in the -14's and -15's.

To get Allan deviation you need to take phase readings between
the UUT oscillator and the reference with something like a time
interval counter. You get your readings, do the math, make a plot
and think you're done. But here's the thing -- these numbers are
not absolute.

Every ADEV point is a measure of the *relative* stability among
the UUT and the reference, and the TIC itself. So not only can
any one of the three components limit the resolution of the plot
at a given tau, but all ADEV points are the rms sum of all three
components. If you know the actual stability for a given tau of
the TIC or the reference you can, with some level of confidence,
rms subtract that known value from the point before it is plotted.

As a simple example, if you know your TIC is not the limiting
factor and you measure the stability of the UUT and a similar
reference to be, say 7e-12, then it could be that both are really
5e-12 (since 5^2 + 5^2 ~= 7^2), which is sqrt(2) better.

Not sure this is a recipe, but you should have the idea by now.

/tvb





More information about the time-nuts mailing list