[time-nuts] Z3805A cooling requirements?
magnus at rubidium.dyndns.org
Sat Dec 22 13:22:21 UTC 2012
On 12/22/2012 04:55 AM, Volker Esper wrote:
> Although I am dog-tired it gives me no peace...
> I come to the following conclusion:
> - The long term Allan deviation gets worse, as long, as the effect of
> EFC compensating is in the range of tau
> - But: It gets back to its normal value after that
> - The short term deviation, however, increases slowly, but it doesn't
> settle. It's increasing more and more.
> So it is the short term stability, that is affected, rather than the
> long term stability.
> Am I right or wrong?
> Perhaps I'm to tired to decide.
You short term will rise with 1/tau in ADEV is due to white noise. This
noise may be due to the oscillator or (many times) your measurement rig.
This is expected from theory (see Allan Deviation on Wikipedia).
A free-running OCXO will have a D*tau/sqrt(2) rise due to linear drift.
If it is beeing steered it would captured by the control loop and
depending on it's properties there will be some marginal errors due to it.
A free-running OCXO will also experience temperature shifts and that
will create a ripple effect in the ADEV/MDEV. If it is being steered the
PLL loop would have issues fighting it which could leave traces in the
For systematic effects like these, the ADEV/MDEV/TDEV plot isn't the
ideal tool. phase/frequency/drift plots and FT of them might be more useful.
However, as you might have realized, a single event of temperature event
like this averages out on the large scale of things. ADEV and friends
isn't a good tool for transient properties, as it is a statistical tool
to establish noise levels of various types. For that purpose single
freak events of systematic effect needs to be averaged out or even taken
Doing your plots to show how they vary over time illustrates this in a
good way. The "event" is not part of the long-term noise properties.
More information about the time-nuts