[time-nuts] OCXO settling
magnus at rubidium.dyndns.org
Wed Sep 14 21:48:07 UTC 2011
On 14/09/11 23:34, Paul A. Cianciolo wrote:
> Thank you.
>>> When an OCXO is unpowered for some time, it undergoes a further ageing
> process when it starts up next. This is specified in some data sheets, and
> can typically require an hour of burn-in for each day left unpowered before
> specifications are met.
> This is one thing I was curious about. I am asking these questions because
> I am working with a Rakon/C-MaC double oven OCXO.
> As I make changes to the power supply, cabinet, EFC power supply,
> surrounding this OCXO, I wondered how long to let the unit run before I run
> a standard test on it using TimeLab.
> Maybe 2 days would be a good compromise?
I let my OCXOs at work simmer for a couple of days before I really trust
the data... but using a direct ADEV is mearly an approximation as the
upper end will be drift limited. I use HDEV to pre-view what a drift
compensated curve would be, because Hadamard Deviation will remove first
degree drift, but true OCXO drift isn't linear, so using a
curve-matching gives better info.
It is actually recommended to run 1h test-runs back to back and see how
the "high end" reveals itself more and more as the drift falls back
while shorter taus remains fairly unchanged in the process.
TimeLab is good in that it will allow you to flip between ADEV and HDEV
while running. Looking at the high-end curve flipping and flapping
around while gently being pushed up by more samples is also educating.
More information about the time-nuts