[time-nuts] Looking for datasheet for Oscilloquartz 8602

Magnus Danielson magnus at rubidium.dyndns.org
Sat Jun 1 19:52:18 EDT 2013


On 06/02/2013 12:58 AM, Jim Lux wrote:
> On 6/1/13 1:46 PM, Bob Camp wrote:
>> Hi
>>
>>> For ADEV, a lot of oscillators have a sort of "floor" where the
>>> ADEV is relatively constant, say from tau in the range10-1000
>>> seconds, and then it rises up (from thermal effects and such), so
>>> the shorthand is that the number quoted is that "floor value"
>>
>> You see a lot of different ADEV plots. Some would suggest flat from
>> 0.1 seconds out. The real world is rarely that simple ….
>>
>
>
> Yes.. but still, a floor of sorts.. high at the low end, high at the
> high end (for tau), and flattish in the middle.
>
>>>
>>>
>>>
>>> Certainly for "OC" applications this might be true. Although, a
>>> sort of trend is that the TCXO resonator has to have a lower Q, so
>>> the temperature compensating components can "pull" it to the right
>>> frequency over temperature, so the phase noise of a TCXO isn't as
>>> good as that of an OCXO, which can have a higher Q.
>>>
>>> A lot of times, though, an OCXO is chosen because a TCXO doesn't
>>> have frequency stability needed over environmental changes. I don't
>>> think ADEV is really the right measure when you're looking at aging
>>> or temperature effects.
>>
>> Well, I've certainly seen TCXO's spec'd and 100% tested for ADEV in
>> the 50,000 pc / year quantities …
>
>
> Sure, but is ADEV *of the oscillator at constant temp* really relevant
> when your application isn't at constant temp. I guess it is, because
> your system ADEV can't be any better than the underlying oscillator, but
> still, I'm not sure it's an entirely appropriate specification to be
> calling out.

It is, but only if you are asking the right question.

If your question is "How much RMS frequency noise does the oscillator 
contribute at observation time tau?" then the ADEV is for you.

If your question is "How stable will my oscillator be over observation 
time tau?" then the ADEV isn't going to give you remotely everything.
As tau increases, systematic effects will dominate. Notice how you 
limited yourself to constant temperature. The constant temperature hints 
that you have things changing with temperature, and that is a systematic 
effect. ADEV was never meant to solve that for you.
Rather, you will have to characterize your oscillators systematic 
behaviors to such degree that you can with needed certainty estimate the 
systematic time or frequency errors you can expect. ADEV is very handy 
to shine the light away from those issues, which is much hairier. We 
have third-degree thermal curves, we have systematic drift (which needs 
appropriate model for it, the linear drift is a simplification in 
itself), you have memory effects and if your temperature shifts, how 
quick does it shift and how will the temperature gradient over the 
crystal be? ADEV never attempted to cover this.

When you see a bump in your ADEV which isn't motivated by a system 
behavior (none of them usually are) then you need to figure out what 
systematic caused it. ADEV might be one of several ways to illustrate 
that you have an issue, but that bump does not represent a good answer 
to the RMS frequency noise question, as the noise and the systematic 
will need quite different treatment to get comparable answers.

When comparing systematic and noise effects on jitter (where white phase 
noise dominates), the BER value of 10^-12 motivates a scale-up factor of 
about 14 for a white noise RMS number to match up against the systematic 
peak-to-peak number. Together the sum then needs to be below one cycle. 
This scale-up number can be different depending on the confidence 
interval you need. It however illustrates that you need to treat them 
separate and then combine them back in such a way that they give you the 
answer to the actual engineering question you ask.

> And, isn't "aging" (in the sense of slow long term drift) usually
> excluded from the ADEV calculation (e.g. you fit a straight line to the
> raw frequency data, and subtract that out)

It is and it should be. You need to make a model and parameters that fit 
your device. The confidence of those parameters will give you the 
confidence interval for it's estimated behavor, just as the confidence 
interval for the ADEV will be needed.

So, what was your engineering question, really?

Cheers,
Magnus


More information about the time-nuts mailing list