[time-nuts] Caveats on Allan Deviation with ultra stable oscillators

[Magnus Danielson]

John,

On 05/30/2014 10:31 PM, John Miles wrote:
>>
>> I usually don't use drift removal as I want to see the effects of drift!
>>
>> The effects of oscillators locking together are very apparent on both the
>> phase and AD plots when using a DMTD system. There was no indications of
>> such locking!
>>
>> My point was that if you are measuring ultrastable Quartz oscillators
>> against each other the AD at the higher Tau will not reflect the true
>> stability of the oscillators.
>
> Well, it will reflect the true stability if they're not exposed to the same
> drift stimuli, right?  Neglecting shared long-term aging trends, they should
> both end up in random-walk territory over the long run.   Otherwise, if you
> run the two oscillators in open air in the same room or even the same
> building, then they will respond similarly to HVAC cycles, diurnal cycles,
> and whatever other environmental changes are common to both.  That will make
> their ADEV look better than reality when you measure them against each
> other.

Indeed. Another way to view it is that you can that way consider it as a 
cancellation of those factors so you know how the noise performance 
behaves, which is what ADEV is all about. That may not be the best 
choice for other measures.

> It doesn't mean the long-term ADEV is necessarily invalid as a statistic,
> just that you probably haven't eliminated all of the common-mode influences.
> To the extent the two oscillators drift independently, the ADEV measurement
> is valid.
>
> For that matter, the isolation amps in your DMTD are also exposed to the
> same environment.  Their residual phase tempco should be much better than
> any quartz oscillator, but if they're worse for some reason, they may
> dominate the long-term measurement.   And of course, if you don't have
> enough isolation, you could be injection-locking the OCXOs in a really low
> bandwidth (days, perhaps).
>
> Shared power supply leads can also induce entrainment -- or even separate
> power supplies, if you run the leads right next to each other.  Some of the
> Wenzel ULNs seem to be susceptible to this if you don't add bypassing at
> their power terminals.
>
>> Only a Maser or high performance Rubidium (HP5065A) will reveal the true
>> behavior.
>
> True, because they aren't as sensitive to environmental effects.  But if you
> benchmark two nearby 5065As or masers carefully enough, their long-term ADEV
> will also look better than it really is, and for the same reasons.  (Some
> people have even reported similar behavior with cesium standards, although I
> don't see how that could happen.  There aren't supposed to be any
> first-order temperature effects in a CBT, and I'd think that any lower-order
> effects would be way beneath the tube's flicker floor...)

Cesiums and rubidiums is indirectly sensitive as it takes time for the 
loop to track in changes in the OCXOs frequency. This process isn't 
phase accurate, so it could creep in phase as a result of environmental 
changes. Depending on the clock, different amount of work have been made 
to handle this.

However, the differences in ADEV here is due to environmental effects, 
and they should not be measured in the ADEV context at all. ADEV is 
about to measure the random noise forms. Systematic effects such as 
environmental sensitivity is noise to that measurement. There are other 
measurements better aimed at such deviations. So, when you have 
cancellation of long term systematic effects you are in fact measuring a 
trued ADEV.

Systematic effects affecting the phase (or frequency) should be 
separated and presented separately to the ADEV. They should then be 
canceled out of the measurement data that you make an ADEV plot on.

In the end of the day, there is an overbelief of what works well in an 
ADEV plot.

Cheers,
Magnus