[time-nuts] How can one measure ADEV of a good oscillator?

Magnus Danielson magnus at rubidium.dyndns.org
Mon Dec 1 17:27:34 EST 2014


Tom,

On 12/01/2014 11:05 PM, Tom Van Baak wrote:
>> I think I have a flaw in my understanding of this.
>>
>> How can something like an SR620 measure the ADEV of an oscillator,  if the
>> oscillator is of a similar or better than the reference fed into the SR620?
>
> The key is to realize that all measurements are actually just comparisons: the instrument itself has no knowledge if the REF or your DUT is the better frequency source.
>
> If you use a SR620 to measure a TCXO then you are [mostly] measuring the DUT.
> If you use a SR620 to measure a cesium then you are [mostly] measuring the REF.
>
> I say "mostly" because no frequency standard is perfect, and the instrument itself contributes noise to the measurement. So strictly speaking, any measurement is always the sum of REF noise, instrument noise, and DUT noise. Never assume any of these are zero.

Indeed. You will need to learn the instruments limits, and assess when 
you are sufficiently away from the instrument or reference limits so 
that you to some degree can trust that your DUT is dominating the plot.

Beyond the tools limits, the way you set things up can significantly 
impact your measurement. Also, some limitations such as drift may 
require you to compensate it to see the random noise behind the 
systematic signature.

One trick to overcome the white-noise limit of the counter, is to plot 
the modified Allan deviation rather than the Allan deviation. This 
causes the white noise limit to have a 1/f^3 slope rather than a 1/f^2 
slope which will have the effect of the measurement hitting the 
oscillator behavior at a lower tau than otherwise. You can gain more by 
making measure more often, as the modified Allan deviation is then given 
the oppertunity to even further surpress that noise. This also serve as 
a tool to verify that your measurement is indeed white-noise measurment 
limited.

>> I see plots of ADEV  for hydrogen masers, but I can't understand how this
>> can be measured from the phase data unless the reference is better than the
>> DUT, which is not going to be possible with a good hydrogen maser.
>
> You measure a good hydrogen maser with a better hydrogen maser, or with another good hydrogen maser.
>
> The same is true if all you have a couple of OCXO.
>
>> I was thinking it might be possible if one has 3 oscillators and 3 time
>> interval counters to perhaps solve 3 simultaneous equations. I can't prove
>> that, but it seems intuitively correct.
>
> Correct. That's what we do. And you can use more than 3 if you want.
>
>> Also I have seen graphs of both Allan variance and Allen deviation.  Both
>> are typically 10^-12 for a decent oscillator, but given the variance and
>> standard deviation are related by a square root, they can't both be around
>> 10^-12.  I would expect to see values of 10^-6 or 10^-24, but I don't see
>> such dramatic differences from 10^-12.
>
> In the 70's it was common to [incorrectly] call it Allan variance. These days it is [correctly] called Allan deviation, or root Allan variance. Note that it's also called the two-sample variance/deviation, especially by mathematicians.

Allan deviation is the root square of the Allan variance.
Any number you usually see listed is most likely the Allan deviation.

There is uses for the Allan variance, but for most graphing purposes, we 
want the Allan deviation.

Dave actually jokes himself about his Allan deviation. :)

>> If I see numbers around 10^-12 on an OCXO,  is that the Allen variance or
>> Allen Deviation?
>
> It's Allan deviation. Not Allen. Not variance.

Indeed.

Cheers,
Magnus


More information about the time-nuts mailing list