[time-nuts] Method for comparing oscillators

[Magnus Danielson]

Steve Rooke wrote:
> 2009/8/7 Magnus Danielson <magnus at rubidium.dyndns.org>:
>> The Hadamard processing (Dev, ModDev or Tot) gives another B term, which
>> considering that B is fairly small gives a drift-gain. As the series
>> progresses, the drift derivate dies away faster (1/t^2 rather than 1/t) than
>> for Allan processing so the longer time sequence used, the better
>> suppression of the drift mechanism (which is true for Allan processing too).
>>
>> So, in this context Hadamard is better... but it still does not nail it.
>> It may be sufficient however. Estimating A and B and remove the trend from
>> the data isn't too hard.
> 
> To a certain extent it depends a lot on the oscillator that is being
> measured. Allen Deviation is considered as being the best tool for
> this type of work but it does suffer when the subject under test has
> significant drift.

Allan Deviation isn't qualitatively the best tool, just the most known.
Look at TotDev or TheoDev as they converge quicker and produce better 
confidence interval for the same amount of data.

> For the quality of xtals and sources that some
> time-nuts are testing, this is unlikely to be a huge problem but for
> lesser sources this is a real factor hence my suggestion for the use
> of Hadamard Deviation.

Having done the exercise on a fellow time-nuts measurements I beg to 
differ. It became clear that estimating the drift as a linear static 
component and then calculate ADEV with raw samples and ADEV with drift 
compensated samples it became clear that raw samples ADEV was infact 
drift compensated as it leveled out on the drift value, which is expected.

You can't make the above assumption unless you know what the drift is 
and know that the ADEV is above your drift level for the intended tau 
range. The reason I keep pointing this thing out is that after having it 
pointed out in several sources relating to how one does real 
measurements I have done the exercise and been able to remove the 
limiting drift component.

It's not advanced processing, so just do it rather than argue against it.

> For practical purposes though, the xtals we use
> are generally embedded as part of a GPSDO which will compensate for
> the drift in the oscillator but cannot practically compensate for a
> noisy xtal and HDEV would make comparing one source with another easy.

The part of the noise being at taus longer than the loop will be 
replaced with the GPS-receivers output noise. Measuring loop-locked 
oscillators isn't the same as stand alone oscillators, so if that is a 
way to remove drift, it will give you false readings if you beleive you 
are measuring the oscillator itself.

Cheers,
Magnus