Measuring Allan-deviation of the best clocks available

[michael2]

Hello,

how is this done, measuring Allan-deviation of the best clocks available, say a hydrogen-maser atomic clock? I thought you always need a better clock to characterize the DUT. But there isn't anything better than a hydrogen-maser, isn't?

And, if this is possible, could this "mystical" measuring techniques applied to hobby scale? Say I have 4 OCXO and would like to find the best one and measure its Allan-deviation without using another helper like rubidium or GPS. How to approach this goal, just for the sake of fun and interest?

[ebastler]

I think that the standard approach is indeed to compare against a clock that is known to have better specs.

If your clock deviations are dominated by random, non-correlated fluctuations, you could compare your four oscillators to each other pairwise (i.e. measure one using the other as the reference). The pair that gives the lowest variation will be the two best oscillators. You won't yet know which of the two is better; it may be possible to figure that out from the other comparison pairs.

Of course this approach only works for uncorrelated drifts. It won't help if the oscillators drift in a correlated way due to systematic effects, e.g. room temperature changes. In that situation, what appears to be the "best pair" might still have massive, but well-matched drifts.

[rf-loop]

Hello,

how is this done, measuring Allan-deviation of the best clocks available, say a hydrogen-maser atomic clock? I thought you always need a better clock to characterize the DUT. But there isn't anything better than a hydrogen-maser, isn't?

And, if this is possible, could this "mystical" measuring techniques applied to hobby scale? Say I have 4 OCXO and would like to find the best one and measure its Allan-deviation without using another helper like rubidium or GPS. How to approach this goal, just for the sake of fun and interest?

Hydrogen maser short term stability is poor.   Example 2E-13 (1s allan)

State-of-the-art quartz crystal oscillators may have frequency stabilities lower than  6E -14
Allan deviation  over 1s  - 100s

Example cryogenic sapphire oscillators (CSO) short time stability is much better, like 5E-15 or something like it.

Short time allan measurement do not need very accurate  frequency, it need short time stability better than dut, and example CSO have it. (type to google: Cryogenic CSO ELISA  and you know more)

[edpalmer42]

Hello,

how is this done, measuring Allan-deviation of the best clocks available, say a hydrogen-maser atomic clock? I thought you always need a better clock to characterize the DUT. But there isn't anything better than a hydrogen-maser, isn't?

And, if this is possible, could this "mystical" measuring techniques applied to hobby scale? Say I have 4 OCXO and would like to find the best one and measure its Allan-deviation without using another helper like rubidium or GPS. How to approach this goal, just for the sake of fun and interest?

One way I know of to accomplish this is to use the so-called "Three-Cornered Hat" technique.  By simultaneously measuring three oscillators in pairs (i.e. A vs. B, B vs. C, and C vs. A) and then doing matrix math on the results, the individual results of each oscillator can be determined.  You can try to do the testing sequentially, but the math probably won't work out nicely.  The square root of -1 seems to pop up frequently.  The tests work out best when the three oscillators are nearly the same - maybe three of the same make and model.  (Let me think......Do I have three hydrogen masers?.......) 

The Timelab software program http://www.ke5fx.com/timelab/readme.htm automates the calculations.  That feature is new, possibly still considered to be in beta testing, so move cautiously.

Ed

[awallin]

And, if this is possible, could this "mystical" measuring techniques applied to hobby scale? Say I have 4 OCXO and would like to find the best one and measure its Allan-deviation without using another helper like rubidium or GPS. How to approach this goal, just for the sake of fun and interest?

three-cornered hat:
http://www.wriley.com/3-CornHat.htm

note that most counters (like 53230A or SR620) have a noise floor of about 2e-11/tau(s) so if you want to measure below that you need a DMTD or other high-resolution technique (TimePod or similar).

If you don't have TimeLab or Stable32 there's a python library for these calculation:
https://github.com/aewallin/allantools

[awallin]

Hydrogen maser short term stability is poor.   Example 2E-13 (1s allan)

State-of-the-art quartz crystal oscillators may have frequency stabilities lower than  6E -14
Allan deviation  over 1s  - 100s

AFAIK the best masers use good BVA quartz oscillators and discipline them with a time-constant of maybe 1s.
So the BVA itself will be as good (NOT significantly better like suggested above) than the H-maser at 1s. But at 10s or any longer time the H-maser will be much better.

FWIW the best primary frequency standards now produce RF at 9.2GHz via an optical resonator (laser light locked to a rigid cavity) and a frequency-comb that downconverts to RF. They are maybe 1e-16 at 1s, limited by the thermal noise of the mirror coatings in the resonator. This produces 'better/cleaner' RF than a BVA at 5MHz multiplied up to 9.2GHz.

[KE5FX]

Hello,

how is this done, measuring Allan-deviation of the best clocks available, say a hydrogen-maser atomic clock? I thought you always need a better clock to characterize the DUT. But there isn't anything better than a hydrogen-maser, isn't?

And, if this is possible, could this "mystical" measuring techniques applied to hobby scale? Say I have 4 OCXO and would like to find the best one and measure its Allan-deviation without using another helper like rubidium or GPS. How to approach this goal, just for the sake of fun and interest?

At the highest levels of play, they basically have to build more than one of whatever they're testing, measure them against each other, and separate the variances as well as they can.  In the simplest case, you build two, assume they're identical, and subtract 3 dB from the observed noise/instability.

Another technique takes advantage of the fact that the DUT's phase noise can be measured using multiple sources whose noise is worse than the expected DUT performance, but removable by cross-correlation.  You then do the appropriate math to convert the frequency-domain noise into Allan deviation, etc.  This is better than a traditional N-cornered hat computation in the sense that it's a vector operation that yields unambiguous data, but it has its own gotchas (see Sam Stein's article among others.)

One way I know of to accomplish this is to use the so-called "Three-Cornered Hat" technique.  By simultaneously measuring three oscillators in pairs (i.e. A vs. B, B vs. C, and C vs. A) and then doing matrix math on the results, the individual results of each oscillator can be determined.  You can try to do the testing sequentially, but the math probably won't work out nicely.  The square root of -1 seems to pop up frequently.  The tests work out best when the three oscillators are nearly the same - maybe three of the same make and model.  (Let me think......Do I have three hydrogen masers?.......)  The Timelab software program http://www.ke5fx.com/timelab/readme.htm automates the calculations.  That feature is new, possibly still considered to be in beta testing, so move cautiously.

Lots of things can go wrong with N-cornered hat measurements, all right.  Bill Riley's info from the Stable32 manual and elsewhere is a must-read.  Currently, only the beta release of TimeLab supports them (http://www.miles.io/timelab/beta.htm), and the only documentation is the mouseover help text for the Source A and Source B fields in the Edit->Trace properties dialog.

[JuKu]

They use portable clocks: The US national reference and the portable reference difference was this when it left, and this when it came back and in France, the diff was this. When the Finnish clock was in France, the results were this and when it was in UK, results were this. When the UK reference was in US, it showed this, in France this, and In Germany, this. And so on, you get the idea. After that, it is statistics and math.

I have some stories about Mr. Clock and his adventures in first class air travel, like if he got served food and how the dessert was divided between him and his assistant, does every passenger with a seat reservation really need to have a passport as well and so on.

[awallin]

They use portable clocks: The US national reference and the portable reference difference was this when it left, and this when it came back and in France, the diff was this. When the Finnish clock was in France, the results were this and when it was in UK, results were this. When the UK reference was in US, it showed this, in France this, and In Germany, this. And so on, you get the idea. After that, it is statistics and math.

maybe 20 or 30 years ago.
AFAIK none of the current primary frequency standards (i.e. cesium fountains - some of them cryogenic) are transportable. Some of the new generation optical clocks are built to be transportable but none have been transported while running AFAIK.
Time transfer by GPS-PPP and/or two-way geostationary satellite seems sufficient for timekeeping at the level which H-masers can perform. Optical fibers between all the UTC-labs would be nice and might be required if everyone wants to run optical clocks continuously for timekeeping.