How to compare the performance of atomic clock?

[poorren]

Hello guys,
My question comes from evaluate two rubidium clock sources, one is bought as used clock with nice price, another is disassembled from an old basestation. I don't know the detailed spec of two clocks. After reading some material on time source measurement, I found the methods on the table (commerial counter TIC or dual mixer time difference method) require to use a more stable clock to measure another one. Because all the methods above are relative measurement, the confidence is based on priori knowledge on standard performance. So, my curiosity drive me to ask how to compare two clock with unknow performance (short term jitter and long term stability)?   Thanks for your time.

[Geoff-AU]

There's an old saying:

A man with one clock knows what time it is
A man with two clocks has no idea what time it is
A man with 3 clocks has slightly more idea than the man with 2 clocks

[ejeffrey]

If you have two identical clocks you can use the allan variance to measure the relative error between them and then assign that 50/50 to each clock since they are supposed to be identical.

If you have 3 clocks, you can use the "three corner hat" measurement to find the variance in each clock individually.

Both of these methods require the clocks to be statistically uncorrelated with each other which is often extremely difficult to achieve in practice.

[MikeK]

Compare to GPS time?

[jmw]

If comparing the PPS signal is what you want, this might do the job: https://tapr.org/product/tapr-ticc/

[poorren]

excellent open source work, I have starred it.

[poorren]

the gps seems to be not good enough for this. the 1pps signal derived by receiver MAYBE only have good lone term performance. the jitter introduced by weather is too much for this application.

[poorren]

Hi ejeffrey, Thank you VERY MUCH! That's the one I am looking for.

[ejeffrey]

Take a look at the plot from Stanford Research at the bottom of this page:  https://thinksrs.com/products/prs10.html

They show the allan variance of the Rb oscillator beating a GPS 1pps for intervals up to ~3 days. 

[coppercone2]

what if you put them all into a logic gate so when one of them drifts there is a short glitch. I wonder how fast it would need to be.