Hi
On Aug 26, 2013, at 2:01 AM, Magnus Danielson <[email protected]> wrote: > On 08/25/2013 08:25 PM, Bob Camp wrote: >> Hi >> >> The most common approach is to *assume* that the two devices are not >> correlated. SInce it's a negative, you really can't prove it. What you can >> do is to disprove it by finding and documenting a correlation. >> >> ADEV it's self has a confidence level based on the number of samples taken. >> What is normally reported is the calculated number, not the number plus the >> uncertainty. The same carries over to the square root of 2. It's simply the >> best estimate of what's going on. As long as they say what the do / do what >> they say, it's not really a problem. >> >> The next step is typically to build a couple more devices and start doing a >> simultaneous N way comparison. That will let you play with math and better >> estimate the performance of each of your devices. The best case would be to >> compare devices made by different labs using different approaches. That >> usually lets you spot the correlation issues between devices. > Exactly. As you have three devices, measuring them pair-wise you get > three measures and three un-knowns, and you can untangle the stability > of each individual. If you have yet more, you can get some confidence > levels also as it becomes overdetermined. > The only real gotcha is common mode drift. If all your gizmos are made same place / same time / same parts then they may drift the same way. In this case "drift" could correlate to environment rather than just to time (aging). Back when HP pretty much made all the Cs standards this was a common thing to worry about when setting up an ensemble of them. Of course you could always move your ensemble to an un-inhabited cave and ….. No, HP did not make the long tube Cs standards at NIST (as the NIST guys always love to point out) and they are very different animals than the ones you can buy. So, the international definition of the second has always been safe from manufacturer induced common mode. Bob > Cheers, > Magnus > _______________________________________________ > time-nuts mailing list -- [email protected] > To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts > and follow the instructions there. _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
