> I have already commented on this in another thread but to reiterate. > The test that John performed that for a range of Tau that was possible > to be calculated for the given measurement period, both methods > produced the same results for each and every value of Tau, not for a > single value of Tau.
Keep in mind, though, that tau is only one axis. I had three reasons for using a noisy HP 5062C cesium standard in the medium-term test ( http://www.thegleam.com/ke5fx/tpll/small/5062c.gif ): 1) To compare the instruments across a wide range of timescales from short-term to medium-term, watching for signs of accumulated error that the software might exhibit in records >> 100K points; 2) To observe a familiar, repeatable source over multiple runs, looking for general cases where one instrument might report different statistics versus the other; and 3) To look at several different noise *slopes*, looking for specific cases where filtering or integration might cause the instruments to respond differently depending on the spectral characteristics of the noise in a given tau region. Bruce's concern is largely that of (3), whereas Warren historically has been satisfied when condition (2) has been met. Both criteria are important. If you look back at the first test with only the two crystal oscillators: http://www.thegleam.com/ke5fx/tpll/small/10hz_with_resid.gif ... you'll see that while the blue and magenta traces agree pretty well, they are also almost flat, with a slight rise at longer timescales. This is characteristic of flicker frequency and random-walk frequency noise. These are only two of roughly 5 different noise slopes that are likely to be seen in various oscillators, so this plot by itself is useful mainly as a quick sanity check. The 5062C plot, though, has a characteristic 'hump' that you see in cesium and GPS sources, which is almost the opposite of the 'trough' you tend to see from undisciplined crystal oscillators. This means the two plots are complementary to some extent: various noise characteristics that don't appear in crystal oscillators are present in the 5062C plot to some extent. It is meaningful, I think, that the two instruments still exhibit good agreement. If you zoom in on the slope at the far left of the 5062C plot, though, you'll see the beginnings of some disagreement between the two. It gets worse at shorter timescales, where the cesium loop is just starting to degrade the internal OCXO's stability. This is an example of Bruce's concern, I think: whether it's due to Warren's hardware or my software, there *is* at least a slight dependence of measurement fidelity versus spectral content. In my opinion, this discrepancy is nowhere near large enough to be of any practical consequence, especially since it's occurring at timescales where I'd normally look at phase-noise measurements instead. I would have no problem using either instrument in everyday work, tweaking GPS clocks and surplus Cs/Rb standards, comparing different crystal oscillators, optimizing temperature controls, and such. But this list is for nuts, right? If you're worthy of being called a nut, then the 2-3 pixels of separation in the blue and green traces near t=0.1s may be enough to keep you up at night. I lack both the moral authority *and* the math skills to argue against that. -- john, KE5FX _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.