[time-nuts] ADEV computed, now what?

Charles Steinmetz csteinmetz at yandex.com
Wed Feb 5 06:47:01 EST 2014


Bob wrote:

>here's the result for 1PPS vs 10MHz for my GPSDO, as measured by a 
>5334B clocked by the same 10MHz.    I don't know how to read these, 
>but 6,3,1,6,3,1 etc. doesn't look normal.

The adev results you obtained look very much like the "adev" results 
reported by Lady Heather, very likely for the same reason -- you do 
not have any independent standard by which to measure.

 From what you say above, it appears that you measured the time 
interval between one edge of the PPS pulse and the next zero-cross of 
the GPSDO 10 MHz, using a 5334B clocked by the GPSDO 10 MHz.  Is that 
correct?  I take it the GPSDO is disciplined by the PPS?

Bear in mind (i) that you are comparing one noisy source to another, 
and (ii) that the errors are correlated more and more strongly as tau 
increases beyond the point where the discipline loop begins 
controlling the 10 MHz oscillator.  At small tau, the GPS PPS is very 
noisy (much noisier than the 10 MHz oscillator), and it gets better 
and better as you average for longer and longer periods until at tau 
above 10k it's reasonably decent.  Your GPSDO should leave the 
oscillator more or less on its own at low tau (where the jitter in 
the oscillator is lower than the jitter in the PPS), and correct it 
beginning at some longer tau where the jitters are comparable 
(continuing to even longer tau where the jitter in the PPS is lower 
than the jitter in the oscillator).

So, measuring as you appear to be doing, at low tau you are 
essentially measuring the improvement of the PPS with averaging -- 
10x per decade -- using the essentially undisciplined 10 MHz 
oscillator as a standard.  At some point, you would expect to reach a 
floor where you would essentially be measuring the residual jitter in 
the disciplined oscillator and the PPS.  In fact, you can see this in 
your results starting around tau = 500, but your series does not go 
far enough to show the floor clearly.

Bob suggested that you are measuring the trigger error of the 5334B, 
and that may be contributing to your results as well.  With good 
measurement techniques, the outer bound of the 5334B trigger error 
should be less than 1nS, probably more like 200-300pS.  Your actual 
error is probably significantly lower than this outer bound unless 
something is wrong with your setup.  The trigger contribution to your 
computed adev should also fall 10x per decade with increasing tau.

Best regards,

Charles





More information about the time-nuts mailing list