[time-nuts] Interpreting the ADEV plot in TimeLab
jmiles at pop.net
Sat Oct 22 01:58:39 UTC 2011
> I use TimeLab and the HP5370B and a Prologix GPIB interface.
> After making several 10 Hr runs with the same oscillator I noticed
> When laying the successive runs on top of each other there are little
> and bumps that line up perfectly on top of each other.
> For example each run has a small blip at 10 seconds in, and a rise from 50
> seconds to 150 seconds
> This was happening when I was using the NI PCM-CIA interface as well.
> If time is the X axis and variation of frequency is the y axis , how can
> something repeatedly happen at 2 points in time.
The x-axis should be thought of as an indication of period, not absolute
time. The whole idea behind the ADEV plot is to tell you how accurate the
clock can be expected to be when observed at recurring intervals of T
seconds, where T is any given tau value on the X-axis.
This can be confusing because when you watch the ADEV trace appear over
time, it seems to advance at a rate reminiscent of the time label on the
X-axis. In reality this is because the graph can only be updated for a
given tau interval once enough samples are available at that timescale.
Say you' re checking your watch for accuracy on a day-to-day basis
(tau=86400 seconds). You couldn't make an accurate judgment if you only
checked it after a single day. You wouldn't feel confident in your t=86400s
accuracy estimate until you had checked it every day for a week, for
instance. Even then, you might observe that your watch gains an average of
2 seconds per day the first week, 3 the next week, and 8 the week after
that. For this reason, you'd need to run your ADEV test for several days if
you want to see what the graph looks like at 86400 seconds... and the error
is likely to be large when that part of the graph first appears.
In your case, if you look at the phase and/or frequency difference plots you
may be able to discern some kind of regular variation or structure that
recurs every 10 seconds. An ADEV test that runs long enough will often
exhibit small bumps corresponding to HVAC cycle times, day/night temperature
changes, and other effects. If you were sampling fast enough, you'd also
tend to see bumps at the power-line frequency and its first few harmonics.
More information about the time-nuts