[time-nuts] PRS10 MDEV
magnus at rubidium.dyndns.org
Wed Oct 17 08:29:20 EDT 2018
On 10/15/18 4:07 AM, Chris Burford wrote:
> Hello All,
> I've collected several data sets over the last few days from my RFS
> and I'm puzzled by what I'm seeing. A link to the plot is here:
> I'm looking at the averaging periods from about 100 thru 400 tau and
> noticed a "plateau" or a brief level off period and then it proceeds
> down and right again. Is this flat area a function of the averaging
> algorithms for the MDEV or is it something else?
Actually, that is where you start to see real stability data.
Look to the left of the plot, all 4 traces have very close traces from
tau 1 s to about tau of 100 s. This is becuase of your measurement
instrument noise, which falls with 1/tau^1.5 with MDEV and 1/tau with
ADEV. This is just the measurement resolution obscuring your real device
performance. It is only as this becomes lower than the actual DUT noise
that you see the actual DUT noise. The cause of the plataue then becomes
somewhat obscured as you don't see the full picture.
Another aspect is that the traces is on the short side for confidence
intervals of the measurements have gone down, as the truncated series in
itself can cause variations of the estimated value around the actual value.
One effect that can create similar behavior is a systematic signal
either as a side-band or as a modulation such as a sine that cause a
additive effect on top of the noise performance. This type of systematic
disturbance may be an effect in the measurement setup or in the device,
so care must be take to locate and eliminate it as it does not represent
the systematic noise of the source.
There is a few other effects that can cause similar behavior.
It is recommended to monitor the phase deviation data plot. Whenever I
see deviations like that I ask for the phase plot, which typically also
get the frequency compenated out and only look at the residues of that
prediction. It may also be convenient to look at the FFT of the data.
ADEV and MDEV is a poor way to estimate stability that has systematic or
semi-systematic behaviours, it's only intended for truely random noise
of the noise-slope types. Other noise-types should be identified by
other means and compensated out prior to ADEV/MDEV processing.
> I have spent considerable time in reading "Techniques for Frequency
> Stability Analysis" by W.J. Riley and a host of other publications
> (NIST, etc) in preparation for this. I would enjoy hearing from any
> readers that may have some insight as to what I'm seeing and or
> possible causes.
> Thanks for reading.
While that is a good reading, it doesn't go in depth onto how to
interpret the plots properly. Some have been written by David Howe as
found in NIST T&F publication list.
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
More information about the time-nuts