On 29/01/14 06:16, Hal Murray wrote:
Whenever I plot an ADEV chart for a given oscillator I see the diagonal line
descending from upper left to lower right.
However at near the end of the plot, in this case 3600 sec at 1 sample per
sec. the trend begins to reverse at about 1200 sec. in.
Is this a function of the shorter term noise being averaged out and some
other drift the major contribution to the ADEV?
If you collect enough data, it should be a V shaped graph.
On the left side, the error is dominated by measurement noise. Longer times
between samples average the noise over a longer time so things look better.
To be clear, for many of the time-counter-setups it will be the
instrumentation which provides this limit. There is two parts of this:
1) Single-short resolution - this one you fight by choosing instrument,
and you guessed it, the better the more money.
2) Trigger jitter - how limited slew-rate convert additive voltage noise
to time-noise, this one you fight in various ways to avoid building up
and then converting it to time-noise.
A common mistake is to assume you can average it out, but that gives you
a different measure which does not represent the ADEV values you are
comparing with. The time between samples will scale down the relative
impact of the time-noise, but not really average it.
The MDEV measurement does provide a form of averaging, and it is done in
a standard way, and helps to distinguish two noise-forms.
On the right side, the error is dominated by the drift in the clock you are
measuring. Longer times between samples give the clock more opportunity to
drift.
The drift typically reduces with the time the oscillator have been on,
so you want to turn it on and let it settle for at least a few days.
There should be several good URLs out there describing that. I don't have
one handy. It's pretty obvious after you see it.
Some of it is covered in the Allan Deviation wikipedia article. Let me
know what's missing (besides plots, gotta fix that).
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.
