[time-nuts] xDEV "spread" on parts of data

[Ole Petter Ronningen]

Hi, All

I am currently working on some lather long time-series, and I find myself a
little unsure of how to present the data. I've attached a TDEV plot below
(that I hope makes it through) for illustration purposes, but the same
dilemma holds for ADEV/MDEV etc. It concerns how instability is averaged
out on long data series.

The pink line below is the TDEV of the whole series and represents about 14
days of data. The blue lines are generated by using the "trace history"
facility in TimeLab, each line representing roughly 12-hour "chunks" of the
same data.

It can be seen that some of the blue plots are "worse" than the pink plot -
not only at the tail end, where it might be expected but also at short tau.
The implications are of course that there are stretches of data where the
TDEV is better/worse than others, but it is also clear that the TDEV plot
of the whole data in some sense masks this.

 ADEV/TDEV/MDEV is what it is, and that is fine - but this is still
information that is relevant when looking at the plot. The question at hand
is how to convey this information in an accurate way? The error bars are
really no help.

Some kind of "min/max TDEV per tau displayed as error bars" comes to mind,
but there are a number of issues with that approach.

I am also hoping that perhaps this has been thought of before - is there
some "standard" (or even good, or half-decent) way of calculating/showing
this "spread of xDEV"?

Thank you all,
Ole


[image: image.png]
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 303751 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20200916/81fc07d7/attachment.png>