[time-nuts] Allan variance Vs Plain Old Accuracy

Tom Van Baak tvb at LeapSecond.com
Tue Feb 26 01:49:57 EST 2008


> Something I rather miss is some good old phase or frequency plots -
> especially if done at different timescales - which seem to be becoming
> rather less common now. 
> As well as having a plot of ADEV or its relations, seeing what the
> reference is doing and when is useful, and most ADEV plots using so
> few values of tau does not help.
> 
> Angus.

Angus,

I agree with you. Each representation has its merits and a
combination of several is often necessary to tell the whole
story. One feature (problem?) of ADEV is that it's a statistic
and so some anomalies, clearly revealed in the phase or
frequency domain, are hidden in the statistics. On the other
hand, the ability of ADEV to resolve noise types, gives it a
greater power than linear strip charts.

A recent example of using all three graph types is here:
http://www.leapsecond.com/pages/8607-drift/

On your comment about too "few values of tau" -- since adev
is calculated by software tools these days, there is less reason
to use only a few tau per decade. What I and others sometimes
use is "all tau" or "many tau" where the tools calculate as many
tau as necessary so that the plot is as close to a continuous,
non-interpolated line as possible. For a well-behaved oscillator
this is overkill (see above example) since all the points land on
the line anyway, but for oscillators with any sort of periodic
frequency modulation (e.g., an OCXO with bad tempco), the
many tau method turns an adev plot into a sort of spectral plot.

A cute example of the power of a many tau adev plot is here:
http://www.leapsecond.com/museum/earth/

This is a free adev tool that calculates as many tau as you like:
http://www.leapsecond.com/tools/adev1.htm

/tvb




More information about the time-nuts mailing list