[time-nuts] Clock quality: alternatives to ADEV

[Magnus Danielson]

On 10/04/14 00:38, Hal Murray wrote:
>
> I've been watching the discussions and graphs for a while.  ADEV seems
> appropriate for cases where the noise pattern is "nice".  How does ADEV work
> if the noise isn't nice?  Are there alternatives?  What's the mathematical
> term for the type of noise that works well with ADEV?

ADEV and friends is intended to handle phase-noise of 1/f³, 1/f², 1/f 
and white and aid in separating them such that the strength of the 
individual noises can be established.

>
> I can think of 3 examples:
>    Crystal jumps

Crystal jumps is rare enough to not contributing much noise power, which 
ADEV is collecting with various filters. It is however a quite annoying 
random effect. Better characterized by other means.

>    GPSDOs going into holdover.

For hold-over, you need to characterize the drift of oscillator and 
other frequency controlling things (such as DAC and voltage-ref). In 
addition, environmental effects will all of a sudden be uncompensated.
ADEV isn't a good tool for systematic effects like these.

>    Power lines make all sorts of interesting not-quite jumps.

This is a systematic effect for which ADEV isn't well suited.

> Is there a way of characterizing that sort of event?  How do I turn a pile of
> data into a useful graph or chart?

You can make a TE plot (i.e. the phase plot) which illustrate these 
properties.

> What does an ADEV graph look like if the data has crystal jumps?

Depends on the length of the data, the more data before or after the 
jump, the less prominent the crystal jump polutes the ADEV plot.

> I'd expect that something like crystal jumps would follow some sort of power
> law: the bigger jumps would be less frequent.  But it wouldn't surprise me if
> the GPS or power lines had an underlying mechanism that turned into a
> different pattern.

Crystal jumps can most probably be modelled as a noise form, but the 
power will be very small, so it will not be giving much information in a 
ADEV plot. Rather, average jump length and average rate should be more 
of interest when considering their impacts.

Cheers,
Magnus