[time-nuts] quartz drift rates, linear or log

Attila Kinali attila at kinali.ch
Sat Nov 12 17:25:37 EST 2016

On Sat, 12 Nov 2016 13:54:14 -0800
"Tom Van Baak" <tvb at LeapSecond.com> wrote:

> 2) attached plot: TBolt-10day-fit0-e10.gif ( http://leapsecond.com/pages/tbolt/TBolt-10day-fit0-e10.gif )
> Here we zoom in by changing the Y-scale to 1e-10 per division. The X-scale 
> is still 10 days. Now we can see the drift much better. Also at this level 
> we can see instability of each OCXO (or the lab environment). 

These look like textbook examples of random walk frequency modulation.
As this is a random process it is not surprising that they look different
for each oscillator. 

> 3) attached plot: TBolt-100day-fit0-e08.gif ( http://leapsecond.com/pages/tbolt/TBolt-100day-fit0-e08.gif )

Do you know what caused the frequency jump of CH18?

> So is it linear or log? The answer is it depends. Now I sound like Bob ;-)

Hehe. There are worse things than sounding like Bob :-)

Yes, depending on the data you show, it is not clear whether one
should do a linear or a logarithmic fit. But keep in mind that
a logarithmic curve can look like a linear curve depending on the
parameters. Also Depending on the exact properties of the noise
and the evaluation function, a linear fit might be better on a
logarithmic curve than a logarithmic fit. We are talking about
system identification under noise, after all. And things become
strange when noise involved. Even more so when it is not white
gaussian noise.

			Attila Kinali
Malek's Law:
        Any simple idea will be worded in the most complicated way.

More information about the time-nuts mailing list