[time-nuts] TCXO improvement

[Magnus Danielson]

Hi,

On 2019-08-01 08:50, Tom Van Baak wrote:
> An email came in asking if it was possible to improve the performance
> of a TCXO if one monitors the temperature of the PCB or enclosure and
> then applies timekeeping corrections in s/w based on that data. I
> don't have specific P/N or other details so treat this as a generic
> question. Has anyone tried this?

I had intended to do exactly this, but never got around to, since I
didn't have to. We put a hood over the oscillator, but the temperature
readings where really useful in that process. That ended up being good
enough for our needs, the time-errors accumulated over a shift reduced
to 1/3.

There will be possible to make some improvements, but there is
limitations to how well you can do it.

You probably want to do a least-square fit to some model. You want to
look at the residuals from actual and square-fit response in order to
see when you hit the limit of the model. The residuals will never be
white noise only which is what you would get as a success.

Crystals is temperature gradient sensitive, so once ability to
compensate that is somewhat limited. Depending on the quality of the
TCXO, it's match may be more or less good. It's for the less good that
you can get improvements.

The drift behaviors is what end up being hard for a simple TCXO to
compensate for, where as we will see it drift after a temperature step.
It will for sure be one of the limits.

There is more subtle differences.

Cheers,
Magnus