[time-nuts] GPSDO Alternatives

Jim Lux jimlux at earthlink.net
Fri Dec 7 15:07:20 UTC 2012

On 12/6/12 11:12 PM, Hal Murray wrote:

> Suppose you just implement a simple bang-bang control.
> Suppose the EFC is 1 volt and the frequency is correct but the GPSDO
> phase is a bit early relative to the GPS PPS.  So the FF says early
> and the software says go-faster.  That keeps happening for a while,
> the frequency keeps getting faster and faster.  Finally, the GPSDO
> PPS catches up with the GPS PPS, but now it's frequency is way fast.
> The FF says go slower, so the control software starts dropping the
> EFC.  But the frequency is still way too high so the error is still
> increasing.  After a while the frequency gets low enough so the
> PPS/phase error starts catching up.  Eventually the PPS error crosses
> over, but by then the frequency offset is way way low.  ...  Isn't
> that cyclic pattern stable?
> Is there a simple tweak to break that loop?  Do you first have to
> recognize that you are in that mode?  If so, how?  ...

yes.. what you've described is essentially a first order control loop. 
You can add higher order terms (e.g. integral or derivative) so that you 
don't get overshoot.

> I might be able to do fix that in software by looking at the times
> when things change state.  Suppose it's 193 seconds between the first
> early and the last early and that the EFC went from X to Y.  I think
> that's enough info to work out the crossover point and work back to
> the desired EFC.

Yes.. that's another approach.. you figure out what the model is, and 
solve backwards.

> But that all sounds too complicated.  What would hardware-only guys
> do with a 1 bit A/D?

More information about the time-nuts mailing list