[time-nuts] GPSDO disciplining algorithms
kb8tq at n1k.org
Sun Apr 19 09:19:12 EDT 2015
> On Apr 14, 2015, at 12:05 PM, Alan Ambrose <alan.ambrose at anagram.net> wrote:
> Hi all,
> Thanks for the interesting feedback everyone, and thanks Attila, for your specific references - some reading ahead.
> In principle, I should be able to remember all the stochastic / standard control theory, but I'm sure I'll have to 're-remember' it.
> An interesting point re 3 use-cases and re the vast majority of commercial algorithms being made for telco purposes. Does that imply that 10MHz lab-instrument feed for frequency / timing etc would ideally have some different algorithms? A series of algorithms (or at least a series of parameters for the same algorithm) targeted at particular time horizons?
I suspect there are actually a multitude of use cases. Time Nuts (in general) have little interest in holdover. For the guy doing a CDMA design back in the 1990’s (when our eBay gear was being designed) it was a major
check box that needed to be addressed. The need for precise time in holdover has continued on into some (but not all of) the newer wireless standards. It’s still a check box that gets addressed. It’s hardly the only thing
that is a variable.
Here on the list we seem to divide into “time nuts” and “frequency nuts”. Let’s say we are designing a GPSDO for each of them. Our OCXO is bumped by the dog and it goes quickly (< 1 sec) off frequency (by ppb’s). What do we do?
For the frequency nut, we just tell our trained flea to adjust the trimmer so it comes back onto frequency. Our task is done, since we know from talking to our frequency nut that his longest gate time is 1 second.
For the time nut, we observe that we’ve lost a lot of phase (many ns). We tell the trained flea to get it back on frequency *and* then adjust it a bit past that point. He holds it there until the phase error walks out (we’re back to nearly zero ns). Then he adjusts it back onto frequency at a rate that zeros the phase error.
Are there really two different measures here (time and frequency) - of course not. People *do* tend focus on one versus the other. Let’s refine things a bit:
Our frequency nut listens a bit and decides that he’s been to simple in his request.
Now we see the bump, and we adjust the OCXO back on frequency (just like above). This time we *do* go past zero, but only by < 3.625x10^-11 (his new limit) until the phase error is zeroed out.
Listening, our time nut has modified his request:
There’s this big S curve in the phase plot and it “bleeds” through to his ADEV / MDEV / (Bob Dev :) and he wants that reduced. Back to playing with the filtering.
On a larger scale - Do you allow step jumps in the PPS? If I move the PPS 100 ns at a time, the “second” it marks is off by a lot. The time it takes to get back to “on phase” is shorter (no need to slew the OCXO). Will this make my frequency nut happy (less OCXO slew)? maybe. Will it make my time nut happy (faster to time) ? maybe.
Lots of choices. Unfortunately, "don’t let things bump the OCXO" is something that you don’t get to pick.
> I have found these recent-ish docs if anyone else is interested...
> Advantageous GPS disciplining of the OSA BVA oscillator
> DESIGN AND VERIFICATION OF A LOW-POWER GPS-DISCIPLINED OSCILLATOR FOR USE IN DISTRIBUTED SENSOR ARRAYS
> Unbiased predictive steering of local clocks utilizing GPS 1PPS time signals
> Regards, Alan
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
More information about the time-nuts