[time-nuts] I've been thinking about a GPS receiver experiment
magnus at rubidium.dyndns.org
Wed Oct 25 21:18:26 UTC 2017
On 10/25/2017 12:42 AM, Mark Sims wrote:
> I did a quick silly experiment where I took a PRS-10 disciplined by an X72 which was disciplined by the PRS-10. The result seemed to have created a rupture in the space-time continuum. Nobody was happy... they didn't seem to agree on who was in charge. I need to try it again now that I have my X72 interface boards back from China and can properly connect to the X72.
No, you set up an oscillator so that is why you have that problem.
Each rubidium has an integrator and each PI loop adds another, and now
that you wire the feedback over them you have an unbalanced 4 pole
system. Already a third degree system is somewhat of a challenge to keep
stable, so fourth degree... sure, it can be done, but not that way.
I have been looking at mutual synchronization and there is some
classical articles on it and recently some work have been done too.
A good mutual synchronization means that you find your aggregate clock
to be at the middle frequency and phase of the two separate clocks. This
also means that you want to drive your clocks towards each other, so one
is to raise frequency and the other lower frequenncy. For un-equal clock
EFCs, you need to scale the driving force accordingly. For more fancy
setups of weighted clocks, you need to apply correct weights so that the
better clock moves less than the clock being worse.
Another factor of the control is that delay needs to be compensated for,
and this have already been analyzed in the classical mutual clock setup,
but also exists in modern setups, and I happen to touch on that subject
in an article, since the delay margin, which is just a different version
of phase margin, needs to be adapted in accordance with the control loop
bandwidth. With the right key-words you can find out more. I did that
work on automatic power-grid stabilization and it was kind of
interesting to forge an analysis out of three independent
research-fields, so I found myself instructing a PhD student how she
would drive here simulations into self-oscillation... and they sure did
and we learned stuff. Ah well.
Anyway, mutual synchronization require some thought, but ends up not
being all that different in the end.
> BTW, the firmware based disciplining on the X72 seems to be rather crappy. It has lots of trouble locking to less than perfect 1PPS inputs. For instance it goes into holdover mode over half the time when driven by a Ublox LEA-5T (which seems to have like +/- 60 ns jitter on the 1PPS. It does lock fine with a Tbolt... but needing a GPSDO to discipline a rubidium sort of defeats the purpose of disciplining a rubidium.
No, not really. The rubidium would be the real hold-over clock.
> Lady Heather now has an X72/SA22 disciplining routine built in that works fairly well using the LEA-5T. Adevs are in the mid E-13 range at 10,000 seconds and the 1PPS output is in the +/- 50 ns range.
More information about the Time-nuts_lists.febo.com