[time-nuts] GPSDO recovery from holdover

SAIDJACK at aol.com SAIDJACK at aol.com
Sun Dec 2 01:13:00 UTC 2012

Hi Magnus,
yup, at the levels we are interested in, a prefix or two sometimes doesn't  
make any real difference :)
Most of the time typical GPSDO's won't ever drift out of a say  +/-100ns 
window. If they do, then the antenna must have been shot off by  someone, or 
something else must have gone horribly wrong.
Just for fun I attached two phase correction examples from a FireFly-IIA,  
and a CSAC GPSDO. Both were essentially brand new and not yet calibrated 
when  turned on, and thus you can see a large EFC variation over the first 15 
minutes  or so as the frequency stabilized.
Then you can see the phase stabilize slowly, this takes about 1.2 hours for 
 the FF-IIA with a much more aggressive loop setting, and about 3 hours for 
The most perplexing fact for me is that while you can clearly see the exact 
 point at which the phase has stabilized, you cannot really see any 
corresponding  change in EFC behavior at that time. You can see a large EFC voltage 
change  initially as the frequency stabilizes after power-on, but then it 
goes into the  noise floor. This shows that the EFC corrections for phase 
error are essentially  smaller than the proportional noise floor of the loop!
The maximum phase error in these plots was about 100ns for the CSAC,  and 
230ns for the FF-IIA. Here we can see that the FF-IIA has a much more  
aggressive loop approach (~5x more gain on the phase correction). Since the CSAC  
is an atomic clock we can increase the time constant quite a bit and make 
the  loop much less aggressive.
In a message dated 12/1/2012 14:39:58 Pacific Standard Time,  
magnus at rubidium.dyndns.org writes:

One can  also wonder if the limit is relevant, as you are about to 
resolve a rather  catastrophic situation where you already cause 
interference, so moving out  of it quickly should be first priority and 
only when back to reasonable  time-error would it be relevant to obey 
frequency error limits.

The  transmitters and the recievers would be able to follow, as they have 
large  enough bandwidth for it.

> But if you set the loop parameters more  aggressively to 1ns/s as in your
> example, it would take less than 20  minutes to correct 1us.. Not 12hrs.
> Unless you meant to say  ms?

What's a off by one prefix among friends?

But still, one has  to be  careful.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: phase_corrections.zip
Type: application/zip
Size: 19678 bytes
Desc: not available
URL: <http://www.febo.com/pipermail/time-nuts/attachments/20121201/c91e2265/attachment-0001.zip>

More information about the time-nuts mailing list