[time-nuts] Good references on holdover?

[Hal Murray]

javier.serrano.pareja at gmail.com said:
> We have never worked on holdover, and I am wondering if we can do something
> smarter than the obvious feeding of some constant voltage to the VCXO, based
> on averaging during the locked state. Does anybody know of any good
> references on holdover?

I doubt if you will come up with anything better.  Maybe if you have big 
temperature changes, it might help to track the recent slope.

Is your PLL analog or digital?  I'll assume digital since it's hard to hold 
analog voltages stable for several seconds.

I suggest capturing some data from a live system.  Say 1 minute or 1/2 hour, 
whatever you can get with minimal work.  Then scan it by eye to get a feel 
for what it does in normal operations.  Then maybe patch your collector to 
only collect data for big values of the types of stuff that look interesting. 
 Do you see any jumps?  What's the biggest change over your worst case 
holdover?  (or a bit longer for caution)

I'm assuming you can compute or measure the frequency change per count.  Do 
you see anything that will go over your limits?  I assume you have some 
target limits rather than are doing this just because it feels good.  If not, 
compute the predicted limits and see if your users are happy with that.

One thing to watch out for is recovery.  Do you have something like a PPS?  
If so, you have a decision to make.  Do you want the PPS back to the correct 
timing as fast as possible or can it get there slowly?  The PPS fixup time 
will be the integral under the frequency offset curve.  You can have a tall 
narrow spike, or a short wide bump.  For the spike, the frequency will be way 
off, but not for long.  For the bump, the frequency will always be close, but 
it will be off for a long time.


-- 
These are my opinions.  I hate spam.