[time-nuts] 1 pps correction

[Chris Albertson]

On Sat, Mar 31, 2012 at 7:00 AM, Azelio Boriani
<azelio.boriani at screen.it> wrote:
> And here:
> www.cnssys.com/files/PTTI/PTTI_2006.pdf
>
> Anyway, using a Dallas/Maxim DS1023-100 delay line (and a microprocessor,
> of course) you read the @@Hn data from the iLotus M12M and apply the
> correction to the delay line. Of course the delay line cannot anticipate
> the PPS, so that you have to set a "zero" at the center of the delay line
> total span. This fixed delay can, in turn, be nulled out using the M12M
> capability to displace its PPS with respect to the UTC. Please, read the
> M12M user manual and the DS1023-100 data sheet


It's clear that a fixed offset combined with a variable delay can work
in theory but how much noise (jitter) is added by the delay line?  The
uncorrected M12M is so god that I wonder of the delay line correction
is worse than the problem it seek to address.

Maybe a better way is to use the method of NTP.  It uses the PPS to
snapshot a counter and stores the value in RAM.  It also tracks the
sawtooth functionsand allies the corrected to the value stored in RAM.
  So you get the correction but no extra noise for a delay line.   But
then in most real-world NTP systems the counter has only uSec
resolution so the sawtooth is mostly moot.  But the idea perfectly if
you have a better counter.   But only for the one system
-- 

Chris Albertson
Redondo Beach, California