[ntp:questions] Re: Using a Rubidium clock with NTP

Dave Andersen dga at no.spam.pobox.com
Fri Oct 8 01:37:11 UTC 2004


Geoff Horner <geoffh at bigfoot.com.au> wrote:

> With all this effort,  I'm starting to think the easiest way to
> get a good stratum 1 server is to use one of the  miniature OCXO
> that I've seen, and substitute the 14.31818 MHZ PC xtal.and with
> one of them and a Motorola M12+ GPS timing receiver, producing good
> results for little outlay. Maybe next clock....

  Actually, many Rb references will accept 1pps in, and
send 1pps out, correcting their internal drift based on the (long term!)
difference from the 1pps reference signal from GPS or whatever source
you happen to have.  See various oscillators like the PRS-10:

  http://www.thinksrs.com/products/PRS10.htm

With this setup, creating a stratum 1 is really quite easy.
Depending on the accuracy that you want your stratum 1 to have,
of course - your major source of error with a GPS-tuned Rb will
be the serial port interrupt latency and any delays in processing
and responding to network packets.

  The "GPS + {osc}" is a pretty general recipe for an easy to
use, easy to maintain refclock.  Devices like the z3801a, z3816a,
58540a (yes, there's an HP/Agilent bias in that list) all do 
exactly this with various degrees of ovenized quartz oscillators.
Higher end systems couple the GPS to an Rb oscillator for longer
holdover times, though most internally use an XO for the 10Mhz.  
Trying to do the drift compensation on your computer
is going to be limited by the accuracy of your ability to timestamp
the GPS input signal and Rb input signals -- in other words, rather
bad, on the grand scheme of things.

  If you're interested in building something a little better
than what a traditional PC will give you, take a look at Poul 
Henning-Kamp's timestamping with a Soekris pc page:

   http://phk.freebsd.dk/soekris/pps/

  Replacing the 14.318.. oscillator on your motherboard will give you
better NTP server performance, because NTP won't have to do as many
second order corrections, but since you're already feeding it pretty
good 1pps, and have other significant sources of error, it probably
won't do too much unless you're going for timestamping accuracy
on the computer itself.

  -Dave

-- 
work: dga - at - lcs.mit.edu                   me:  angio - at - pobox.com
      MIT Laboratory for Computer Science           http://www.angio.net/
      (note that my reply-to address is vaguely despammed...)
      bulk emailers:  I do not accept unsolicited email.  Do not mail me.



More information about the questions mailing list