Frequency Reference for NTP Use?

Brad Hards bhards at
Sun Jul 14 10:54:57 EST 2002

On Sun, 14 Jul 2002 00:27, Alex Satrapa wrote:
> On Saturday, July 13, 2002, at 10:06 , Brad Hards wrote:
> > A few seconds a month is probably going to require some kind of
> > temperature
> > stabilisation.
> When you move from temperature compensated oscillators to ovenised*
> oscillators, you move from 1 second/week to about 1 second/year
> stability.  You also move from the order of $x00 to $x000.
This type of stability issue is a bit misleading. Allen variance is a curve
that varies over the time you are examining. There is nothing that stops
an oscillator having 1 second/week and 1second/year stability. 
The dodgy NMEA outputs on GPS receivers are probably in this 
sort of case (although hopefully it is a bit better than 1 second).

> > Unless you are really concerned with GPS availability, I'd look for
> > a GPS with a 1PPS output.
> The catch is that such units are usually designed to be rack mounted,
> and they're usually in the order of $1500.  My budget is significantly
> less than that.
One of the links I posted had a thing called gpsclock 200 at about US$380.
There are likely others, but my knowledge of SPS receivers is a bit
scant, except in aviation.

> I'll probably try to read through the NMEA driver to figure out where it
> gets the second-edge from, and find some way of reducing the jitter -
> such as the first character in the stream, rather than the first
> character of the time stamp.
That will still likely suck. The problem is that the output routines run with lousy
priority in most GPS receivers. So everything else that is going on
(downloading almanacs, user input, constellation changes, etc) makes
for jitter on the NMEA output. 

-- 22-25Jan2003. Perth, Australia. Birds in Black.

More information about the linux mailing list