Tony Hoyle <[EMAIL PROTECTED]> wrote:
> That's transmitted in plaintext over a relatively slow RS232 link, 
> interpreted by the operating system, then the NTP driver.. by the time 
> you get it it's way off.
>
> PPS helps (enormously.. without it it's hard to get within 200ms)

200ms happens with some serial clocks, but not all. Many are good
to +/-1 character (which is 1ms at 9600 baud) and a few to +/- one bit
(better than 1ms), although all of these have to be tuned via the
fudges.

> - but 
> even with that it's damned hard to get the GPS sync within 1.5 
> milliseconds of 'known' time, and the jitter is consistently 5ms or 
> more... that's just inevitable in a multi-tasking OS, even a relatively 
> idle one.

Not at all inevitable. PPS support in the kernel gives me jitters of
+/- a few microseconds. (Z3801A, Linux with Ulrich's PPS patches).

The difference between PPS out on a GPS receiver and "true" time at the
receiver is usually guaranteed to be less than a microsecond. RS-232
smears the pulse but you can always use differential signaling instead.
Interrupt latency is the remaining uncertainty but even then there are tricks
around this microsecond-level uncertainty.

Tim.

_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to