[ntp:questions] Re: Errors with NMEA refclock

David J Taylor david-taylor at blueyonder.co.not-this-bit.nor-this-part.uk.invalid
Tue Jan 31 16:20:21 UTC 2006

Dusty Bin wrote:
>> David J Taylor wrote:
> [...]
>>> Connecting the same GPS to a Windows 2000 system gives the error
>>> message:
>>> clock GPS_NMEA(1) event 'clk_fault' (0x03)
> [..]
>> Seems I'm getting these same errors on the XP system as well.
>> Sigh!
> Just out of interest, does it stop the system syncing?  I'm sure I
> reported the same issue ages ago, but in my case the system seemed to
> sync, but the clock had a vast amount of jitter.
> I was using a Garmin GP25 with W2K, but it's not connected at the
> moment.
> As far a baud rates go, I guess the real point is to use PPS if you
> want accuracy - just for us windows users, that isn't possible...

No, it seemed that the sync still worked OK.  I saw the same 0x03 errors 
on the FreeBSD system before I rebuilt the kernel to have the PPS_SYNC 
option.  Since then, there's no console on that system, so I might still 
be getting the errors!  [FreeBSD folk - what file should I "cat"?] 
Perhaps these errors are related to driving the PPS software interface 
where one doesn't exist?

My current conclusion is that GPS 4800 baud NMEA syncing is no more 
accurate than Internet syncing - they both suffer from the diurnal thermal 
effect dominating once the poll interval gets up to 1024.  That's one 
reason I'd like to see a simple daily clock drift model added to ntp. 
Even if you have a GPS/PPS on one server, that doesn't do anything for the 
"long poll allows thermal offset" problem on the clients to that server. 
On the FreeBSD system, before the PPS was working, I would have said there 
was more short term jitter from a 4800 baud NMEA sync than from an 
Internet server.  Short-term here meaning tens of minutes.

I haven't looked at what the PPS mods do to the kernel, so I don't know if 
it would be possible in Windows - perhaps at the device driver level.  I 
do suspect that increasing the baud rate could reduce the jitter.  I don't 
know if the Windows implementation includes the "software PPS driver", if 
I can call it that, and therefore if that could be used for experiments.


More information about the questions mailing list