[LinuxPPS] still the strangeness

[Bernhard Schiffner]

Am Mittwoch, 20. Dezember 2006 13:40 schrieb Udo van den Heuvel:
> Bernhard Schiffner wrote:

> I would have to check. Currently the status has improved.
> Later: 4 sats. hdop 3.5m.
Ok


> >> /dev/ttyS0, UART: 16550A, Port: 0x03f8, IRQ: 4, Flags: low_latency
> >
> > May be this answer is independent from my remark.
> > (What does low_latency say ? interrupt at the beginning of each received
> > charakter?)
>
> No more fifo. Ans thus the old 8250 behaviour of 1 char per irq.
Ok


> >> Does the NMEA driver this on NMEA data only?
> >
> > !!! YES !!!
>
> Of course, since it is a NMEA driver but it would be interesting to have
> a good method (besides the difference in accuracy) to see if the PPS is
> OK and received or not, ....
cat /proc/pps/...

> ..., from within ntpd.
May be there is an additional flag set in ntpq -c pe. Don't know.

> > Offset your local NMEA-receiver until it's in shape with the others.
> > Forget about a lot of peers in ntp.conf in this state of affairs.
>
> Only the few I posted. ntp.xs4all.nl should be closest but it does not
> have an own GPS receiver.
Once more:
127.127.20.0
ntp.xs4all.nl
!!!ONLY!!!

and now adjust 127.127.20.0's offset so that's reported within +/-3 ms to  
ntp.xs4all.nl.
(fudge 127.127.20.0 offset as it is + or -30 ms)
Then you will have a base to justice later wich slope at DCD is the right one.

> > cat /proc/pps/0*/* must react!
>
> It doesn't.
IT MUST. BY ANY MEANS!
(Connect other sources, use other programs, do what you want!)

> But I do have a feeling that I have PPS now.
It helps but we speak about precision timekeeping, no feelings! ;-)
Call me Thomas, but I want to see /proc/pps reactions!

If you use pps (as provided by GPS), you can reach 1..10µs accuracy.
1..10 ms is for (checked, see above) string receiver.
But before you start with pps, make sure you took the strings to their limit.

> Time offset was much closer to 200ms (one PPS length), jitter is lower.
My receivers:
the "right" slope is _very_ stable, the other is 100+-1ms away. So it's 
important to find and use the right slope.

Bernhard