[LinuxPPS] Why is my ATOM clock a falseticker?

[Cirilo Bernardo]

On 2/6/09, Paul Simons <paul at thesimonet.org> wrote:
> And here are the results.  It is good that I'm not losing seconds; it is bad
> that something is still wrong.
>
> # watch -n1 cat /sys/class/pps/pps1/{clear,assert}
>
> Every 1.0s: cat /sys/class/pps/pps1/clear /sys/class/pps/pps1/assert
>                                  Thu Feb
> 5 15:57:39 2009
>
> 1233878259.496184165#347453
>                    Seconds pass Midnight = 57459
> 1233878258.996184378#347453
>
> Every 1.0s: cat /sys/class/pps/pps1/clear /sys/class/pps/pps1/assert
>                                  Fri Feb
> 6 13:13:16 2009
>
> 1233954795.497913974#423989
>                Seconds pass Midnight = 47596
> 1233954795.997913264#423990
>
>
>     423990
>  - 347453
>      -------
>       76537
>
>     86400
>  - 57459
>     ------
>     28941   =  Seconds in the rest of the previous day
>  + 47596   =  Seconds in the current day
>     -------
>     76537
>

That shows that there is no problem with PPS or your electronic interface.

I have looked at the NTPD driver (refclock_true.c) and the driver
performs some voodoo which I don't understand (some ad-hoc atmospheric
propagation delay).  However, if this reference driver is tied to the
ATOM driver (which assumes a precise interval reference, not
necessarily synchronised to TAI/UTC/GPS) it will calculate an 'offset'
for the ATOM signals and eventually tie into it.  All I can suggest at
the moment are:

1. fiddle with the option settings of the driver; see the driver
documentation.  This may be sufficient to discipline the TrueTime
output.
2. check that the ATOM driver is correctly patched to use LinuxPPS

The ideal behavior here (based on how the drivers work) is for the
TRUE driver to calculate an 'offset' for the ATOM driver; that offset
needs to be used (either in the ATOM driver or LinuxPPS - it doesn't
matter).  If the calculated offset is ignored, then the PPS will
become a 'false ticker' (because the offset can be as much as 0.5
seconds).

- Cirilo