Re: [gpsd-users] Large ntp.conf time1 offset needed with GPSD shared memory driver

[Jon Brase]

On Tue, 09 Aug 2016 16:13:55 -0500, Gary E. Miller <address@hidden> wrote:

> Yo Jon!
>
> On Tue, 09 Aug 2016 15:23:04 -0500
> Jon Brase <address@hidden> wrote:
>
> > In my case I'm using GPSD for coarse time, but not for PPS, because I
> > have not been able to get PPS working with GPSD on the Raspberry Pi
> > (but ntp picks it up just fine).
>
> Many problems with that.  Not the least of which is that ntpd does not
> know when PPS is valid or not.
>
> I have three RasPi's.  A B, a 2 and a 3.  Each with a different PPS/GPS.
> They all work great.
>
> Have you looked at the RasPi HOWTO?

Which RasPi HOWTO are you referring to. Several turn up on Google searches  
and I've seen several mentioned on this list, plus one mentioned by a  
member offlist. Different ones recommend different setups. Anything that  
recommended using 28.0 and 28.1 has come to grief trying to get GPSD to  
pick up PPS on the Pi (I haven't even gotten as far as configuring NTPD  
with that setup). Using 28.0 and 22.0 seems to work best of everything  
I've tried (28.0 + 28.1 fails on getting GPSD to recognize PPS input, 20.0  
alone with flag1 set goes out half a second as soon as PPS skips a beat  
and then comes back on the wrong second when PPS returns, and 28.0 + 22.0  
intermittently ends up on the wrong second if I don't offset it, but works  
fine with a half-second offset).

> > Thus I'm using 28 for coarse time,
> > 22 for PPS.
>
> Which makes it real easy to get the PPS on the wrong second.  Or worse,
> following PPS when the PPS is not valid.

But that's the point of the offset fudge on the coarse data, right? To  
bring it close enough to external time sources that there's enough  
agreement on what second it is that each PPS pulse can be assigned to the  
right second?

> > I'm needing to offset the coarse time by nearly half a
> > second,  and am just wondering if this is normal.
>
> Yes, that is normal.  But that brings it's own problems.  You need to
> set 'prefer' on the PPS and 'noselect' on the NMEA time.  And of course
> at least 3 other chimers.

Yes, pretty much everything recommended other chimers.

And one of the guides I was referencing said that prefer has to be marked  
on something *other* than PPS because PPS doesn't include a second count  
and only indicates top of second.