Again we are talking effectively what happens. There is a good deal of logic what to do if the NMEA and/or pps does not come in and if the system clock is significantly different from either the NMEA or pps. A whole day on the lab bench with a scope on both the pps and the NMEA these never happen. The pps comes in and by definition it must be represent the second + 1 of the previous NMEA string. This is exactly the same logic that gpsd uses and is why we tell NTP that the NMEA has an offset.
We scope the oscillator also and it is a very expensive oceanographic device that adjusts for temperature automatically. Sounds like a common problem and I note when we add an additional network clock the problem seems to go away. I note there are multiple SHM segments in NTP. Anyone tried writing the pps time into two locations to fix this problem? -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of unruh Sent: Thursday, July 14, 2011 5:50 PM To: [email protected] Subject: Re: [ntp:questions] Single GPS/PPS time source gets marked as a falseticker On 2011-07-14, Michael Eder <[email protected]> wrote: > Not using gpsd, just writing the NMEA time and receive time into SHM > (0) like gpsd does. The pps does the same to SHM (1). Effectively > the pps code sounds like a bad procedure. You want to make sure that the nmea actually gets associated with the second marker that the interupt marks. Your nmea time will be off by 600ms from the true time, and could well confuse ntpd. > just increments the second from the NMEA string and writes it to SHM. > We need certain values from the NMEA string so have not looked into > anything but the ASCII strings. PPS comes in on a high priority > interrupt so it gets serviced very quickly. Actually the PPS > conditions an oscillator that we So when the pps comes in, you can read the system clock and you know how many usec the clock is out from true time-- you just do not know how many seconds. Then you can read the system time when the nmea comes in and determine how many seconds the system time is out from the true time (but not how many usec.) But combining the two, you can zero in on how many seconds and usec your system clock is out. interrupt-> read system time nnnn.uuuuuu nmea ---- rrrr at system time mmmm.vvvvvv d=(mmmm.vvvvvv-nnnn.uuuuuu)*1000000)/1000000 (d should be 0, since it should be within a second of interrupt) Then in the shm ntp packet, the sent and received times are nnnn.uuuuuu, and the remote received and sent are (rrrr+d).000000 where I am assuming that the time between the nmea packet and the pps interrupt is under a second-- ie the nmea sentence designates the time of the interrupt that is less than one second in the past. One could always throw the item away if d is not equal to 0. If, like in the Garmin 18x, the time of the nmea is close to a second late, then you could always put in a shift in defining d. d=int(mmmm.vvvvvv-nnnn.uuuuuu)*1000000.-500000.)/1000000 Ie, you want whatever writes into the shm to associate the nmea time with the pps time, and since (except for unusual circumstances) the system clock is assumed to keep reasonably good time (eg be out by not more than say 10000 PPM) you can use it as the flywheel to enable you to do that association. > use to keep time in the event we lose the GPS (out in the ocean on a buoy). > The tests I am running are in the lab so we are always getting the > NMEA and PPS. > > > > -----Original Message----- > From: [email protected] > [mailto:[email protected]] On Behalf Of > Rob > Sent: Thursday, July 14, 2011 4:13 AM > To: [email protected] > Subject: Re: Single GPS/PPS time source gets marked as a falseticker > > Michael Eder <[email protected]> wrote: >> We have looked at our GPS on a scope, the PPS it is dead on and the >> NMEA (just one sentence) is also reliable with about a 680 ms latency >> and 10 ms jitter. > > Again, are you using gpsd? > > If so, you may want to try removing the (huge) 680ms offset inside > gpsd instead of in ntpd. > There is a place in the code where a fixed offset is added to time > obtained using NMEA (because the gpsd author does not want > configurable items) and it cannot be correct for every possible receiver. > > Again, it is better to switch from NMEA to the native binary protocol > of the receiver. > > _______________________________________________ > questions mailing list > [email protected] > http://lists.ntp.org/listinfo/questions _______________________________________________ questions mailing list [email protected] http://lists.ntp.org/listinfo/questions _______________________________________________ questions mailing list [email protected] http://lists.ntp.org/listinfo/questions
