Hi The easy way is to take a pps off of your external oscillator and feed that into a port on your NTP server. Let NTP tell you where that pps is. Don't let NTP lock to the pps, just let it report it's position.
After that all you need to do is write some code to read the location of the pulse and implement a *long* time constant loop. Taking the 1 ms number and a 1x10^-10 goal, the time constant would need to be around 4 months. There are a few minor details about drift of the local reference and how valid 1 ms is over long time periods. A reasonable GPSDO will likely be much more stable and much more accurate. To get it into the range of being practical, you have to get the NTP setup into single digit microseconds. In the us range you still would not beat the GPSDO, but at least you would have a useful device. 1 us can be done on a LAN with NTP. It's tough to do better than 1 ms over the net. Since PTP suffers the same issues over the net that NTP does, it's not a lot of help in this situation. Bob On Jul 22, 2011, at 5:30 PM, Javier Herrero wrote: > I've found a plot of the ntp-synthesized GPS output compared with the > UTC-aligned GPS from a Thunderbolt. The generated PPS output was 1us wide, > and it is represented in infinite persistence to get an idea of the jitter. > The offset was around 50us, and the jitter around 8us, so not very bad (it > was at least one order of magnitude better than my requirements, so I did not > bother to optimize it further). > > The ntp source was a M12-based ntp server (a blackfin running uClinux, not a > Soekris :) ). > > Driving the PPS output to a serial port from the ntp is not as trivial as you > think. This PPS output is from an "oscillator" disciplined to the system > clock - really a stearable divider from the system clock (it is an embedded > system using uClinux). If you try to drive a digital output directly from the > timer interrupt to get the PPS, you would get far more jitter and error. > > Best regards, > > Javier > > P.S. not sure if the attachment will show up... > > El 22/07/2011 22:19, Chris Albertson escribió: >> This is exactly what an NTP server does. It adjusts the rate of a >> local clock so that the local clock advances at the same rate is the >> set of Internet servers that have passed a clock selection test. NTP >> does this very well considering the uncertainly of the lag over the >> internet. There is a good argument the NTP is optimal but the best >> you can get using Internet time servers is "about a milli second" or >> 0.001 second. >> >> It would be nearly trivial to write software to produce a PPS output >> on one of the control line of a serial port. SO you NTP disciplined >> computer can produce PPS with about .001 second error >> >> Send this PPS to the "normal" GPSDXO in place of the GPS' PPS. The >> computer is "only" about 1000 times worse than a GPS and might work OK >> if you drastically increase the time constant on the GPSDXO's control >> loop. >> >> I think your "NTPDXO might be as good as GPSDXO is measured over a >> long enough period, like months. Short term it might be about as good >> as the TTL can oscillator inside the PC. >> >> On Fri, Jul 22, 2011 at 11:27 AM, Jason Rabel >> <ja...@extremeoverclocking.com> wrote: >>> I was just thinking (dangerous I know)... Has anyone attempted to build a >>> stand-alone oscillator that is disciplined via NTP? >>> >>> i.e. NTP keeps it on-frequency... And I'm not talking about NTP that is >>> locked to a local GPS, I'm curious about purely syncing to >>> other NTP servers over a network. (The presumption is that you have no >>> access to GPS, WWVB, Cellular, or similar.) >>> >>> Is it even possible or am I just day dreaming? >>> >>> >>> >>> >>> >>> _______________________________________________ >>> time-nuts mailing list -- time-nuts@febo.com >>> To unsubscribe, go to >>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts >>> and follow the instructions there. >>> >> >> >> > <PPSAGPS.pdf>_______________________________________________ > time-nuts mailing list -- time-nuts@febo.com > To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts > and follow the instructions there. _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
