> Now, while the internal clock accuracy of a GPS - once it has locked > onto the time of the satellites - must be somewhere in the order of > only a few hundred ns, or better
someone pointed out to me once that 33ns is 10m... The NMEA data does not deal with resolutions below 1s, and the time given in the NMEA message is the time valid at the previous/next (forgot which) one-pulse-per-second edge. If you don't have this 1pps signal, your accuracy will aproach 1s rather quickly. > The extra pulse mentioned by Volker would be nice, but how does it get > from the GPS to the PC? Exactly :) If you want to use the process scheduler of the OS, you won't be any better than the serial NMEA data. You'll have to hook into an interrupt line, e.g. the parport's strobe. You then modify a Linux interrupt service routine to deal with your time keeping, and somehow interface it to the NTP daemon. Any semi-decent setup will have to work in a similar way to this. You should achieve in the 10-100s of microseconds without too much trouble. THere's paper from 2001 which achieves 1us accuracy over NTP and 100M ethernet. You don't want to know how much specialised hardware that requires on both sides... Volker -- Volker Kuhlmann is possibly list0570 with the domain in header http://volker.dnsalias.net/ Please do not CC list postings to me.
