In article ,
Phil Kane wrote:
Something must have changed (or been fixed) then - we made
measurements about three years ago and there was about six seconds
offset - an eternity for accurate time measurements. 340 nanoseconds
we can tolerate. Six seconds we can't.

It's changed. GPS and UTC now differ by 14 seconds, according to
http://tycho.usno.navy.mil/gpstt.html. This is because GPS time does
not include leap seconds.

This 14 second difference is part of the GPS broadcast, so can easily
be backed out of the GPS time data to produce UTC. Once corrected,
the UTC values have the stated accuracy.

Don't be confused by the latency of some GPS units in producing time/fix
products. I've seen them produce fixes several seconds later. That's why
the time is included in postition data, so you know when you were there.
If you want time from your GPS, you need either the 1PPS pulse output or
a unit with a known and predictable period from real time to character
output. For many uses, simply assuming that the first character of the
output string (NMEA) occurs at the time in the message is adequate,
but that's not going to get you your 340ns accuracy.

For example, I am using a Trimble Acutime to feed an home-brew time
demon. Tests comparing system time from this demon to ntp stratum 1
servers gave a few millisecond difference. Good enough for me.