In message <[EMAIL PROTECTED]> you write:
(...)
>Systematic errors: I assume the gps pps sets out synchronized to UTC,
>suffers cable delays (well, a few ns), hits the DCD pin, goes thorough a
>gate or two, is picked up by the os, then the pps driver finds out about
>it, and it is only then that a (pps time, system time) stamp is made.
For 100 (or so) Z (impedance) cat5 (what I use for my differential pps
signal), figure about 1.5 nanoseconds per foot. With my 50ft cable,
call it 75 nanoseconds. The differential driver adds 20 nanoseconds,
and the receiver adds another 30 nanoseconds. If you then convert
the TTL to true rs-232 levels, an MC1488 driver adds 230 nanoseconds
according to page 8-59 of National's Jan 1974 data book. Total about
350 nanoseconds or so.
>Has anyone any feel for these delays on a modern pc running linux?
An ISA bus 8250 ("COM" port) on a 400Mhz K6-2 gets down to 1-2
microseconds of jitter, according to NTP on my servers.
>Would the error be swallowed up in the inherent noise of the ntp process
>over the internet?
A 10 megabit ethernet (half duplex) gives on the order of 1.5 milli-
seconds delay on a hub here in the lab. A 100 megabit switch (full
duplex) gives on the order of 180-200 microseconds delay. Both of
those figures include whatever delays are involved in the OS IP/UDP
driver.
A very nearby NTP server shows 15 milliseconds delay when configured
on my 1.5M/380k (down/up) ADSL circuit. The jitter of this server
is on the order of 400 microseconds when behaving well.
../Steven
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
