[time-nuts] 10MHz LTE-Lite - PPS accuracy?
hmurray at megapathdsl.net
Wed Dec 10 15:37:44 EST 2014
david-taylor at blueyonder.co.uk said:
> But having the PPS on the DCD over USB is not as useless as you might first
> think, because in tests here using the DCD/PPS over USB produced better
> results with NTP than an internet connection alone. It is worth checking -
> your results may differ.
The Ethernet is also on USB so it will have the USB jitter as well as any
jitter from the network. Even if the remote NTP system is perfect (or at
least very good relative to the R-PI), I'd expect a local PPS via USB to be
slightly better than internet time.
tic-toc at bodosom.net said:
> I'll admit, to my shame, that I have yet to deduce how to use USB provided
> DCD for PPS. I've looked, really I have but to no avail.
Linux has two APIs to PPS.
gpsd uses TIOCMIWAIT, an ioctl that lets a userland program wait for the
PPS/DCD change. You can feed that to ntpd via SHM.
The ATOM and NMEA drivers in ntpd use the API described in RFC 2783. It's
in sys/timepps.h On Fedora, it comes from the ps-tools-devel package. This
needs a running ldattach 18 /dev/xxx for each PPS source. The interrupt
driver grabs a timestamp so the timing accuracy should avoid most of the
jitter associated with getting to userland.
I think most real serial ports have support for both. Support for USB serial
devices is not so good. I haven't checked recently. I think TIOCMIWAIT
support is generally better.
These are my opinions. I hate spam.
More information about the time-nuts