[ntp:questions] Re: drift/offset
david at djwhome.demon.co.uk
Sat Feb 5 10:44:00 UTC 2005
In article <mailman.39.1107502786.583.questions at lists.ntp.isc.org>,
folkert at vanheusden.com (Folkert van Heusden) wrote:
> Ah no, it is a linux system running kernel 2.6.9 on a P-III (1GHz).
OK. What sort of serial interface do you use for the DCF receiver?
You must use one that really interrupts.
If you use something like a Comtrol RocketCard, your times will
be delayed until the next time the driver polls (10ms in that case).
If you use a USB one, I hate to think what consequences there will for
latency, especially if something else is doing a burst transfer.
The 20ms jitter also corresponds to the signalling unit length at
50 baud. As I understand it, the UART trick with raw DCF uses the
leading edge of the signal pulse as a start bit, and idles between
leading edges, so ought to have a sampling error of the UART sampling
clock period, which is generally 16 times the baud rate, or 1.25ms,
in time terms, in this case. I don't think one could produce a reliable
UART with a one to one ratio, so I can't really see how even unusual
hardware would result in a full signalling unit uncertainty.
Incidentally, I think that your speed of light correction will be
dwarfed by the uncertainties in the filter delays in the receiver.
More information about the questions