[time-nuts] Timing on Ethernet

Magnus Danielson magnus at rubidium.dyndns.org
Mon Aug 6 05:49:49 EDT 2007


From: Dr Bruce Griffiths <bruce.griffiths at xtra.co.nz>
Subject: Re: [time-nuts] Timing on Ethernet
Date: Mon, 06 Aug 2007 15:04:48 +1200
Message-ID: <46B68FD0.8080903 at xtra.co.nz>

> Hej Magnus

Hej Bruce,

> One could always make use of the 2 spare pairs on a 100Mbit/s connection.
> Its not that much more difficult to build a custom interface using these 
> than one doing precision timing on the receiver outputs.

Indeed. The trouble is that cross-talk from the transmitter side will be your
enemy. It will smooth out over time, as the transmitted data is DC balanced,
but there is a small symbol dependent component. By listening to the
transmitted data and choosing a "wise" point in that data you are operating
synchronously with that data. You can use the spare pairs for queueing or for
a separate site-channel with measurements.

Considering that delay in cables varies very slowly, the two one-way measures
does not have to be taken exactly at the same time. Time-stamping the events,
transfering the data over to the other side and then take the difference
between the time-stamps and divide by two you have the time-difference between
the two nodes clocks (which was part of the time-stamping) and using that it is
easy to steer the slaving clock since this is the delay-compensated time-error-
detector in action. It is a trivial control-loop to keep the slaving node on
time.

Cheers,
Magnus



More information about the time-nuts mailing list