[802.3af] nanosecond timing over Ethernet++
Relative to tight synchronization between stations,
such as Ethernet speakers in the root.
A couple points of interest:
1) Point-to-point connections have the capability
of synchronizing devices at the nanosecond level,
as long as the out and back links have about the
2) To do (1) might require a distinct physical layer,
which acknowledges the presence of point-to-point
links, rather than a broadcast bus.
3) 802.17 may develop the protocols that implement (2).
In concept, one could migrate 802.17 into the home.
I think that the lack of a physical layer support would
always limit the achievable worst-case accuracy.
Statistical averaging isn't necessarily acceptable in
high-fidelity audio, if the worst-case is orders
of magnitude worse.
I don't particularly favor placing the synchronizing
pulse on the voltage supply, since power distribution
has enough problems already. Some form of 802.3
packet-signaling enhancements (supported by the MAC)
seems more appropriate. Probably useful to the customers,
although accurate network timing and good QOS are
likely to be perceived as threats by IEEE 1394 zealots.
David V. James, PhD
Network Processing Solutions
Data Communications Division
110 Nortech Parkway
San Jose, CA 95134
> -----Original Message-----
> From: owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx
> [mailto:owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx]On Behalf Of
> Kessner, David
> Sent: Tuesday, July 31, 2001 7:48 AM
> To: stds-802-3-pwrviamdi@xxxxxxxx
> Subject: RE: [802.3af] clocking over Ethernet
> > I would also lke a method of achieving synchronisation to a few
> > microseconds via Ethernet.
> Achieving that level of sync over a large network is
> difficult at best.
> I work for Peak Audio (part of Cirrus Logic), and we've
> been using Ethernet for distribution of digital audio
> over an Ethernet based network. The difference between
> us and "the other guy" is that we're doing 64+ channels
> of uncompressed real-time audio. This requires
> synchronizing the audio sample clocks across the entire
> network. Our spec says that we can get 1/4 sample clock
> accuracy at 48 KHz over a 2 km wide Ethernet network.
> We're actually getting better than that (1/10th sample,
> I think).
> Here's a link to some CobraNet (that's what we call it)
> stuff. It's somewhat audio-biased rather than timing
> biased, and not very technical, but it might be at least
> a little informative.
> That being said, I think we're getting off topic here.
> If anyone has questions about this feel free to email
> me directly.
> David Kessner