| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Will, Yes it makes sense to use auto-negotiation
to select PHY port type and higher-level system configurations such as pause
and EEE capability and then use link training mechanisms to adjust PMA level
coding once a particular PHY type has been selected. Arthur. From: William Bliss OK
Arthur, Yes,
letting Precoder choice be part of link training can work, and is
probably a good idea here, as it might be more amenable to ‘late
changes’, at which time the RX can have learned more about the particular
link and how to optimize it. Note
that its difficult to say a priori that the Precoder on is always the best
choice for each channel and each particular RX architecture. The idea is
to give each RX implementation team as much leeway as possible. On a
related topic, it seems any FEC choices (assuming we don’t have a fixed
always on FEC) could fall into the same category (decided during training
vs in AN, again giving the RX more time to decide what is
‘best’). Thanks, will From: Arthur Marris Will,
Seeing as this is specific to a particular PHY type and might be asymmetric
would it not be better dealt with by link training similar to how a 10GBASE-KR
receiver sets a transmit equalizer as defined in Clause 72? Also if the
precoder is useful and has miniscule cost why not just make it mandatory rather
than an option to be negotiated? Arthur. Hi
Arthur, Please
consider adding to AN; -
Precoder. The
RX specifies its preference for either having a 1/(1+D)mod(M) precoder or not
(bypass). This allows RX implementers to choose what is best for them,
either fixed or an a link by link basis. The cost to the TX is
miniscule. This proposal also applies to an NRZ only solution. Thanks, Will
bliss P.S.
looks like I have a problem posting to LISTSERV for now, but wanted to get this
to your attention now |