Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [10GBT] Discussions on precoder decisions

Sailesh -
What you're seeing is a response from the PAM-12 group to fill the holes
in the proposed constellation from the last meeting, which would serve
to mitigate some of the minor technical points that you have raised over
the past few months.

Before you come off with a quick - "why don't you just save the effort
and converge around PAM-8", speaking for myself, I can say that this
work is being done primarily because we have looked at the issues, and
determined that the extra analog bandwidth, higher clock rates, and
increased complexity of signal processing in the cancellation are to be
avoided, especially when confronted with the reduction in noise-limited
performance that comes from higher baud rates.  We have looked carefully
at PAM-8 and PAM-12-based designs and found that there are no major
relaxations in specifications that offset the higher bandwidths,
increased signal processing, higher-rate sampling clocks, and lower
intrinsic performance.  An honest effort has been made, and, as a
result, some observations from the PAM-8 solution are being adopted into
the lower-baud rate systems to make them better.

I believe that at this point in the disucssion, there is general
agreement that the performance at the two baud rates is similar, with
one having better ANEXT noise margin and the other having increased
constellation point spacing.  Most PHY vendors that I have spoken with
make the value judgement that the improved noise margin and reduced
complexity at the lower baud rate is better than the increased
constellation point spacing at the higher baud rate.

What we have here is an attempt to make progress benefiting from
discussions with you.  By filling the hole in the proposed constellation
to make a more efficient constellation, and thus to get the benefit of a
lower baud rate, well matched to the optimum for noise performance, and
making the job of 10GBASE-T easier.  In our judgement, unless you have
some OTHER advantage, the complexity and drawbacks of going to 952 Mbaud
just aren't justified by the benefits.

-george z.

-----Original Message-----
From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
Behalf Of sailesh rao
Sent: Wednesday, September 15, 2004 11:07 AM
Subject: Re: [10GBT] Discussions on precoder decisions


Is the "Double Square" constellation the same as the "Checkerboard"
constellation you were telling me about last week? If not, this would be
5th constellation I've heard about from the PAM12 camp.

Here's the list so far:

1. Original PAM-12 "donut" constellation.
2. PAM-12 Compapiano-Glazer "cross" constellation.
3. PAM-12 multi-dimensional "Kabob" constellation.
4. Checkerboard constellation.
5. PAM8/16 Double Square constellation.

A general observation is that instead of researching PAM12
the task force would be well served if we can converge around the PAM-8
proposal and conduct a working session on the standards draft at the


>From: Gottfried Ungerboeck <g.ungerboeck@BLUEWIN.CH>
>Reply-To: "IEEE P802.3an" <STDS-802-3-10GBT@LISTSERV.IEEE.ORG>
>Subject: Re: [10GBT] Discussions on precoder decisions
>Date: Wed, 15 Sep 2004 10:56:34 +0200
>the 128DSQ constellation ("double square") is the natural go-in-between
>PAM and 16 PAM, much more so than 12PAM. Contrary to 12PAM, all
>decoding, and precoding operations are based on powers-of-two
>128DSQ has been known for a long time in the literature and its good
>with precoding has been well described.
>On precoding, please see ungerboeck_1_0704.pdf (Portland). One fixed
>IIR-type precoder is enough. If the group cannot agree on a single
>precoder, then let it be a very small set (2?) of fixed precoders, one
>selected preferrably during auto negotiation.
>Regards, Gottfried
>   _____
>From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
>Of George Eisler
>Sent: Mittwoch, 15. September 2004 06:03
>Subject: Re: [10GBT] Discussions on precoder decisions
>Thanks again for taking the badly needed lead on urging along the
>making. The precoder definition is clearly in need of consideration,
>although I think it will take simulation work to really nail it down.
>course you can't finish it if you don't get started.
>On last week's call, I was surprised by the "double square"
>with it's sort of 16/8 PAM flavor. It seems to open the case up rather
>converging, but I wonder what's your take on it? I was going to call
>only to realize that I don't have your phone contact.
>George Eisler
>310 459-9225

Express yourself instantly with MSN Messenger! Download today - it's