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

Re: [10GBT] Updated Tables


Thanks for the opportunity.

At the meeting, Jose pointed out that the extension of the SNR/BER curve I
made on the (2048,1723) code from 6E-11 BER down to 1E-12 BER (slide 10 of
rao_1_0704.pdf) was not based on real simulations. I had assumed that the
1E-12 BER assertions on slide 4 of seki_1_0304.pdf was based on real
simulations, but it was apparently just a "hand extrapolation" of my
original SNR/BER curve.  Unfortunately, this point was not clarified to me
during my e-mail correspondence with Dr. Seki.

At the moment, I stand by the 6E-11 BER point in the BER vs. SNR curve of
the (2048,1723) LDPC code since I personally supervised the generation of
this simulation point. In the worst-case, assuming Murphy's law applies, we
can expect the BER/SNR curve for the (2048,1723) LDPC code to undertake a
slope change in parallel with the uncoded Gaussian curve and the intercept
for the 1E-12 BER will occur slightly higher than 19.9dB (my extrapolation
shows 20.4dB, not 20.9dB).

However, this is a moot point. If Amir Mezer's LDPC simulations show that
the (2048,1723) LDPC code has a slope change at 6e-11 BER, I will instantly
recommend adding a row to the Djurdjevic construction for the (2048,1723)
code and pushing the slope change down below the 1E-12 BER point. Or, for
that matter, if Amir Mezer's simulations show that the (992,829) LDPC code
has no slope change until 1E-12 BER, I will recommend we switch to that
code. I am aware that Amir Mezer has close to infinite computing resources
since he is employed at Intel (Amir, I hope things haven't changed!!!), and
therefore, we can rely on him to validate the specific LDPC code the task
force is using down to 1E-12 BER and beyond.

The bottom line is that the PAM8 modulation scheme is not wedded to a
specific Djurdjevic LDPC code and we should really dis-associate the
specific LDPC code choice from the choice of PAM levels. For instance, the
PAM8 proposal will work just fine with the (1024,833) code the PAM12
proponents are using, though I would personally hesitate to add such a
SONET-on-Steroids like framing complexity in an Ethernet standard. Heck, we
had interoperablilty problems with the scrambler definition on 1000BASE-T -
do we really want to deal with interoperability issues with such a complex
framing scheme in 10GBASE-T?

Sailesh Rao.

>From: Vivek Telang <vivek@VITESSE.COM>
>Reply-To: "IEEE P802.3an" <>
>Subject: Re: [10GBT] Updated Tables
>Date: Mon, 19 Jul 2004 16:17:39 -0500
>Hi Sailesh,
>At the meeting last week, there was a question raised (by Jose) about the
>PAM-8 SNR requirement of 19.9 dB for BER=1e-12 (column 3 of your table). I
>think he said that it was actually ~1dB worse than that (~20.9dB), but I'm
>not positive. There was a discussion, but I don't think the matter got
>resolved, so can you clarify?
>-----Original Message-----
>From: []On
>Behalf Of sailesh rao
>Sent: Monday, July 19, 2004 2:02 PM
>Subject: [10GBT] Updated Tables
>In the attached, I've updated the 3 tables in our July presentation based
>the following:
>1. Change PAM12 symbol rate to 825Ms/s from 820Ms/s.
>2. Delete PAM10 entry.
>3. As Luc pointed out, add a 1.2dB emissions penalty for PAM12 due to its
>higher transmit PSD.
>4. As Jose pointed out, subtract 0.4dB from the PAM12 emissions penalty due
>to the THP peak voltage adjustment.
>Next, I integrated the WGN for 1E-12 BER over the Nyquist frequency range
>get a "wideband noise tolerance" measure for the two proposals. Finally, I
>summed the noise immunity penalty and the emissions penalty for the PAM12
>proposal to form a "Total EMI Penalty" metric over the PAM8 approach.
>In Models 1 and 3, the penalty works out to be 2.6dB and 2.2dB respectively
>for PAM12 over PAM8. However, in Model 2, which represents the existing
>cabling infrastructure, the penalty for PAM12 over PAM8 works out to a
>whopping 4.0dB!!
>Sailesh Rao.
>MSN Toolbar provides one-click access to Hotmail from any Web page - FREE

Donít just search. Find. Check out the new MSN Search!