Re: [10GBT] Request for Cat6 Emissions Characteristics
Sailesh,
As Glenn described, when you corrected spectra.m (which improved PAM12)
you did eliminate your original "EMI PSD" plots.
We all understand that EMI is a complicated problem and hopefully it's
not an issue, but I don't think you should omit results that can be
useful to the group to evaluate. Attached is your revised code with all
four plots included (I just had to copy from your original code) plus
the FCC and CISPR EMI shapes.
Based on your corrected code and the EMI limits you provided, it appears
that the most problematic region is around 200MHz and in this area PAM12
is up to 1dB better,
Jose
PS On the peaky PSD issue, the sync pattern highest peak is 23dB below
the signal PSD. This adds about 0.02dB of ripple. I don't think test
equipment can even measure this. But if you think this ripple is too
high we can always scramble it :)
10*log10(1 + 10^(-23/10)) = 0.0217 dB
-----Original Message-----
From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
Behalf Of sailesh rao
Sent: Friday, July 23, 2004 10:03 PM
To: STDS-802-3-10GBT@listserv.ieee.org
Subject: Re: [10GBT] Request for Cat6 Emissions Characteristics
Glenn,
There are several reasons I stated for why this calculation is
incorrect.
1. In the passband (upto ~0.25fs), where we can do nothing about the
transmit PSD, PAM12 is 0.8dB higher than PAM8. This is the killer
portion of the frequency band as far as emissions are concerned, since
we can filter out the higher frequencies without incurring much penalty
in the receiver performance.
2. You appear to be focusing on a narrow region of the frequency band
above 200MHz. If you are concerned with this narrow frequency band,
please note from the attached that the FCC and CISPR limits increase in
that frequency band. These changes in limits should be taken into
account in any such "peak EMI PSD" calculations that you are doing. The
reason I deleted this portion of my matlab code was because I didn't
have the exact frequencies at which these steps occur, and I didn't want
to make another mistake and provoke another round of highly
disrespectful e-mails on the reflector.
3. Emissions optimizations cannot be done by just doing some 20log10
calculations on average PSDs and selecting the system that shows the
lowest peak. For instance, the PAM12 proposal uses periodic frame start
symbols that have "peaky" PSDs (see, e.g., the PSD "bumps" on slide 26
of tellado_1_0704.pdf). You are guaranteed to get much higher peaks in
the 200-300MHz range for the PAM12 proposal because of the addition of
these PSD bumps. Therefore, just looking at the average PSD is
insufficient for "estimating" emissions.
In any case, the PAM12 EMI penalty is well over 4dB for existing
cabling.
This is a huge hole that the PAM12 proponents are trying to climb out
of, and I don't see how we can go before IEEE 802.3 and claim
1. 10GBASE-T is an extremely hard problem, which requires the 10GBASE-T
PHY solution to operate very close to the Shannon limit.
2. The task force decided to choose PAM12 that has a 4dB EMI penalty
over PAM8, because the task force decided that these 4dBs are not that
important.
(2) contradicts (1) and therefore, don't you think that we will get our
clocks cleaned at the working group level?
Regards,
Sailesh Rao.
srao@phyten.com
That is a completely untenable
>From: Glenn Golden <gdg@zplane.com>
>Reply-To: gdg@zplane.com
>To: STDS-802-3-10GBT@listserv.ieee.org
>Subject: Re: [10GBT] Request for Cat6 Emissions Characteristics
>Date: Fri, 23 Jul 2004 21:24:32 -0600
>
>
>On Thu, 22 Jul 2004 16:07:12 Sanjay Kasturia wrote:
> >
> > This makes me doubt the objectivity of your analysis.
> >
>
>
>On Thu, 22 Jul 2004 16:52:22 "Kardontchik, Jaime" wrote:
> >
> > I would suggest to delete the remark about "objectivity".
> > It is really unwarranted.
> >
> > Everyone is trying to do the best possible technical analysis in
> > real-time.
> >
> > It happens that the ones that dare offer their code for public
> > scrutiny in order to advance the discussions, end up being
> > criticized for the incompletness, inaccuracies, etc, of their code.
>
>
>I'm not sure that Sanjay's remark had to do with Sailesh's program
>having a bug. Everyone understands that honest mistakes of this sort
>will occasionally be made, and that public stonings on the reflector
>are not appropriate. I think Sanjay's remark may have had to do with
>what happened after the error was pointed out and corrected.
>
>In his original posting (based on the computations prior to the bug
>fix) Sailesh showed four plots: Two large scale, and also two zoomed-in
>"EMI PSD" plots. Only the latter two include the 20log(f) radiated
>emission frequency dependence, which, in past postings, Sailesh himself
>has specifically pointed out as being relevant to EMI considerations.
>On those zoomed-in plots, he made a point of noting that the
>differential between the PAM8 and PAM12 peaks was trivial:
>
> >
> > In both cases, the peaks for the so-called "EMI PSD"s are within
> > 0.01dB of each other for PAM8 and PAM12.
> >
>
>But in the subsequent posting, after the bug was fixed, the zoomed-in
>EMI PSD plots with the 20log(f) factor were omitted, and nothing was
>said about the relationship between the PAM8 and PAM12 peaks. This
>omission was justified with the following dismissive comment:
>
> > Besides, in any case, the FCC limits for emissions increases in the
> > neighborhood of 220MHz, thereby rendering Jose's emissions
> > contentions completely irrelevant.
>
>and later
>
> > I consider the point Jose made about the emissions issue to be pure
> > specmanship, and I've already stated that it is completely
irrelevant.
>
>I think Sanjay's response probably had to do with the lack of clear
>technical basis with this claim was made, and with the curious way that
>plots which had been relevant prior to the bug fix became "totally
>irrelevant" after it was corrected, since the actual plots -- had they
>been included in the posting -- do not seem to support that contention.
>
>The plots are attached, along with original (erroneous) plots:
>
> 7525_orig.gif 0.75_0.25D filter, original (erroneous)
results
> 7525_corr.gif " " corrected results
>
> unger_orig.gif Ungerboeck filter, original (erroneous)
results
> unger_corrected.gif " " corrected results.
>
>These were produced using the original and corrected spectra.m programs
>respectively, with changes made only to the axis limits so as to focus
>in on the claims being made about the behavior near the peaks.
>
>In the corrected EMI PSD plots -- the ones that incorporate 20log(f)
>and were omitted from Sailesh's second posting -- even if one
>completely ignores the energy above 216 MHz [I think that's where the
>actual Part
>15 mask breakpoint is] PAM12 has a lower peak value for both filters.
>For the Ungerboeck design, both peaks occur below 216 MHz, and the peak
>differential -- which Sailesh was interested in pointing out earlier
>when his numbers were wrong and showed no penalty for PAM8 -- is now
>0.8 dB in favor of PAM12.
>
>So, as far as the actual data at hand goes, it seems disingenuous to
>use the step in the EMI mask as justification for dismissing Jose's
>claim as "totally irrelevant", when in fact for one of the filters, it
>supports exactly the opposite conclusion.
>
>It may well be true that when the smoke clears on this issue, neither
>system is significantly different as regards peak EMI. But the present
>data does not make that case, and dismissing it with a verbal flourish
>does not help to advance the discussion.
>
>
>Glenn Golden
>Principal Engineer
>Teranetics, Inc.
>ggolden@teranetics.com
>
><< unger_corrected.gif >>
><< unger_orig.gif >>
><< 7525_corr.gif >>
><< 7525_orig.gif >>
_________________________________________________________________
Overwhelmed by debt? Find out how to 'Dig Yourself Out of Debt' from MSN
Money. http://special.msn.com/money/0407debt.armx
EMIPSDSaileshCode.pdf
spectraAllPlots.m