RE: [10GBT-Modeling] RE: [10GBT-Cabling] [10GBASE-T] a channel capacityestimation program for your evaluation
George,
From the past few days' discussion, I finally understand the meaning of
"scale to fit" used in the channel models. Compared to the limit, the
scaled, selected channel model will give a few dB better SNR. Once the
channel models are available, they can be directly plugged into the program
and get the desired results. I believe there will be a difference.
I also agree with you that modeling ANEXT will be a quite challenging job.
Since it is the primary noise source in the system now. We have to give it
a clear shoot before we can move forward.
Best regards,
Xiaopeng
Marvell Semiconductor
"George Zimmerman" <gzimmerman@solarflare.com>@majordomo.ieee.org on
02/24/2003 11:35:14 AM
Sent by: owner-stds-802-3-10gbt@majordomo.ieee.org
To: <stds-802-3-10GBT-Cabling@ieee.org>,
<stds-802-3-10GBT-Modeling@ieee.org>, <stds-802-3-10GBT@ieee.org>
cc:
Subject: RE: [10GBT-Modeling] RE: [10GBT-Cabling] [10GBASE-T] a channel
capacity estimation program for your evaluation
All ?
(you might want to note that there is a message length limit on the
reflector ? hence much of what went before is truncated.)
It is worth re-iterating that the limit lines provided in TIA & ISO specs
are defined, not as models, but in fact, as limiting specifications,
against which real cables are evaluated at each frequency, over a specified
bandwidth. This means a number of things, some of which are pointed out:
1) That real cables will generally touch the limit line at a small
number of frequencies (often a single point) over the specified bandwidth
for the measurements. (Xiaopeng ? this is the PRIMARY reason that capacity
with real models is significantly greater than your limit-line models. You
may continue to believe otherwise, but the numbers are clear, and there is
a valid physical reason for it ? namely the noise is lower everywhere that
the real transfer function doesn't touch the limit line)
2) That, as a result, in general, real cables will be better than the
measurement, since manufacturers design with margin, and qualification
specifications are designed so that installers have margin as well.
3) That extension of these lines beyond the specified frequency range
has limited value, and representative measurements would be preferred (real
cables can be better or can be worse than an extrapolation ? we've seen
both, depending on the characteristic)
4) That any scaling to a specification probably out to respect the
frequency range over which the spec is defined.
It has been our experience that, in fact, the limit lines are most often
most closely approached in the low frequency bands (under 10 MHz) ? this
actually detracts from any attempts at extrapolation to high frequencies.
For existing cabling, no doubt we will ask for some form of qualification
measurement. I do not expect that it will require strict adherence to an
extrapolated limit line, and would oppose such a qualification as
unnecessarily restrictive. Meanwhile, a simple total power measurement
doesn't quite provide the right level of input either. I suspect that some
coarser, relaxed frequency averaging, perhaps as a ratio of insertion loss
to impairment (e.g., NEXT) level will be the solution. This is work for
the Task Force, and for communications with the cabling standards bodies.
For informational purposes, the material presented for capacity in the
November tutorial were scaled by adjusting the levels of the insertion
loss, NEXT, and ELFEXT specifications were scaled to touch the limit lines
defined for the category of cable (Cat5E) within the 1 MHz to 100 MHz
region of the frequency band. The exception to this scaling was Alien
NEXT, as discussed at the time of the tutorial, the variance between
observed results, and the rather loose limit line recently put forward is
not a small amount for margin, but a rather large amount to account for the
uncertainty. Estimates on achievable cancellation can only be put forward
in conjunction with an observed model, hence it is also unreasonable to
apply the 8 dB number we used to a situation where the Alien NEXT is
higher. I would conjecture that it is likely that effects resulting in
higher alien NEXT would increase the pair-to-pair correlation, thereby
increasing the potential for cancellation.
We feel strongly that simply scaling the within-sheath NEXT response is
inadequate for alien NEXT measurements, because we have observed both a
different frequency shape in the high frequency bands), and because, if you
take the NEXT response at -3 dB relative to the within-sheath response, and
assume that that case, in fact, is valid, you will rapidly come to such
odd, and contradictory conclusions that NEXT cancellers are unnecessary for
1000BASE-T, or that 1000BASE-T will not work due to Alien NEXT either, both
of which are known falsehoods. I know that folks in the cabling community
are working hard on alien NEXT modeling, measurement methodology & new
specifications. These will undoubtedly come out of the work from TIA and
other groups in the near future. Meanwhile the work in IEEE can move
forward to provide them with meaningful targets to shoot for.
George Zimmerman
gzimmerman@solarflare.com
tel: (949) 581-6830 ext. 2500
cell: (310) 920-3860
-----Original Message-----
From: Albert Vareljian [mailto:albertv@ieee.org]
Sent: Monday, February 24, 2003 1:10 PM
To: William Jones; Chris DiMinico
Cc: xichen@marvell.com; stds-802-3-10GBT-Cabling@ieee.org;
stds-802-3-10GBT-Modeling@ieee.org; stds-802-3-10GBT@ieee.org; Sterling
Vaden
Subject: Re: [10GBT-Modeling] RE: [10GBT-Cabling] [10GBASE-T] a channel
capacity estimation program for your evaluation
Bill, Chris
In this regard, as the frequency band of interest expands considerably,
wouldn't it be beneficial to introduce an integral limit for the NEXT,
FEXT and ANEXT power (say max curve RMS in a given freq. band [500 MHz]).
This would be complimentary to the existing practice of the limit envelop
line.
The integral metric would be easier to make both strict and at the same
time
much less conservative than the envelop. The RMS spec would not be allowed
to
exceed, whereas failing to meet the envelop at the specific frequency
point(s)
(say beyond the conventional band) could still be well tolerated in the
system,
and therefore could be made discretionary, subject to application.
(Integral impairment metric for NEXT, FEXT and ANEXT would also be
consistent
with other UTP systems, such as EFM MC and SC, etc)
Regards,
Albert
William Jones wrote:
Albert
Thanks for making this crystal clear with your example. I believe this
brings to the forefront the distinction between limit lines and models.
Specifically, limit lines must be extremely pessimistic since the cabling
folks must necessarily insure, with extremely high probability, the
deployed cable will not exceed these limits at any point in frequency.
Further these line must be smooth due to the unpredictable structure of the
xtalk. The effect is even more pronounced in the case of alien xtalk.
See, for example, the tutorial where the peak alien xtalk measurements are
almost 16 dB below the limit line.
regards
Bill
---- (remainder of preceding conversation truncated) ---