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

Re: [10GBT] Channel models

Several PHY vendors have asked for 4x4 channel data to evaluate PHY
proposals with more accurate modeling data.

There are several methods to scale matrix channel data to the set of
limit lines. After discussions with a few cable experts, I sent the
email below to most PHY vendors. The results of this "unofficial PHY
straw poll" were

-6 said case 2 was fine
-1 said case 2 was fine, but would like the max to not include the lower
100MHz or so since there could be some artifacts there
-1 said case 2 was fine, but case 1&2 (min of case 1 and case 2) could
be better
-1 said case 2 was fine, an added that 1000BT used the equivalent of
Case 3, which is pessimistic.

To summarize, case 2 uses PS, but had one scaling for all the 4x4 matrix
data, based on the worst PS measured from all 4 receivers.

Below is the detailed email

Jose Tellado


Sanjay requested data for all 4 transceivers in a email several weeks
ago. A couple of cable vendors have volunteered this data, but there are
questions regarding the details of scaling, specifically for 4x4 FEXT
and NEXT data since 1000Base-T simulations probably did not include data
for all 4 receivers. The main issue regards Power Sum vs Pair to Pair
limit line scaling of measured data. If you scale NEXT or FEXT data to
meet one of the two limits, it's very likely it will not meet the other.
The difference in the resulting scaled data is small (typically <1dB)
and since both NEXT and FEXT will require significant PHY cancellation
the residual error effects on the receiver will be very minor. I don't
have a preference, but I have been talking to a few cable vendors and
they don't seem to have a consensus answer. I was tasked to poll PHY
vendors to check if you have any preference. So far I've got responses
from several and Case 2 below seems to be the preferred choice.

Let's consider NEXT Xtalk, represented as:

Next = [ N/A    N12    N13    N14;
         N21    N/A    N23    N24;
         N31    N32    N/A    N34;
         N41    N42    N43    N/A]


PP is pair to pair limit for NEXT
PS is power sum limit for NEXT
PS1 = N12+N13+N14, PS2 = N21+N23+N24, PS3 = N31+N32+N34, PS4 =

Here is a list of scaling possibilities (not exhaustive, but probably
representative of the options):

Case 1. Find K1 = min_ij{ PP-Nij } and scale all Nij by K1. This finds
the worst case NEXT coupling relative to PP and scales all responses by
the same amount K1.
Pros: Xtalk data is symmetric
Cons: Only one xtalk pair is at PP limit. Might violate PS limit

Case 2. Find K2 = min(PS-PSj), j=1,2,3,4 and scale all Nij by K2. This
finds the worst power sum next for all 4 receivers and scales all
responses by the same amount K2
Pros: Xtalk data is symmetric.
Cons: Only one rx victim pair is at PS limit. Might violate PP limits

Case 3.
Compute D1=PS-PS1. Scale N12, N13, N14 by D1
Compute D2=PS-PS2. Scale N21, N23, N24 by D2
Compute D3=PS-PS3. Scale N31, N32, N34 by D3
Compute D4=PS-PS4. Scale N41, N42, N43 by D4
Pros: All rx have NEXT at PS level
Cons: Xtalk data is not symmetric. Might violate PP limit.

Case 4.
Compute D1 = PP - N12. Scale N12 by D1
Compute D2 = PP - N13. Scale N13 by D2
repeat for other 10 pair2pair responses.

Pros: All pair2pair Xtalk is at PP limit
Cons: Xtalk data is not symmetric. Might violate PS limit.

Due to reciprocity I would assume the cases that preserve the cable
symmetry (Case 1 or Case 2) are more realistic. If I had to pick one, I
would probably get case 2, since it has some "averaging" effect in the
scaling from adding 3 Xtalk responses to choose the scaling factor and
it keeps the symmetry

Please tell me if you have any other preference. Otherwise, I will
assume that there is no big preference and case 2 could work,



From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
Behalf Of Sanjay Kasturia
Sent: Wednesday, May 12, 2004 4:11 PM
Subject: Re: [10GBT] Channel models


My requests were driven by requests I had received for:
a) 4x4 or full matrix data so you could actually simulate all the
crossterms and hence simulate a system operating over 4 pairs
b) a few other sample measurements so that you get a better
representation of the cable plant

I have subsequently seen questions on how you scale actual cable
measurements to the limit lines - the questions arise specifically when
you are dealing with a matrix channel. Do you scale each parameter to
the limit line? Do you scale the sum etc.

I think Jose Tellado had summarized a few different ways of scaling. If
you don't have a summary of what he had summarized, please ask him for
it and put it on the reflector.


Sanjay Kasturia
Editor-in-chief, 802.3an
cell (650) 704-7686
office (408) 653-2235


From: stds-802-3-10gbt@IEEE.ORG [mailto:stds-802-3-10gbt@IEEE.ORG] On
Behalf Of CDimi80749@AOL.COM
Sent: Wednesday, May 12, 2004 12:47 PM
Subject: Re: [10GBT] Channel models

Sanjay Kasturia had posted a request for additional cabling channel
models in an e-mail sent to the 10GBASE-T reflector

4/10/2004 (attached below). I had responded to Sanjay that the TF
cabling Adhoc had contributed a set of worst case

models to the TF modeling Adhoc group and requested that Sanjay be more
specific on the need for and types

of additional models required

I'd like to continue this thread in order solicit input from the TF on
the need for additional cabling models in order

to coordinate and focus the cabling adhoc on the deliverables. The TF
established a baseline cabling specification

supported by the current set of models. What are the requirements and
objectives for additional cabling models?

We need to focus on the development of the standard and generate a list
of objectives and requirements for

additional channel models that apply to that effort. Towards that end,
what are we lacking in channel models??


Chris DiMinico

Chair: 10GBASE-T Cabling Ad-hoc

In a message dated 4/10/2004 1:19:36 PM Eastern Standard Time,
SKasturia@TERANETICS.COM writes:

        We have four channel models defined from the March 2004 plenary

        For detailed phy performance simulations, we will need sample
channel data that complies with these models.

        We will greatly appreciate it if any task force participants can
provide full matrix (4x4) channel data. This would ideally include:
        a) Insertion loss data on each of four pairs - real and
        b) Near end crosstalk coupling (6 unique pair-combinations) -
real and imaginary
        c) Far end crosstalk coupling (6 unique pair-combinations) -
real and imaginary
        d) Four unique echo transfer functions (one for each pair) -
real and imaginary

        This can be obtained by measurements on actual cable and then
scaling to meet the specified channel models.
        Any cable companies volunteering to provide this data?
        There is some data from the cabling Ad-hoc available on:
        This data has some of the items listed above but not all. It
does, however, provide enough to get some sense of performance.

        Sanjay Kasturia
        cell (650) 704-7686
        office (408) 653-2235