| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Sanjay, this is exactly why some group discussion on the modeling methodology is required. I was heartened to see a call for discussion on the PHY parameters in your presentation on putting together the draft.
You’ve raised an interesting point that parameters such as “margin” which seem simple, are really not. In fact, what margin is required for a system may be dependent on how the system breaks down when the SNR is violated. (but I digress)
If one is putting their estimates together from a detailed hardware design, the complexity and size can be substantially different when you require 2-3 dB better ANEXT performance, because you have to properly scale the ANEXT models, and these don’t simply work linearly. Likewise, areas of hardware complexity, like echo cancellation may be substantially different in a workable design that will perform over ALL varied & valid combinations of patch-cord and cable lengths from those that are simulated from our current modeled configurations (remember, connection points can be substantially away from the end). Such effects as connector mismatch, far-end-echo, multipath, will have to be dealt with in the real system.
If we want to go down this approach of comparison, we first need a benchmark for determining the technical maturity of the estimate. That may mean significantly more models generated by the cabling ad hoc, and would result in delay of our work, but a more accurate result at the end.
The other approach is to simplify our model generation, yet end up with less accurate, but faster, results. Either way, comparison would be on equal footing.
Recollection from the 1000BASE-T estimates is that they were far off from the complexity of the early devices. (I recall estimates like 250k gates being thrown about). I would not want to be that far off the mark with 10GBASE-T. It will make the difference between feasibility and not in a 3-year (2 Moore’s law cycle) timeframe.
As a way forward, but knowing that the agenda is full, perhaps Brad can set aside 2 hours (maybe work through lunch one day?) to discuss this as an entire group? I wouldn’t want to do it in an ad-hoc or after hours session, because the results of the discussion will effect the whole group’s decision process.
-george
-----Original Message-----
Looking through the proposaldetail spreadsheets I have received, I have seen some numbers that may may indicate discrepancies in interpretation and am sending this out as a clarification. This may or may not be relevant to you and does not imply an error in what you have submitted.
Since the most vocal debate on channel models has centered around ANEXT modeling, the entries called for in row 11 aim to understand sensitivity of your proposal to the ANEXT coefficient. This is not the same thing as the gap between SNR required and SNR available at the "Slicer input" of the DFE. To get the number requested, adjust the ANEXT coefficient so that the SNR available equals the SNR required.
I think some have submitted the gap between the SNR required and SNR available which is the more traditional interpretation of margin. It may make sense to include this number also.
Regards,
Sanjay Kasturia
Editor-in-chief, 802.3an
cell (650) 704-7686 office (408) 653-2235 fax (408) 844-8187
Teranetics Inc. 2953 Bunker Hill Lane, Suite 204 Santa Clara, CA 95054
|