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

Re: Open Request to Del Hanson and David Cunningham: Link Model for10 Gig Et...




Comment:

The "Gigabit Ethernet Model" served very well for many members who do not 
deal Fiber Optic Communication issues daily.   The benefit and contribution 
of the Model to committee were well appreciated by all members, which 
provided instant, quantitative answers to various, complex input parameters 
without being an expertise in this field.   Truly, it was a SUPER 
CONTRIBUTION.

The Model is a Cook Book of the GbE optical communication issues.   In the 
process of accomplishing a Cook Book, many assumptions have to be made to 
represent most likely occurrences, but not every possible occurrences.   As a 
result, in the event of a different occurrence from the original assumption, 
the Cook Book will provide an approximate result, but not an accurate result. 
 In most cases, the approximated answer is good enough; however, it is also 
possible that the approximated answer may do disservice to users.

The ISI issue in the GbE is a typical case of the Cook-Book failure. 

The ISI power penalty in the IEEE 802.3z specification is 2.6 dB which is 
over  40% of the useful power 5.83 dB in 1000BASE-SX specification using 62.5 
um fiber.  The ISI calculated based on the Gimlet's  ISI power penalty 
equation and the system BW equation (BER of 10^12), is around  0.5 dB, but 
not 2.6 dB.   Experimentally, it is impossible to have an ISI penalty of 2.6 
dB measured from any 230 meter of 62.5 um fiber using a transmitter with 
negligible ISI.  The actually measured ISI penalty was, again, near 0.5 dB, 
but not 2.6 dB.  The impact of the error is very significant: over 40% of the 
useful power were wasted, and there is no power budget left for distance 
extension. - poor cost-effectiveness. 

When this 40% of ISI wasted power issue was brought up, we were unable to 
determine where the errors came from.    The Model, a Cook Book, did not 
provide the analysis and derivation process to demonstrate how the model was 
made.   A big disadvantage.   Even many of the experts  in the Fiber Optic 
communication field could not determine what  is the problem due to the 
lacking of the traceably of a model. 

The GbE model was established in a very short time period without sufficient 
confirming tests.

I believe it is necessary for a model, or a cook-book, to not only provide 
the spread-sheet, but also the brief analyses which created  the "Model" 
spread sheet for optimization.        

Besides ISI issue, there are many other issues: 

1. Fiber bandwidth.

Using the BW of 160 MHz-km for a 62.5 um fiber in the Model is wrong.   A 
fiber itself does not have a BW in the first place.
  
2. The K factor, which caused lots of headache for us, without knowing how 
the K factor was established in the Model.    

3. A random jitter of a fiber, true?

Recommendation:  
In addition to the spread  sheet, provide the mathematical equations, and 
reasoning in the Appendix.

Regards, 

Ed Chang
NetWorth Technologies, Inc
NetWorthTK@xxxxxxx 
     


 
I would like to request, in behalf of the HSSG, a presentation at the York
meeting regarding the applicability of the "Gigabit Ethernet Model (i.e.,
Hanson; Cunningham spreadsheet)" for use in our work on "ten Gig."

This model was extremely helpful to us in 1000BASE-X in discussing
Variations to the specifications and resolving issues. I expect, as with any
model, that there are certain underlying assumptions which might have to be
corrected (or at least tuned) for application in 10 Gig. At very least,
there will be new techniques needed to use it for multilevel encoding, as an
example.

Some things to consider:
1. Use of truly single mode LW lasers (Vs multiple longitudinal; single
transverse mode)
1.1 Compensated by adjusting K only?
1.2 Chirp?
1.2 Measurement of spectral width (rate of drift of "single mode" Vs
Characterization of spectra)
2. The inherent difference between rise and fall times
3. Assumptions about ISI penalty maximums for a link
4. Simultaneous support for:
4.1 Serial up to 12.5 Gig
4.2 Parallel / WDM down to ?
4.3 Support of various multi-level schemes
4.4 Channel to channel crosstalk issues
4.6 Assumptions for SMF and high performance BW MMF
5. Replacement of the E/R (per FC proposals)
6. The various recommendations made by Petar Pepeligoski (IBM)