This file explains the background for 
  DDJ100.txt
  DDJ400.txt 
  ISI100.txt
  ISI400.txt
  EMB100.txt
  EMB400.txt
  PIED100.txt
  PIED400.txt
  Rand100.txt
  Rand400.txt

These files sample the updated 850 nm OM2 5000 fiber model, provided by Corning, and the 2000 TIA launches. Only fibers that pass the OFL bandwidth requirements and launches that satisfy the 802.3ae encircled flux launch requirements are included.

See 05-878v1 for more information.
 
To explain the files, I began with Eric Borisch's results (thanks Eric!). Recall that he took all 1e7 combinations of fibers and launches and analyzed DDJ (psec), ISI (dBo), and effective modal bandwidth (EMB, MHz-km). For each of these 3 cases, I first removed non-compliant fibers and launches, then sorted the remaining data (~7e6 combinations remained) by their respective result (DDJ, ISI, EMB). Then, for each sort, I linearly downsampled into two levels - approximately 100 combinations, and approximately 400 combinations. The downsamplings are uniform/linear.  That is, the files with ~100 values sample at ~1% per step along the CDF, and the files with ~400 values sample at ~0.25% per step along the CDF. The best-case value for each is at ~1% or 0.25% respectively, and the worst-case value is at 100% for each option.
 
I also created a group of 50e3 PIE-D results. These were created with random (uniform probability) combinations of fiber and launch. This group was also sorted and the CDF similarly downsampled into groups of 100 and 400.

Each file/array has 3 columns; each row contains the sort value (ISI, etc., for information only), the fiber index (out of 5000), and the launch index (out of 2000). The files should be obvious by their names.
 
The Randx00 files are based on the same 50e3 set used for PIE-D, except that they are randomly sampled with no sorting or selection criteria. The PIE-D values are given in their first columns.
 
I am providing these in hopes that they are convenient groups of selected fiber and launch combinations for which to run full SFP+ system simulations.
 
Again, see 05-878v1. With enough samples, it should make little difference in the end, since each group samples the entire population. I expect that whatever is used for full system simulation, the results will not monotonically track the input cases - however, they can simply be resorted.
 
I'd appreciate any comments or questions.
 
Thanks, Tom
Tom Lindsay
ClariPhy Communications
tom.lindsay@clariphy.com
phone: (425) 608-0209 or (949) 480-9210
cell: (206) 790-3240
fax: (425) 608-0232