Re: [10GMMF] Corrected Monte Carlo Model as LRM standard model fo r OM1
Jonathon,
Thanks for your thoughtful comments.
My responses corresponds to your numbering:
1. John Abbott and I believe that a Monte Carlo model should be used
validate any spreadsheet approach to 10GbE, not vice versa. The spreadsheet
models assume that "bandwidth" means a Gaussian pulse, and the ISI
statistics are calculated based on that. This is overly simplistic for 10Gb
transmission; the standard for 10 GbE work was set by 802.3ae where the
Monte Carlo delay set approach was applied. The Monte Carlo set has
non-Gaussian pulses and a log normal distribution to the BW.
2. The OFL-BW measurement scheme used for grading MM fiber has long been
based on gamma-scaling from 6-8km to 300m fibers.
3. I agree that effect of mode-mixing on single mode launch is complicated,
and I have pages of notes detailing different cases. That is why John Abbott
and I got away from trying to sort out that complex problem in favor of
focusing on a problem we know how to treat quantitatively: the impact of
mode-mixing on OFL-BW data that was used to construct the MC delay set,
using a universally accepted expression and following extensive precendent
within the industry for length-scaling of OFL values. (Unfortunately a
careful experimental study of DMD variation with length would be very
expensive in both fiber and time, given the complication of axial
variation.)
4. The manufacturing margin is consumed by the need to account for 1)
gamma-scaling when cutting from 8km to 300m length as well as 2) sensitivity
of MM fiber to axial variation. Fibers cut to 300m can be expected to have
OFL-BW all the way down to 500MHz-km. Regardless, a 50MHz-km shift in
minimum OFL-BW (a large value) would only change a given PIE-D %tile by
0.05dB.
Robert
Robert Lingle, Jr, Manager
Fiber Design and Transmission Simulation
OFS R&D, Atlanta, GA
-----Original Message-----
From: Jonathan King [mailto:jking@BIGBEARNETWORKS.COM]
Sent: Thursday, July 07, 2005 4:13 PM
To: STDS-802-3-10GMMF@listserv.ieee.org
Subject: Re: [10GMMF] Corrected Monte Carlo Model as LRM standard model
for OM1
Hi Robert
thanks for your e-mail, and the summary of the sensitivity analysis to
OFL Bw cutoff
I personally still have a few objections to the revision of GEN67YY you
are proposing. I have two main objections and a couple of perhaps less
significant ones, in order of significance:
1) Its not been 'baselined' against other applications: GEN67YY has been
shown to be largely consistent with preceding applications which have
been successfully proven in the field (Ewen_1_1104.pdf). I could not
support transitioning to a model which has not undergone similar
scrutiny.
2) Gamma factor scaling from 6-8km fibre length measurements has not
been validated with short fibre (300m) measurements. I've raised this
issues a few times, but to date it has not been addressed; I apologize
for repeating myself.
the ones I'm less sure of are
3) I am not convinced that gamma correction is as straight forward for
OSL launches as with OFL:
My thinking is this: there are three mode mixing regimes -
zero mode coupling, which corresponds best to the <300m LRM application,
weak mode coupling (where significant power couples into initially
unexcited modes) which corresponds to several km or so of fibre , and
strong mode mixing (significant power couples back and forth between
modes); a fibre would be expected to transition from zero mode mixing at
short fibre lengths, through weak mode mixing at intermediate lengths,
to strong mode mixing at long lengths.
Normalized OFL bandwidth would be expected to increase monatonically
with fibre length, because all modes are excited and any degree of mode
mixing would tend to average out propagation delay differences between
modes.
However, for launches exciting just a few modes (eg OSL), weak mode
mixing couples power into initially unexcited modes, resulting in an
increase in measured propagation delay spread, and a decrease in
normalized bandwidth with length - the opposite of effect on OFL.
(One symptom of weak mode mixing would be an emerging correlation
between measured bandwidths or PIE-D for OSL launches with different
offset values - I believe this was visible in the OFS 1998 fibre set
Lingle_1_0305.pdf)
So I think that whereas the correction for mode mixing effects for OFL
may be relatively straightforward (albeit requiring an educated guess
which requires validation with measurements), the necessary length
correction for mode mixing effects on an OSL would be difficult to
predict, and may require the opposite of that required for OFL.
4) The effect of OFL bandwidth manufacturing margin should probably be
included in any model revision - the installed base model in GEN67YY is
sharply truncated to 500MHz.km. In practice, there would be very few
fibres actually at 500MHz.km, because standard manufacturing margining
practices would dictate a test criteria at a higher level than the spec
limit to guarantee compliance of any individual measured fibre. The
effect on the ensemble of fibres is a shift to a higher cutoff OFL Bw.
This effect would be similar in magnitude to the sensitivity analysis of
the lower OFL Bw cutoff, which you summarize below, and like the
wavelength issue would probably result in a 0.1-0.2dB correction in
PIE-D values, depending on the details.
best wishes
Jonathan
tel: 1 408 524 5110
e-mail: jking@bigbearnetworks.com
fax: 1 408 739 0568
Jonathan King
Director, Optical Systems
BigBear Networks
345 Potrero Avenue
Sunnyvale, CA 94085
-----Original Message-----
From: owner-stds-802-3-10gmmf@ieee.org
[mailto:owner-stds-802-3-10gmmf@ieee.org] On Behalf Of Lingle, Jr,
Robert (Robert)
Sent: Saturday, July 02, 2005 2:33 PM
To: STDS-802-3-10GMMF@listserv.ieee.org
Subject: [10GMMF] Corrected Monte Carlo Model as LRM standard model for
OM1
All,
At the London meeting, several of us made a presentation on the Monte
Carlo
set corrected for mode mixing (the gamma correction). I showed duplex
coverage PIE-D results for the recommended correction of gamma = 0.9,
and
recommended that this model be used henceforth for LRM coverage
calculations. See
http://grouper.ieee.org/groups/802/3/aq/public/upload/Implicationsofdupl
exco
veragewithcorrectedMonteCarlomodelforTP3stressors.pdf
At the meeting Jon Abbott and I received useful feedback from Jon King
that
sensitivity analysis of the choice of lowest OFL-BW value admitted into
coverage calculations was needed to assess the robustness of the fiber
model. Fiber modeling for LRM typically uses 500MHz-km as the lower
cutoff
for OM1 fiber in all calculations to date. Kasyapa Balemarthy (GaTech)
has
done this analysis, and it increases the credibility of using the Monte
Carlo model to assess OM1 coverage for LRM.
Kasyapa varied the LOWER limit of OFL-BW fibers retained in the MC model
and
found that the 99%tile duplex PIE-D point changes linearly by 0.05dB for
each 50MHz-km change between 400 and 600 MHz-km. This relative
insensitivity arises from the fact that a wide range of both low and
high
PIE-D fibers exist in a selected range of OFL-BW, and the distribution
of
PIE-D values within a range of OFL values varies smoothly as OFL rises
from
400 to 600MHz-km.
This is very good news and means that the MC model is robust to the
choice
of this parameter, i.e. coverage results obtained are rather insensitive
to
picking 500 MHz-km vs. 475 or 525 MHz-km.
I would like to assess whether there are any remaining technical
objections
to using the Monte Carlo model at 1355nm with a gamma=0.90 correction as
the
standard model for LRM coverage calculation.
Robert
Robert Lingle, Jr, Manager
Fiber Design and Transmission Simulation
OFS R&D, Atlanta, GA