Re: [10GMMF] channel models based on measured DMD
Jonathan, I agree with Robert Lingle -- the measured DMD data from
1998-99 should be used to help anchor our LRM estimates. The mode
mixing/mode coupling question you raised is a small effect, and this
needs to established to the group's satisfaction. The analysis using DMD
data did not include the connectors used in the Gen67 work of John Ewen
& others, and these connectors themselves add significant mode mixing.
My suggestion would be compare the results based on DMD measurements to
the Gen67 results with connectors, and to modify the LRM estimates.
As Robert noted in his email, there is an error in his histogram plot
and the OFS data shows a gamma closer to 1.00 as I will describe in the
Corning data.
I will follow up with an upload to Piers with plots & graphics which
hopefully complement the text below.
SUMMARY
Let me emphasize that I agree some mode mixing does occur, I just don't
think it is significant enough to affect the results. A number of
cutback studies have been done at Corning to look into this issue (see
below) and the most recent work concludes that gamma = 1.
I am using a set of about 5000 Corning fibers from 2nd half of 1999 to
estimate the length effects. The median BWs at 2.2,4.4,6.6, and 8.8km
were 870, 910, 880, and 990 MHz.km respectively -- if one takes this at
face value gamma would be ~1.00 over the 2.2-6.6km range and 0.91 going
from 2.2 to 8.8km. There are other factors in play and I would not
conclude that mode coupling is significant.
Starting in mid-1999 Corning began making RML BW measurements -- these
are non-OFL BW measurements with a fixed launch condition. The two ones
used at Corning are the "23um RML launch" (see FOTP 204 Section 3.2.2 on
this RML launch) and a "4um RML" BW which is relevant to LRM estimates.
This is a 1300nm BW measurement using a single-mode fiber to generate
the spot and a 4um offset --- thus it is similar to the DMD launch with
a 4um offset. The median 4um RML BWs at 2.2, 4.4, 6.6, and 8.8km for the
5000 1999 fibers were 1005,1010,1014, and 1080MHz.km, again
corresponding to a "gamma" of ~1.00 in the 2.2-6.6km range and a gamma
of 0.95 going from 2.2 to 8.8. Again I think the higher average BW at
8.8km is due to other effects -- for example, the short lengths arise
disproportionately when there is a problem drawing the fiber or with the
first and last reels of a blank. If one looks at the total distribution
the bottom 25-30% of the 2.2km 4um RML BWs overlays the 8.8km
distribution.
DETAILS
(i) These are measurements on uncabled fibers on large measurement
spools, which minimizes mixing/coupling effects.
(ii) The measurement and analysis does not include connectors. The
99%tile PIE-D levels of 5.1dB (Corning) and 5.2dB (OFS) should be
compared to the Gen67 analysis WITH connectors, which should address
part of the concerns.
(iii) The variation of the modal power distribution as polarization
changes or the fiber position shifts as been discussed in the LRM group;
this supports the idea that experimental measurements should supplement
the modeling to ensure we are not oversimplifying things.
(iv) (slide 6 comments). Since mid-1999 Corning has routinely done RML
launch measurements on MM fiber; the 62.5um FDDI measurement includes an
SMF launch with a 4um offset discussed above.
(v) I mentioned in my Atlanta presentation in responding to a question
that gamma was approximately 1.00. A Corning paper from 2001 LEOS for
reference is http://ieeexplore.ieee.org/xpl/abs_free.jsp?arNumber=969104
Petar mentioned the number 0.85 and Robert's note indicates their data
supports gamma in the range of .8 to 1.0. The Corning LEOS paper
concludes that gamma = 1 should be used. Since the 2.2,4.4,6.6km OFL BW
distributions overlap as discussed above this conclusion is supported
independently by the current data. We also find a similar result for
the 4um offset launch BW.
The Corning BW vs length results will be summarized graphically.
(vi) (Slide 8 comments) I also found a square root dependence on slide 8
as did Jonathan, and was quite confused since nothing like this is seen
at Corning or in modern glass optical fiber to my knowledge. I
eventually asked OFS to check that plot and they have confirmed that the
gamma = .8 to 1.0 number is correct and without the y-axis scale for
slide 8 one can't make use of the bar lengths (i.e. the bottom of the
histogram corresponds to a relatively high BW rather than zero MHz.km).
(vii) Jonathan raises a point about the size of the DMD data sets which
needs to be noted and addressed in more detail. BWs are measured on
each FDDI fiber but not DMDs. However, the DMD data sets are large
enough for our purpose and I think the OFS one is the same data set used
in setting up the Gen67 Monte Carlo set. This also needs a table in a
*.ppt slide for clarity.
Jonathan's point that the entire installed base is larger than
these data sets is true. These 98-99 data sets and the Gen67 Monte
Carlo set hopefully accurately reflect a portion of the installed base.
The earlier fiber with large center perturbations is under-represented,
as well as fiber from other manufacturers not participating in
IEEE/TIA/IEC work. We see center perturbations in current fiber from
some manufacturers.
The argument that gamma ~1.0 is consistent with how the Gen67
Monte Carlo mode delay set was constructed to agree with OFL BW data,
which is simply long length measurement data from the manufacturers.
Jonathan might raise the point, which would be consistent with his
concern about mode mixing, that the mode delays in the Gen67 data set
should be slightly expanded to account for a conversion from 5000m to
300m data, using a gamma of .95 for example (these specific numbers
would scale the mode delays by a factor of about 1.15x). Corning has
the consistent position that the Gen67 data set should be benchmarked
against OFL BW data using a gamma of 1.00, and that the measured DMD
data should be used to fine-tune the LRM estimates.
-----Original Message-----
From: owner-stds-802-3-10gmmf@IEEE.ORG
[mailto:owner-stds-802-3-10gmmf@IEEE.ORG] On Behalf Of Jonathan King
Sent: Friday, April 01, 2005 10:34 PM
To: STDS-802-3-10GMMF@LISTSERV.IEEE.ORG
Subject: Re: [10GMMF] channel models based on measured DMD
Hi Robert
thanks for your e-mail - here's my full and frank commentary on your
slides ' validation of using measured DMD from long fibre spools to
characterize the installed base'.
On your slide 2, you correctly summarize my question/comments as being
directed at the inadvisability of using data from very long (5km) fibres
to predict the PIE-D of short (300m) lengths. My primary concern being
that the effects of significant mode mixing/coupling in long fibres.
Your slide set attempts to justify the use of long fibres, but I would
argue that the results you show actually do just the opposite, and in
fact confirm the necessity of making measurements on short fibre lengths
(ie 300m). Let me try to explain myself.....
Firstly, commenting on your slide 4, temporal resolution requirements
alone do not require long fibre lengths: photo-detectors with
significantly higher bandwidths than 10GHz are easily available (I'd be
pleased to loan you one if that would help).
Secondly, on slide 6 - your comments are all true, but only for OFL
launches. LRM does not use OFL launches.
On Slide 7, I agree that the length normalized OFL Bw would apparently
improve with longer fibres as a result of mode mixing. OFL bandwidths
are (of course) measured with an over filled launch at the input to the
fibre - all modes are excited right from the start; as light propagates
down the fibre, at first little mode scrambling takes place, and the
full range of differential mode delay would be seen; as propagation
continues down the fibre more mode-mixing takes place, the effect of
which is that launched light spends only a fraction of the time in a
particular mode or mode group; for sufficiently long fibres the effect
would manifest itself as a root-length dependence of OFL Bw, which is
nicely shown in your slide 8. Slide 8 also provides confirmation that
mode mixing effects are significant in fibre lengths >1km. I do not
agree with your statement that the root length dependence of OFL Bw
shown in slide 8 leads to a conclusion of long fibre lengths giving an
optimistic PIE-D result, because the LRM application doesn't use an over
filled launch.
In contrast, LRM specifies a choice of 2 input conditions, each of
which excites selective mode-groups (i.e. very different to OFL), and
the choice of which results in the lowest PIE-D. My heuristic
understanding is LRM deliberately avoids exciting mode-groups which
would result in high DMD / high PIE-D. In this situation, mode mixing
effects would cause power to be coupled out of the initially selected
'benign' mode-groups into 'worse' mode groups - consequently, the
normalized DMD gets worse with fibre length, and the benefit of the
precise launch definition is lost. Your slide 8 shows that mode-mixing
effects are significant for fibre lengths>1km (root length dependence of
OFL Bw is maintained down to 1km)- i.e. DMD measurements on fibres
longer than 1km cannot be relevant to LRM.
On Slide 10, 1st bullet: linear scaling down to 300m is not justified
from your data because the mode mixing effects (which you clearly show
to be significant in slides 7 and 8) means you don't have the equivalent
of an OSL mode power distribution for a fibre length >1km - ie you
aren't measuring the DMD of an OSL.
Slide 10, 3rd and 4th bullets: I disagree with your statement that
linear scaling underestimates DMD for the 300m case: you haven't
measured DMD for an OSL, because mode-mixing destroys the OSL launch
MPD; If it didn't, you wouldn't see root length OFL Bw dependence
A comment on your last slide: I believe your fibre measurements to be a
small subset of those used to steer the development of the GEN67 Monte
Carlo model, and a vanishingly small sample of the installed base; as
such I think it would be wrong to place a greater significance on them
than I would the results of GEN67.
In conclusion, in my opinion the 99% PIE-D figures in balemarthy-1-0105
are actually quite seriously misleading, since they are derived from
measurements on long fibre lengths which actually conceal the benefits
of the precise launch conditions defined in the LRM draft.
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: Friday, April 01, 2005 6:43 AM
To: STDS-802-3-10GMMF@LISTSERV.IEEE.ORG
Subject: [10GMMF] channel models based on measured DMD
All,
Several questions were raised in Vancouver about possible issues with
using
measured fiber DMD data to model the installed based as presented in
balemarthy_1_0105. I had to miss Wednesday in Atlanta, but I also
understand that similar questions were raised during John Abbott's
presentation.
During the Atlanta meeting, Piers uploaded a presentation addressing
questions raised by Nick Weiner and Jon King in Vancouver on this
subject
at:
http://grouper.ieee.org/groups/802/3/aq/public/upload/Validate1998OFSfib
erse
t.pdf
I would appreciate comments and feedback on that.
Robert
Robert Lingle, Jr, Manager
Fiber Design and Transmission Simulation
OFS R&D, Atlanta, GA