I completely agree with Jack's assertion that we don't have much
time. We should move this discussion along crisply.
We should seriously consider the potential of using a Serial PHY for
at least 100 meters of installed MMF, preferably 300 meters. In
principle, there are three ways to attempt it - (A) Enhance the
effective modal bandwidth of the fiber, or (B) Use equalization
alone, or (C) use a combination of the two.
(A) alone will not work. No matter how clever a launch technique we
try, the effective bandwidth of fiber will fall well short of 1
GHz-Km, even at 1310 nm. We need equalization; question is how much?
(B) and (C) yield interesting possibilities, depending on the amount
of equalization we agree to attempt. (B) eliminates the need for
mode launch techniques, while (C) eases the burden on equalization.
By equalization, I mean to include whatever DSP necessary to
overcome DMD dynamically.
If we want to try (C):
1. A 1310 nm Serial PMD will take care of P802.3ae Objectives 1 and
2 simultaneously. To achieve 8 GHz on 300 meters of 500 MHz-Km
fiber, we need an equalization gain of 15.6 dB. Is this feasible?
If yes, we may have something truly exciting to talk about. A
Silicon IC based solution, that supports Objectives 1 and 2 with a
1310 Serial PHY - which is already endorsed because it meets
Objectives 3 and 4? This is too good to be true... so where is the
2. An 850 nm Serial PMD will support Objective 1 with 5 to 6 dB of
equalization gain. If an equalization gain of around 16 dB is
feasible, however, it can support 300 meters objective too. The 850
Serial has the additional benefit of being lower cost.
If we want to try (B):
One alternative is to apply the 16 dB gain to 850 nm, unconditioned
launch, 100 meters installed MMF. This will meet Objective 1.
Another alternative is to stay with 1310 nm, unconditioned launch.
This will meet both Objectives 1 and 2 if higher equalization gain -
about 22 dB - is feasible.
Instead of jumping all over the decision tree, let's first nail this
down: How many dB of equalization gain is the maximum we should
attempt, taking into account cost-effectiveness, risk and schedule?
[Footnote: To all the equalization experts on this reflector, I have
to ask - where have you been? :-) It takes time to build consensus;
P802.3ae members may find it overwhelming to adopt a new idea at a
late stage. I urge you to present feasibility of this solution at
the next meeting, and to start a technical dialog on this reflector,
quickly, showing proof of feasibility. Speak up now, or forever hold
your silence...the bride is under pressure to choose the groom!]