Re: [10GMMF] Notes from Sept 7 meeting on TP3 definition
Yes, EDC indeed makes the spectrum of the interferer an issue.:-)
Lew and Jens gave us a terrific start (aronson_1_0704.pdf), and invited us
to improve on it. In the spirit of what they were trying to do, I believe
the next step should be to replace the sinusoidal interferer with
something better. We have three options here, and the decision is not
easy, but as a starting point of discussion, I will suggest a noise
generator.
First, why sinusoid is not a good idea: In response to a sinusoid, a
typical adaptive filter (MMSE/FFE/DFE) will adjust its coefficients
(impulse response), trying to minimize mean-squared error. Its frequency
response will develop a notch. The severity of the notch depends on
several parameters like number of taps and SNR. This attempt to "notch
out" interference energy concentrated at a single frequency is a very
different behavior from what you would expect in the presence of wideband
noise (bandlimited by preamplifier). Under wideband noise, the changes in
filter's coefficients are smaller and more uniformly spread. Since they
invoke different responses, a sinusoid and noise cannot be considered
equivalent stimuli. Different responses may lead to different
probabilities of passing the compliance test.
And there is a broader issue here. Future EDC implementations will use
more sophisticated methods of exploiting the statistical and spectral
properties of noise, and may have complex EDC structures. A noise
compliance test based on a deterministic, single-frequency stressor --
even if it sort of works today -- may lead to false judgment tomorrow. We
should avoid creating a spec that people may struggle to design towards,
and yet will not guarantee interoperability because it strays too far from
reality. The safest way out is to treat the receiver as a black box and
feed it inputs that are faithful emulations of signal and noise presented
by worst case transmitter and channel.
One possible solution: In subclause 68.6.4 of D0.1, replace the sinusoidal
interferer with a noise generator, 0 to ~10 GHz bandlimited-white,
Gaussian, as a starting point for further work. It doesn't have to be the
proven best option; we only need to be convinced sufficiently that it's
better than all other options at this stage.
Not that it will be easy. Further work should confirm that such a noise
generator emulates RIN and broadband component of modal noise adequately
well. Maybe the narrowband component of modal noise, triggered by
mechanical perturbations and thermal strain, will be factored-in by the
receiver dynamic penalty test, because it will be impossible to separate
it from "changes in fiber impulse response". The broadband component of
modal noise is caused by laser mode partition -- random distribution of
intensities among laser oscillation modes that couple into fiber
propagation modes, showing up as fluctuations in interference pattern
among fiber modes, in the presence of a mode-selective loss element.
(Lasky et al., ISBN 0-12-437160-4. This phenomenon is not to be confused
with MPN Penalty, which is proportional to chromatic dispersion.) Further
work should also determine if the inaccuracy of modeling broadband modal
noise as stationary-Gaussian is within acceptable bounds. And we will need
the help of test equipment experts to ascertain if noise generator
calibration is manageable, if its incremental cost over a sinewave
generator is small, and what new features beyond the plain vanilla
stationary-Gaussian are available. (Phew! Feeling tempted to stay with the
sine wave?;-) ...I am.)
If the noise generator idea does not turn out to be practical, we have two
other alternatives. For discussion's sake, I will provide
counter-arguments to them, hoping others will enlighten me: (1) PRBS: Nice
idea, but there are practical difficulties. An unfiltered PRBS has
bi-modal probability density function. If you smoothen it, it becomes more
Gaussian, but then its bandwidth will be much lower than signal's. And its
auto-correlation function will be different from AWGN's. It falls short of
being a spectrally and statistically faithful emulation of noise. (2) OMA
reduction: Maybe, but... OMA reduction effectively exaggerates noise
internal to receiver. The spirit of TP3 compliance is best served when we
treat the receiver as a black box and feed it inputs that are faithful
emulations of signal and noise presented by worst case transmitter and
channel.
I recognize that this is a complex topic. Perhaps the decision can wait
till the November meeting. I urge the group to commit to making a decision
and to avoid analysis-paralysis.
Vipul Bhatt
Vipul_Bhatt@ieee.org
+1-650-941-6290
-----Original Message-----
From: owner-stds-802-3-10gmmf@IEEE.ORG
[mailto:owner-stds-802-3-10gmmf@IEEE.ORG]On Behalf Of Piers Dawe
Sent: Thursday, September 09, 2004 11:05 AM
To: STDS-802-3-10GMMF@listserv.ieee.org
Subject: [10GMMF] Notes from Sept 7 meeting on TP3 definition
<snip>
re items d and e: the Infineon folks will consider how to calculate the
amount of sinusoidal interferer to emulate the modal noise and RIN. Tom
suggested the use of a PRBS instead of a sine wave, if EDC makes the
spectrum of the interferer an issue - see
http://ieee802.org/3/10GMMFSG/email/msg00256.html .
<snip>