Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [10GMMF] TP3 3-impulse test proposal

I agree that we should strive for simplicity in test methodology.  But this should not be at the expense of rigor.  

I do not doubt that you have observed the pulse shapes you state.  But it is not at all obvious that selecting specific surrogates, such as those consisting of three impulses, is sufficiently rigorous.  If all the test is intended to do is probe how quickly an EDC device can track changes in waveform, then something even simpler could be used.  However, if the purpose is to see how the device behaves not only in regard to its adjustment speed but also to its ability to follow waveform shape variations, then waveform variety must be included.   The possible exception is to show that, independent of EDC architecture and design trade-offs, a specific class of waveforms represents the most rigorous case.  I don't believe that anyone has demonstrated that yet.  

Paul Kolesar
1300 East Lookout Drive
Richardson, TX 75082
Phone:  972.792.3155
Fax:      972.792.3111

"Joerg-R. Kropp" <Joerg.Kropp@INFINEON.COM>
Sent by: owner-stds-802-3-10gmmf@IEEE.ORG

09/22/2004 03:42 AM
Please respond to
"IEEE P802.3aq 10GBASE-LRM" <stds-802-3-10gmmf@IEEE.ORG>

Re: [10GMMF] TP3 3-impulse test proposal

I agree to your statement. Our observations in the lab (with standard
offset launch patchcords as well as precise controlled offset lauch by
micro manipulator) show , that the pulses appear in the form Gaussian
like for many cases. But we also observe  superimposed 2- or 3-impulse
like shapes with pre- and/or post-cursors.
Therefore it seems adequate to me use a 3-impulse model. Keep it simple
as possible.

-----Original Message-----
From: owner-stds-802-3-10gmmf@IEEE.ORG
[mailto:owner-stds-802-3-10gmmf@IEEE.ORG] On Behalf Of Abbott, John S Dr
Sent: Dienstag, 21. September 2004 15:21
Subject: Re: [10GMMF] TP3 3-impulse test proposal

Ben thanks for the clear summary and thanks to all for clearly stated

[9] When one considers the offset launch at 17-23um, the resulting mode
power distribution looks more Gaussian than a "3 impulse" example. There
are always more pulses.  The fibers which give a relatively low offset
BW and which represent a test of EDC have an "alpha error" in that
region, so that there is a sequence of pulses at roughly equal intervals
with a Gaussian-like envelope.
       Since EDC seems to be more stressed by Gaussian transfer
functions why are we using a 3-impulse model rather than a 5-impulse
model (for example)?  Is the 3-impulse worst case or is it primarily
easy to implement (answer [4])?

-----Original Message-----
From: Ben Willcocks [mailto:ben.willcocks@PHYWORKS-IC.COM]
Sent: Tuesday, September 21, 2004 8:32 AM
Subject: Re: [10GMMF] TP3 3-impulse test proposal

Thank you for the various comments and questions relating to our TP3
3-impulse test proposal.  I will endeavour to address them all here:

Paul Kolesar asked:

1] > What is the purpose of this dynamic test?

Our aim was to produce a test which would verify that the equaliser can
track a changing channel response, as suggested by Lew Aronson.

2] > Does its purpose explain why the pre and post cursors never exceed
the amplitude of the central pulse?

Our intention was to provide stress commensurate with the Cambridge
fibres throughout the test.

3] > Has anyone analyzed other test cases with more or less than three

Yes, Petre Popescu has done some excellent work on this topic recently.

4] > What has led to the choice of three?

Ease of implementation.  We simply took Lew Aronson's 3-impulse
proposal, and we believe we have verified that it can produce an
appropriate level of stress, measured with the PIE metrics.

Sudeep Bhoja commented:

5] > I would suggest that we use Tx rise times consistent with TP2
specifications for computing the link dispersion penalty from the fiber
models. Hence, using 30ps for the Tx rise time under estimates the
dispersion penalty. For reference, the -LR rise time is 47.1ps for the
20-80% rise time of the Tx.

We chose the 30ps rise time filter for consistency with Petre Popescu's
work.  As we are making a comparison between PIE metrics for the
3-impulse test and PIE metrics for the Cambridge fibers (using the same
filters in both cases), the exact rise time of the tx filter has very
little effect on the result.

Changing from a 30ps rise time to a 47.1ps rise time, the increases in
the PIEs for the 80th centile Cambridge fiber at 220m are matched (to
within 0.1dBo) by similar increases in the PIEs for the 3-impulse tests.

Tom Lindsay asked:

6] > Please clarify for me - is this work focused only on a dynamic
test, or does it also suggest a static test? Would the dynamic test be
separate or combined with a static stress test?

Lew Aronson has proposed separating the static and dynamic tests, and we
were focusing on the dynamic test.  It may be advantageous to define
something simpler to implement for the static test, for example a fixed
low-pass filter.

Lew Aronson commented:

7] > We should do this calculation (and all similar calculations) for
300m as well as 220m.

I will add 300m figures to the slides.

8] > I am concerned about deriving the details of the impulse response
parameters solely from PIE metrics... I would suggest that, at least in
the case of the dT choice, that the fiber models and Petre's work be
used to determine the approximate choice, and that we live with a larger
difference in PIE-L and PIE-D that results.

This is a good point.  However, we felt that it would be desirable to
make the dynamic test straightforward to implement, with the idea that
one could rely on the static test to verify that equaliser
implementations have adequate time span.