Brian,
I feel that
your response here is very
reasonable, but would only make a couple of comments. See below.
-----Original
Message-----
From:
owner-stds-802-3-blade@listserv.ieee.org
[mailto:owner-stds-802-3-blade@listserv.ieee.org] On Behalf Of Brian Seemann
Sent: Friday, October
08, 2004
8:32 PM
To:
STDS-802-3-BLADE@listserv.ieee.org
Subject: [BP] Channel
Inclusion
I was one who,
during the recent Ottawa meeting,
opposed inclusion of certain channels in the "group of 20".
Here are some thoughts on that.
1.
We have "about 20" cases to be in our
official evaluation list. These should be used to carefully stake out
the
edges of the reasonable problem space.
2.
"Relevant" channels are those that
either:
o
a) Fit the project objectives AND meet
the channel
model, or
o
b) Meet the channel model
[DAmbrosia,
John F] As a "channel" person I have
also received directions from the Signaling Ad Hoc, as to some others
that they
wanted to see, namely the margin case.
3.
Minor deviations from the channel model
should be
considered "relevant" and studied.
4.
Major deviations from the channel model
usually
represent implementations inconsistent with the project objectives and
are not
"relevant" or "reasonable", but may be "of
interest". Namely, they may have a characteristic that would be
interesting to study, but it doesn't mean that they fit the target
application.
[DAmbrosia,
John F] I agree with your statement regarding
the 4000-6 as non-relevant, however I did explain that to give the
requested
margin case, I would use the 4000-6 line cards. So this is not an
attempt
to use more lossy line cards to see the lowering of the model, but
rather an
attempt to provide a margin case that falls below the proposed line.
5.
The channel model we have today is
simplistic, but
still has value. Here is why:
o
If all frequency domain plots (SDD21,
SDD11, NEXT,
FEXT) are well behaved, it is a good indicator that the channel will
work.
o
A large number of channels have been
presented that
are well-behaved within the channel model. This demonstrates that
backplanes can very feasibly be built to the requirements.
o
If a channel doesn't work, the frequency
domain
plots are discernably bad.
6.
Some channels in between require more
complex
analysis. This is the false positive / false negative region. This
region may need to allow trading off between the different frequency
domain
parameters, such as the ANR and Voltage Ratio methods.
7.
I'm disappointed that only one connector
is
represented in all the channels offered to the group.
8.
Relative to the above points, here are my
opinions
on certain channels:
o
D'A Case 1: Fits objectives,
minor deviations from model - Include as Relevant.
o
D'A Case 2: Fits objectives,
minor deviations from model - Include as Relevant.
o
D'A Case 3: Doesn't fit
objectives (50% of length is not Improved FR-4), significant deviations
from
model - Don't include.
o
D'A Case 4: Fits objectives,
meets model - Include as Relevant.
o
D'A Case 5: Fits objectives,
meets model - Include as Relevant.
o
D'A Case 6: Fits objectives,
minor deviations from model - Include as Relevant.
o
D'A Case 7: Fits objectives,
meets model - Include as Relevant.
o
Peters Cases: Without going
through each one, I'd apply my criteria above. There were a number of
channels that I feel fit in the "major deviations / inconsistent with
objectives" category.
9.
I'm supportive of using our methodologies
to look
at specific "major deviations / inconsistent with objectives" cases
for the informative purposes for that backplane maker to see whether he
could
use our standardized ICs. But I am unsupportive of perpetuating those
types of design practices by changing the objectives of this project.