Signaling ad hoc (October 22, 2004)

Mike Altman begins his presentation (http://ieee802.org/3/ap/public/signal_adhoc/altmann_s1_1004.pdf)

Slide #2

[Joe A.] What are the simulation conditions?
[Mike A.] They have been implicitly included in the ‘link elements definition’

Slide #3

[Joe A.] In one case you say that the methodology should not be simulator-dependent. Yet both methodologies presented so far rely on a specific simulator. Where do we stand with this ?
[Mike A.] This was a discussion we had in the past. So far the direction has been to be simulator independent.
[John d’A.] There never was a straw poll.
[Mike A.] Right, but it never appeared to be a subject of disagreement.
[Charles M.] It’s not clear that different simulators would yield same outputs.
[Mike A.] Let’s table this discussion for now. If you feel that a single simulator is ‘a must’ I must ask you to make a formal presentation on the subject.
[Joe. A.] Fair enough. The only reason I am bringing this up is because the discussed methodologies were certainly based on the concept of a common simulator.

Slide #9

[Tom] Would it be simpler to just measure the eye opening?
[Mike A.] I am not thinking so much as a comparison of implementations but rather as an attempt to define a “penalty box” to be used as a common derating factor.
[Jeff S.] I agree that these effects should be included so as to capture the “shape” of the eye.
[Petri] This voltage margin number seems related to the BER calculation. Is it not redundant?
[Joe A.] I agree.
[Mike A.] Well, that is true if I include random effects. I could consider just ISI for the purpose of determining these margins.
[Richard M.] Are you concerned about the asymmetry of the eye?
[Petri] The point I am making is that this voltage/timing margin should be a subset of the BER calculation.
[Adam H.] A fairly common method has been bathtub curves, The voltage and timing margins are just a different way to report the shape of the same function.
[Xiao-ming] If you add noise that would be better.
[Mike A.] This is for us to agree upon.
[Adam H.] I think that this group at a bare minimum should provide guidance on;
[Petri] In the past different signaling schemes were selected by looking at the channel capacity.
[Adam H.] 10Gbt was the only one case where this criteria was used.
[Petri] CX4 did not use this criteria because there was not a debate on signaling.
[Steve A.] If you start from channel capacity this may lead to a complex solution.
[Vivek] I was involved in the 10Gbt. Capacity defines the best we can achieve. The next step is comparison of the margins associated with each signaling scheme. This still does not shed a light on the implementation complexity. If we agree on the channel model, this should be straightforward.
[Mike A.] Capacity would be a concern only if there was a doubt a bout the capacity of the backplane. We are past that hurdle. What I am hearing is that focus should be on the complexity of the implementation.
[Vivek] A shortcut would be to compute the SNR for the different schemes.
[Mike A.] But you would still need to report BER, voltage and timing margins.
[Vivek] Voltage margin is what I am talking about. Timing margin is trickier.
[Mike A.] My reasoning for specifying the BER is that this is what I ultimately care about.
[Brian] I am very skeptical about using SNR. BER is the metric to consider. SNR has an implicit assumption of Gaussian statistics that fall apart at high BER.
[Mike A.] So you see the use of Gaussian statistics as incorrect?
[Brian] Yes.
[Joe A.] I have a question on power which I consider a critical metric. The concern I have is that power is a competitive metric. Assuming we go beyond confidentiality issues, it will be extremely difficult to do an apple to apple comparison.
[Vivex] In 10GbaseT most people had agreed on the basic architecture and building blocks (transmitter, equalizer …). Once the blocks were agreed upon (independently of signaling) it became possible to compare them in terms of taps, speed etc. So we built a matrix with all this information. The various participants reported numbers for textbook implementation approaches.
[Joe A.] This seems reasonable.
[Mike A.] The concept of a ‘vanilla architecture’ is interesting and is part of what I was considering in the presentation I made in Portland. Two questions: is this agreeable? Is there an example we can refer to?
[Vivek] Yes.
[Mike A.] Would you be able to make an attempt at defining it?
[Vivek] Yes. I will circulate this among the participants.

STRAW POLL #1: Should we establish a detailed power and complexity reporting matrix?

Yes: 27 No: 0 Abstained: 0

STRAW POLL #2: (a) Should we establish BER as a signaling quality metric?

Yes: 19 No: 2 Abstained: 0

(b) Should we establish a min BER target?

Yes: 7 No: 9 Abstained: 5

STRAW POLL #3: Should we require reporting voltage and timing margin at the BER levels of 10^-12, 10^-15 and 10^-18?

Yes: 18 No: 0 Abstained: 3

Follow-up meeting scheduled for Friday 29 October 2004. Bridge and mtg info will be distributed..

Attendance:

Joe	Abler
Andrew	Adamiecki
Michael	Altmann
Stephen	Anderson
Majid	Barazande-Pour
Brian	Brunn
Joe	Caroselli
John	D'Ambrosia
Adam	Healey
Mike	Lerer
Cathy	Liu
Richard	Mellitz
Charles	Moore
Tom	Palkert
William	Peters
Petre	Popescu
Shannon	Sawyer
Jeffrey	Sinsky
Fulvio	Spagna
John	Stonick
Dmitri	Taich
Vivek	Telang
Brian	Von Herzen
Nirmal	Warke
Chris	Wittensoldner