Signaling ad hoc (September 28, 2004)

 

• Meeting called to order.
• Mike reviewed ad-hoc work items. Focus on :
  • Aggressors treatment (e.g. random variable etc.)
  • Channel test case selection (tyco channels, synthetic channel etc.)
• Discussion opened on treatment of channel selection:
  • Ali                    John has brought variable data. But not all possible combinations are perhaps not feasible/realistic.
  • Mike                 Are you proposing that we standardize some form of driver impairments for driver, package etc.
  • Ali                    The problem is that the current models have a lot of reflections which previous models did not have.
  • Mike                 This is a positive direction if we want to consider an end to end simulation. We need some proposal for what to do with respect ot this impairments. Did you have something in mind Ali
  • Ali                    Yes I can come up with some typical s-parameters for package and ESD as a starting point.
  • Richard             Why typical and not worst-case ?
  • Ali                    Because it may not work.
  • Richard             Then there is something wrong with the channel?
  • Ali                    No, if you use everything at worst case nothing will work.
  • Charles             Just to give a direction to this. We are trying to do a signaling direction, no trying to ascertain over what range the system will work.  We just need to have a large enough samples so that we do not have unfair bias toward a signaling method and not another.
  • Joe                   Should the sample be restricted or should we also include some outliers.                   
  • Charles             It was actually your presentation that showed that channels in-spec  behaved worse. But I think it’s reasonable to say that we should exclude the models that do not meet the template.
  • Mike                 However we had a discussion in a previous meeting in which we reached a different conclusion (to evaluate the level of margin).
  • Jeff                   You are right that you should pick cases that are representative of real life but that meet the template. We, as signaling group, should not augment the population by adding other cases.
  • Mike                 The original purpose was not to provide a simulation set that the signaling must pass, but rather provide a bench to evaluate the relative performance. This is way real life channels were requested and provided.
  • Steve                Vendor #2 (which required very little equalization) may provide additional data under the same conditions Tyco did.
  • Mike A              That is fine, we had originally talked about a maximum of twenty models, we are at eight so we are fine.
  • Mike L.              I agree with Jeff that the channel ad-hoc went through a lot of work to set a template and we should restrict ourselves above that line. Joel has gone through considerable effort to build test cards and I do not see his data as part of this suite of test cases.
  • Mike A              So you would exclude channels that would violate the template.
  • Mike L.             Till that channel ad-hoc decide to move the template we should not move it for them. We should increase the suite of test cases. Joel and Intel are potential sources for additional models …
  • Joe                   … although none of the Intel ones fall within the line.
  • Mike A              So we need a specific proposal.
  • Ali                    How do we treat proprietary models …
  • Adam                … these are not to be used to define compliance but only to compare signaling …
  • Ali                    …
  • Adam                … so give us a proposal.
  • Charles             I would like a straw poll to accept John’s 7 models, Steve synthesized models, Joel’s models
  • Mike                 We already did, except for the latter. At the time we explicitly accepted the idea to have some models below the template line.
  • Joel                  I have easily 12 more models
  • John                 I can also provide more models. And these are the test cases you requested. Everyone of those channels was requested and now you want to reverse yourself.
  • Mike                 This is an engineering effort and just because something was requested three calls ago does not mean that they are needed.
  • Ali                    This is the first we saw of these models …
  • Mike  A                 … not true.
  • Brian                 We can certainly ask for a superset.
  • Mike A              How do we get to a concrete proposal. (Joel hands the chair a list of channels, #1,2,3,6,7,8,14,17,18). OK, we have nine channels. Joel, would it be possible to graph them.
  • Joel                  I would need the list back …
• Some vivacious discussion ensues on the subject of the backplane template and on whether we should exclude channels that violate the templates.
  • Mike  A             There is a request for a straw poll to accept, as additional cases, Joel’s models #1,2,3,6,7,8,14,17,18
  • Jeff                   Did one of those have 3 connectors?
  • Joel                  we would have to add channel 19 but I am uncomfortable in including this as it violates the template.
  • Jeff                   I just do not want this to drop of the edge of the table
  • Mike                 Well, if that can be fixed it can be added later.
  • Bill                    The problem with this data is that the data does not fit the goal  of the group (as being connector less and not 40” and not representative of realistic channels)
  • Joe                   So that  means it’s an easier channel.
  • Glenn                Myself, wasn’t to get some kind of model for simulation. However, there were at least three presentations in which it has been hinted that the template will change. So what we may decide  may be in vain.
  • Mike                 Well, back to the purpose (signaling performance evaluation etc. etc.)
  • Straw poll (inclusion of Joel’s models):
    • Yes                  13
    • No                     6
    • Abstain              8
  • Adam                The reason this task force was formed was because there was no way to compare different results. 
  • Justin                What I object to is to weight  the results …
    • Discussion erupts once again.
  • Mike A              There will be channel discussions. But for signaling we need to start somewhere, and this is our charter. If we do not get pass the selection of channel models we will not be able to accomplish our goal.
  • Straw poll (should we only consider for future inclusion models that do not violate all the templates as set by the channel ad hoc):
    • Yes                  22
    • No                    8
    • Abstain             2
  • Mike L  I would like to  propose a straw poll to revise the previous models.
  • Adam    Why should we not let the channel data drive the signaling selection. Later we can decide the weight to assign to the cases that do not meet the template. If we do anything in the signaling ad hoc contingent on the channel ad hoc we will never start and have no data to work with.
• Meeting is adjourned.