Minutes of 5 May 2004 Channel Model ad hoc group call Attendees (at least...): Joel Goergen Graeme Boyd Mike Altman Jason Roe John D'Ambrosia Joe Caroselli Adam Healy Zhi Wong Roland xxxx Petre Popescu Brian von Herzen Brian Seemann Reference: Graeme's 2-port and 3-port paper Jeff Cain's Crosstalk and Group delay presentation Joel's Agenda and SDD21 proposal Proposed Agenda from Joel: 1. Review 2port physical to 2port differential conversion (Graeme) 2. Review and discuss mask approach for channel definition (Joel) SDD21 MagMask using Four Port data SDD11 Magand SDD22 MagMasks using Four Port data -Group Delay Mask (input from Jeff) NEXT Magand phase for diff aggressors (input from Adam and Jeff) -multi-disturber. (Jeff) FEXT Magand phase for diff aggressors (input from Adam and Jeff) -multi-disturber. (Jeff) **measurements from NEXT and FEXT should be 2port differential with open ends terminated aggressor and victim. 3. Discuss proposal for the interim meeting in Long Beach (Joel) Minutes: Graeme's Conversion of 2 -> 4 port. True 4-port is a 4x4 matrix. What the actual ports refer to in your design is arbitrary. Agilent sez that how you assign the ports depends upon your calibration scheme. Page 5 shows how you set up your 2-port measurements to ultimately derive the 4-port versions. Big issue: Must properly the unconnected terminals. Point: Coupling between differential pairs is almost completely excluded, and could be exacerbated by incorrect terminations. Response: Most the backplanes we deal with do not have tight coupling between differential pairs, so incorrect terminations would not have much effect. NEXT: The port designation is redefined, and you can still use the same Eq. 5. Must terminate properly in this as well. FEXT: Same as NEXT. Q: Do we need to assign ports consistently, so that we can all share channel data? A: No. As long as the data is self-consistent, it will work. Anthony's scripts will accommodate this. Straw Poll: Is there anyone that would not accept Graeme's presentation on 2-port to 4-port differential transformation for NEXT and FEXT? Comment Petri: OK if we can include explanation of the effects of termination on whether the differential pairs are loosely or tightly coupled. Passed unopposed. ACTION: Joel work with Graeme this to present to task force. Q: If the channels are indeed non-linear (e.g. superposition doesn't apply, and 1V drive gives different results than 2V drive), do we have to be careful about the size of the applied signal? A: Agilent says that best measurement uses the highest signal to get the best SNR. You just don't want to break down dielectrics. 0dBm should be OK. Joel: His prime objective was to limit the variance in measurements amongst different parties. 20% of channels do not follow a "linear" model. JohnD'A: It looks like there are problems in detecting the phase properly, especially in channels that have crazy dropouts. Channels may be pushing the limits of the equipment. MikeA: Are we talking about superposition not applying, or equipment issues? ZhiW: Howard Johnson sez that S-parameters are appropriate for linear, time-invariant channels. Group agreed that time-invariant refers to the time span of the measurement, not long term environmental changes. Joel: We have noted the problem, and recognize that some of our channels are either behave strangely, or have measurement issues. We need to move forward, having noted this. ACTION: Joel - Document the measurement techniques from the equipment suppliers. Look at the variants in techniques of 2 and 4 port differential measurement. Work with John, Graeme and Petri and discuss the anomolies between the 2 and 4 port measurements. ACTION: All - measure your backplane channels in 2 and 4 port and submit the results to the group. Channel Models and Masks Joel: Need to have something solid for the May meeting. It can be informative. Propose making 6 INFORMATIVE mag and phase mask sets for SDD21, NEXT and FEXT. Thinks that such a method would cover 80% of the channels that we would deal with. The informative nature allows us to move forward, while the ultimate compliance method may be something like Stat Eye, or something else. People's positions on Joel's proposal: Graeme: Thinks stat eye is too far for this group, given that we have not defined the Rx. JohnD'A: Things coming together for the stat eye, but.... Need to come to agreement - As a first step we need to do this methodology to allow people to have an easy way to look at the problem. Mike Altman: Informative is the way to go. Need to be careful about when we transform to the normative mode. Zhi Wong: Don't disagree, but concerned about how group delay would be generated. Which backplanes are you planning to use to take these measurements? BrianS: There is value to a loose/informative model. Need to be very careful about the tranistion as MikeA pointed out. Problem with separate mag and phase is that there is no 1:1 linkage between the two. But pulse response does have this. Well behaved channels will generally not need this 1:1 linkage. Roland: Agree (with Joel). Adam Healy: For time being it is the fastest way to get the group to move forward. Straw Poll: Is there anyone who disagrees with using informative models for SDD21 mag, SDD11 Mag, SDD22 Mag, NEXT & FEXT mag and phase? Q: Why are we concerned with the phase of the NEXT and FEXT, but not phase of SDD21? Comment: CX4 does not define phase for NEXT and FEXT - not much value. Joel: Withdrew straw poll as written above. Proposes that we have a SDD21 mask, based on a 3rd order equation in order to accommodate more than a single slope. Don't have to accept the graph he sent. Q: Is this meant to cover a 40 inch channel, as well as a shorter channel? A: It is for the worst case channel. We should be able to draw a limit line. We can always come back and make a limitation on the minimum as well (for the other end of the range). Comments: BrianS: no opinion yet. Graeme: no opinion yet. Straw Poll: Does anyone disagree with using SDD21 magnitude as one of the defined mask sets? No disagreement. We can go forward with some kind of SDD21 mask. Joel: Is there anyone that disagrees with using SDD11 as one of the mask sets? Graeme: Partially - most return losses look bumpy. If I violate a mask in a few small places, have I failed? TDR masks utilize error integration. If we do not allow these types of excursions, we can falsely exclude good channels - it is coming up in the package world now. It results in unneeded, expensive packages. BrianS: We got into these masks sets as informative models and now they are incrementally getting more complex and merging towards normative. Troubled that we may be making these overly complex, when we may not even end up using these methods for compliance. Joel: Is there anyone that disagrees with using SDD21 and SDD11 as one of the mask sets as an informative point, understanding that the mask has to be defined well to accommodate anomolies? No disagreement. Joel: Does anyone disagree with using the NEXT and FEXT magnitude mask sets as an informative point. Graeme: How do we add multiple aggressors and account for worst case phase? As long as this is defined, he's happy. AdamH: It is usually a statistical power summing. MikeA: Can move forward as informative. Need to look at whether interferers are looked at as statistical or worst case. Joel: Is there anyone that would disagree with using NEXT mask set, similarly defined as in CX4. No disagreement. Joel: Is there anyone that would disagree with using FEXT mask set, similarly defined as in CX4. No disagreement. Summary: We seems to agree that the following masks are at least needed: SDD21 mag, SDD11 mag, SDD22 mag, NEXT mag, FEXT mag. Phase Joel: Consider using group delay. Petri: Group delay does not tell you much. Group delay ripple provides information that is valuable. Group delay is dphase/df. Group delay ripple is d^2phase/df. Brian: Can see that Petri is right that group delay ripple will cause a smearing of different frequencies. Receivers work in the time domain on time domain signals, so it makes the pulse response attractive. XXX?: Difficult to take 2nd derivatives of actual lab data. Brian: How do we test receivers? We cannot derive a particular time domain signal from these masks to test our receivers. Petri: We always deal with this problem when we define compliance as masks. You have to try a signal and see if it works. ACTION: Joel to work with Petri and JeffC to come up with some ripple constraints. ACTION: All - Forward your opinions on the various magnitudes. Next call: TBD - some time before next meeting. Joel to notify.