09 July 04 IEEE 802.3ap Adhoc Conference Call Meeting Minutes ------------------------------------------------- Attendees: john.dambrosia@tycoelectronics.com michael.w.altmann@intel.com joel@force10networks.com popescu@quaketech.com jjlynch@us.ibm.com mike@mike-lerer.com qing-lun.chen@intel.com abler@us.ibm.com rmoubarak@ti.com charles_moore@agilent.com brian.seemann@xilinx.com shannon_sawyer@agilent.com ahealey@agere.com BrianVon@FPGA.com glen@vitesse.com David.McCallum@molex.com why@pmc-sierra.com schelto.vandoorn@intel.com cathyl@lsil.com gerald.merits@hp.com william.r.peters@intel.com Bob Noseworthy Pravin Joel Goergen - ad-hoc channel chair Brain Seemann recording the minutes Minutes: 2 presentations sent out: Agenda VNA set ups A third was later sent while the call was inprocess from John D and Bob N on crosstalk Agenda: Informative Channel Model needs to be voted on next week. Adjustments to freq range Max Freq: 12.5 or 15GHz? Comments: Discussion on 3rd harmonic - is it a problem for short channels? Some wanted to do 15GHz to get all the data to make an informaed decision. Others said this was shown previoiusly to not be a problem. Short reach channels will have useful signals out to 15GHZ, but they will be comfortably above the specification. Concern that the higher harmonics come into play in lengths less than 1m. Suggestion to use a 60dB floor. Nice to have the initial frequency be a multiple of the step size - simplifies DFT. Finding notches is not dependent on step size, but more a function of IF BW. If we have data to 15GHz, it results in a pulse response with 3 samples/bit, whereas 12.5GHz gives 2.5 samples/bit. Long channels with good material may also have data above the noise floor. Is this freq range for simulation limits or for the channel compliance limits? Joel sees them as the same. Charles sees them as different needs. Some preferred to take more data rather than less. Can throw away data later. We don't have a signaling scheme selected. We should trim the frequency range down later, after the signaling selection. Need to screen interesting/non-interesting channels. The Some are worried about irrelevant data getting in, some worried about relevant data being included. Wrap-up straw poll: #1 - Do you agree that the freq range we measure and simulate to should be the same as the bounds we use in the informative mask set? yes - 1111111111111111111 -(19) no -11 (2) abstain - 11 (2) #2 - We have discussed 2 frequency ranges 50-15000 MHz & 12500 MHz Do you agree that 50-15,000 MHz is the freq range is the appropriate measurement, simulation and informative mask set range? yes -11111111111111111111111 - (23) no - (0) abstain -1 (1) #3 - Do you agree that 50-12,500 MHz is the freq range is the appropriate measurement, simulation and informative mask set range? yes -11 - (2) no -1111111111111111111 (19) abstain -1111 (4) No disagreement with proceeding with 50 - 15,000MHz. No disagreement to Joel presenting to the greater IEEE committee. Group delay Joel adjusted equations and result is on Page 5 of Joel's presentation. May interim meeting: ~+/-22ps - ~+/-190ps @15,000 MHz. Borderline SDD21 channels tend to fall within this group delay. Concern that the bounds are so wide that it does not exclude any channels (Charles). Some channels are actually excluded (Joel). Simulations showed that the average group delay was 3ns. Range variance at 100MHz: 1ns to 7.5ns. So starting point range needs to be very wide. Joel concerned that if we narrow the range too much, we will have to narrow the Dk/Df too much, such that Nelco 4000-13 will be eliminated. Group delay = d(phase)/d(freq). Why was the initial range so far off? Concern was that there were so few backplanes that would meet the mask set. Why is the Y axis Group Delay Variation, but the title of the slide is Group Delay Mask? Should change title to Group Delay Variation Mask. Joel: Graph title should be Group Delay Mask - Joel intends to go back and say this is GDV and this is how we should look at it. COncern expressed that group delay seems to be making a major impact in material choice. If it is Variation, what is the variation with respect to? It is defined as Zero at the 50MHz (which is the reference point). #4 - Are you acceptable to the new mask set that Joel sent out for presentation to the greater IEEE committee? In the presentation that J delivered on behalf of the AH group, we Change title to Group Delay Variation? Charles - does not agree. Plans on presenting that at the meeting next week. Feels that in order to have an asymmetrical pulse response, the GD will have curvature. Joel - all he is asking is to change the title. JoeA - Are you changing both the title and the numbers? Joel needs to go back and make sure that Group Delay variation of certain channels fits within the channel. He will work through the errata. Try to get data for Monday's discussion. VNA Measurements Reference: Joel's sent out data. Reference: John D'A sent out data. IF BW @ 300Hz - Joel sees as not an issue. 10 and 100Hz did affect it. Output power - is -5dBm OK for NEXT/FEXT? As long as dynamic range >= 80dB or better we are OK with -5dBM. UNH Data in JD'A's presentation: NEXT only. Seeing deltas between measurements at the 2 different power levels. Need to examine... Deep nulls show differences. Channel measured was adequately designed for a 3G application. anyone opposed to adding 80dB - Bob has issues on equipment meeting anyone against -5dbm until we get more info - no VNA averaging, stay at 16 until we can prove otherwise, Graeme feels 16 is the min average we can go and get better measurements. - no issue staying at 16 until we get data to prove. will discuss at next ccall Channel model test card Crosstalk NEXT & FEXT - Accidently dropped by 2dB instead of 10dB is everyonbe okay leaving the current equations and limit line intact and noting the discrepancy to the 802.3ap body. yes - 11111111 (8) no - 11 (2) abstain - 111111 (6) Keep the curve and equatiosn but I will point out that earlier minutes indicated a 3db increase at 10gig. comments were made if the model was going to address the three connector set. It was noted that lots of work was being done here to show this is not an issue.