May 11, 2005 IEEE802.3ap Channel Ad Hoc Meeting Agenda: Link Budget Model Update Reminders: Deadline to submit presentation time is May 11 @ midnight Deadline for draft comments is May 11 @ midnight 802.3ap Interim meeting is May 16-18 in Austin, TX Carry Over Items: none Link Budget updates (Charles Moore) Charles - added separation for channel loss and equalization effect Rich - How about a TX that has low DCD? How do you separate that from re-reflection? Charles - DCD is pattern independant Petre - 64/66 data has baseline wander Charles - DCD is both additive (baseline) and multiplicative (eye width) Petre - Even if you have a perfect TX, you still need a DCD limit. Charles - I will add the DCD due to AC coupling of RX to channel Rich - What is the criteria? Charles - You need to have >0. RX margin is min amplitude for slicer, so you don't have a need for an open eye. Rich - How do you know what the RX margin is? Petre - If you make assumption of noise floor, offset and slicer input, everything is covered. Xaio Ming - If the noise floor is defined before equalization, what about noise amplified by equalization? Not sure it's feasible to separate the equalization noise in the chip for the standard. Straw Poll: Use square root of power sum of cross talk power integrals as described in healey_c1_0505 multiplied by TBD as crosstalk penalty. Brian - lanes are plesiochronous, so cross talk has the opportunity to be phase aligned Adam - any offset between victim and aggressor will have walking function, and there are many opportunities for trace offsets between channels Brian - offset is 1-10 clock cycles Adam - BER analysis between small tolerance and large tolerance is difference between bit error and burst error Brian - bit errors is a collection of events, and high probability that many are transitioning at the same time. RMS is overly optimistic Adam - would you say StatEye is optimistic? Brian - I believe it is different Adam - According to CEI, this is how StatEye treats crosstalk Brian - 4-8 aggressors, each use 1/3 interval to transition, so you have 1/12 time to crosstalk this provides 8% Adam - probability of phases align is low. skew tolerance of 50ps between lines will give constructive iterference? Charles - Brian assumed all the energy coupled over during the time of transition of the aggressor signal. This is not a valid assumption with our 802.3ap channels. I get 50% contrast between worst and best phase as I shift the phase between victim and aggressor using RSS method. Xiao Ming - Are you including the phase? Charles - When you do RSS, you assume phase is incoherent. You can predict the statistical likelihood that they will align. When you sum linearly, you assume all are constructive. When you do RSS, you assume a random mixture. Brian - Plesiochronous signalling has worse crosstalk. Charles - You are worse in that the coupling is larger Rich - If you had syncronous TX they could slam the RX at the same time. Charles - The path lengths are different. The delays are different. Looking at the pulse response of our channels some are negative some are positive. Adam - TX have skew on the die. Rich - They may not have any skew. Adam - I think it's highly unlikely. Bound the contribution of any single aggressor, and the contribution of any two aggressors in a worst case sum. Rich - When everyone's training, you get the worst crosstalk. Adam - Mission mode should be better. Brian - If we are dealing with 4 local aggressors, you get an 8% probability Adam - You get a gaussian dist with a lot of aggressors or long scrambled codes Charles - sign me up for doing a presentation for how the gaussian distribution is valid. The argument is that you have four aggressors only, indeed you have high probability they will all combine. Gaussian RMS X 7 or 14 you get a larger number than the sum and indeed it is pessimistic. Brian - Root Mean Square vs Root Sum Square Charles - The difference is taking the square of either the mean or the sum of the values. Adam showed that the phase doesn't matter. Adam - depending on how you use scale factor is how optimistic or pessimistic is determined Brian - will one scale factor suffice? Charles - the value is 14.x Adam - which is fine for a link budget analysis. Maybe a different number is better. Brian - small number of aggressors with gaussian dist is overly pessimistic. Straw Poll votes: Y 5 N 2 A 12 Is there aggreement with the link budget approach? Rich - let's see the numbers then decide Charles - I take your comment seriously, and ask for a straw man that we can throw darts at. Rich - For example RX noise margin, I don't know what to put there. I am not seeing the simple picture. Charles - I'll record reluctant consensus. Xiao Ming - I'd like to see more data. Simulation results.