May 4, 2005 IEEE802.3ap Channel Ad Hoc Meeting Agenda: Link Budget Cross Talk Duty Cycle Distortion Deadline to submit presentation time is May 11 @ midnight Interim meeting is May 16-18 in Austin, TX Carry Over Items Rich - Molex channels pulse response look similar. We need a good method to evaluate channel quality. Xiao Ming will bring more data. Currently time domain analysis doesn't show goodness/badness. Charles - this affects normative and informative projects Rich - Correct. The problem is how to analyze DFE and FFE. Link Budget Model (Charles Moore) Charles - John - RX SDD11 is device or channel? Charles - either RX SDD11 penalty, or compute interaction between nominal RX and channel SDD11 John - add in the channel SDD21 parameter like ACR relationship. Prefer Hybrid Eye loss (multiplicative #6) Charles - breaking up RX SDD11 into two effects Brian - Ultimate usage of link budget? Charles - calculate a link budget. Parts dependant on channel. Brian - link budget normative for channel compliance? Charles - correct Rich - When I've added up dB budgets there is a chance of double counting. Charles - the numbers need to add up correctly, otherwise fix it. Xiao - RX SDD11 data shows it's significant? Charles - Don't have data right now. Anirudda - #5,6,7 added up, need to be positive. The margin needs to be at least 0. Equalization is required, so where is that added in? Charles - page 6 shows "channel loss" which implies equalization compensates. page 7 Anirudda - don't think we'll get a positive number after adding up page 6. Charles - how about a channel loss and equalization recovery Rich - popescu talks about a power budge Charles - I looked at the cunningham article, and applied it to Tyco channel 2. With 1V launch and 400mV of eye opening, i thought that number was optimistic. Would like to see a convincing channel model. Spcifying Crosstalk (Adam Healey) Adam - frequency based power sum specification is sufficient to define crosstalk. Avg xt power is the integral of the PSD after the agressors are summed in. Rich - do you think the reason the xt is increasing, is that the sum is double counting? Adam - no, as you go to longer patterns, the crosstalk approaches gaussian prediction John - why did you choose these two cases? Adam - presented two cases that represent the general trend. Test case 5 is tyco, test case 22 is molex. Charles - slide 10, the pulse response magnitude sum at cursor intervals provides the worst case bound Adam - that's what I planned to do Charles - general comment, validity of central limit theorem depends on how few or how small the terms are. Adam - As you integrate over phases, it smears out. At 1E-4 or 1E-5 the gaussian distribution fits. Beyond the mean, the gaussian is pessimistic. Charles - the integral is pessimistic, so now you take the integral of the bounding curve, so we may be leaving too much on the table. How about bounding the integral instead of bounding the curve. Rich - As long as the data matches, I have no issues. Charles - We said we should apply the same equalization for the aggressor as the victim. Adam - Xiao Ming - Is there similarity between time domain and frequency domain? Do we need time domain additionly. Charles - Adam's point is that freq and time domain analysis are mathematically identical. Adam - We had a time domain method that is good. The frequency domain provides a shortcut easier to measure. Unless you violate the setup assumptions, time and freq methods are mathematically proven identical. Duty Cycle Distortion (Shannon Sawyer) Brian - Does this replace Phase Noise Amplification? Shannon - The voltage effects are linear with timing, and capture all the change to the signal through the backplane. Xiao Ming - what about different data patterns. Shannon - alternating 1010 provides worst case Adam - What algorithm can be used to spec DCD? Charles - Since the timing and voltage affects are linear, we can find out how much amplitude reduction is acceptable, and place that value wrt the input signal (timing or voltage) as the spec. Charles - Biggest item we need is data items to provide the link budget. The second biggest is...