20050406 802.3ap Channel Ad Hoc Telecon notetaker: Shannon Sawyer Agenda: 1. email notification 2. schedule of events 3. carry over items 4. Presentations: Mellitz's Normative Channel Model John D - Does the pulse response include the TX FFE? Rich - It could, it assumes the channel does it's job. Charles - looks like a link budget. Determine system losses, provide compliant Tx, multiply RX equalization, then deduct the channel and RX/TX mismatch impairments. Rich - you will throw away budget if you have large reflections. The S-params have to be used as an aggregation of TX and RX. Charles - like the link budget... if you're careful Rich - yes, I'm trying to get back to time domain Howard - page 4 only defines TX pulse? Rich - channels are measured, pulse is defined, package spec is applies Howard - what about RX Rich - package on both sides Howard - why assume FFE, why not ignore and do the whole channel pulse response Rich - either way is fine. Want to see what correlates best with our channels. Howard - keeping channel independant from RX is preferred John - page 4 is thru pulse response and xt pulse response with same channel. How much will pulse change? Rich - We'll see John - thru and xt pulses are related, how about a family of pulse shapes Rich - we'll see Charles - final point is...does this correlate with our channel measurements and pulse response methods. Will you have correlation? Rich - yes. John will present ordering. John's Informative Channel Model John - Rich, can you comment on page 3 Rich - there are issues with the data, need to evaluate to clarifyWant to add some Matlab code, need Matlab 7.0 Adam - have to drop off, Charles please send meeting results. Charles - slide 6, key slide to proceed Rich - great stuff Charles - if you were to write a spec from slide 6, where would you draw the horizontal line? Rich - we are missing new intel channel data John - not sure, 0.5-1 maybe? Molex channels had very different results, judging S21 may be incorrect. If the channel model is right, and xt can be done differently, need ACR Rich - can put number at 0.25 Matt - everything less than 1 contains a 0, so is a 1E-12 BER failure, and wouldn't work Luke - not getting Intel boards soon, so new xt/channel measurements will be available later Charles - based on slide 6 and slide 8. Do all sims have same quantity of sims John - no Charles - what's the difference John - considered sims from same vendor one at a time Charles - 1-3 ranking is per vendor John - Yes, Molex had 3 sims, Intels had incomplete, Tyco was from everyone. Exluded Xilinx since couldn't read spreadsheet Charles - Xilinx needs to update spreadsheet John - 0-10 = 0, 10-20 = 1, 20-30 = 3. Should I look at V or T? Charles - both correlate Rich - need both John - I will do both Xiaoming - only consdier amplitude John - Adam may be able to contribute xt information. September presentation shows ACR explanation Charles - John I'll talk to you about XT convolution later 5. walk in items none 6. Straw Polls (topic) Are we ready to report the channel informative model as complete? No (topic) What should the normative channel model look like? John - Rich's presentation includes pkg,xt,etc similar to budget approach is this like Charles, Petre's or is it a hybrid. Charles - problems caused by impedance mismatch. My method treats re-reflections from TX and channel and is treated as XT on the pulse response Petre - my view is that the package can't be included, although my method includes TX/RX package. I included reflections from the ends. John - Mellitz's model may be a hybrid of 1 & 2. Charles - You are assuming a TX or RX, and when you change them, the response changes. Moore's method delt with mag of reflection, not the sign. With so much XT present it's safe to ignore the XT phase. Rich - let's wait to see what matches measurements. Should we use the package or not in the simulation? Shannon - Rich provided good range of packages. Joe found the package to cause 10-20% channel performance degredation (vote) Straw Poll #1 Should the affects of the worst case RX and TX return loss be included in the normative channel model? Results: yes=14, no=2 (vote) Chicago rules (everyone can vote for everything) 1. Full channel sim with XT, RX, TX like stateye. 2. link budget like petre or charles Charles - includes TX pulse times RX equalization subtracting channel losses to get eye size 3. Hybrid of 1 & 2 (Mellitz proposal) Rich - let's defer until we get data Results #1=3, #2=11, #3=10 (vote) Straw Poll #2 Are you interested in using an RX normative spec instead of or in addition to a normative channel spec? Charles - RX Interference Tolerance. Normative spec would be a compliant TX with a compliant channel to get a compliant BER. This means the RX was good enough. Xaioming - means the RX and the channel are together Charles - bad channel will screw up RX Xaioming - design of the equalization makes a lot of difference Charles - RX spec alternative to spec the channel Anirudda - Based on previous straw poll what additional information does straw poll three provide? Charles - Current assumption is to spec the TX, spec the channel, expect RX to work. Anirudda - Petre's link budget specs the RX. Charles - He specs what the RX should do, but can't be measured. This SP#2 is about a measurable RX spec. A normative spec implies it is measurable. Present method is in D802.3ap annex 73A. Results: yes=3, no=5, abstain=6