Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[10GMMF] Notes for TP2 call, 8/19/04

All - here are my notes for the 8/19 TP2 call. Please send corrections if needed.
Attendees (no order)
  • Martin Lobel, Intel
  • Jan Peeters Weem, Intel
  • Sudeep Bhoja, Big Bear
  • Bob Zona, Intel
  • Jens Fielder, Infineon
  • Lew Aronson, Finisar
  • John Ewen, JDSU
  • Yu Sun, Optium
  • John Dallesasse, Emcore
  • John George, OFS
  • Bala Mayampurath, Vitesse
  • Pekko Sipila, Modulight
  • Abhijit Shanbhag, Scintera
  • Piers Dawe, Agilent
  • Tom Lindsay, ClariPhy
  • Norm Swenson, ClariPhy
  • Paul Voois, ClariPhy
  • Chet Babla, PhyWorks
  • Sudeep Bhoja, Big Bear
  • Lars Thon; Aeluros
  • I apologize for missing others who did not respond to my requests to send emails.
Summary objective (repeated from before)
Present a proposal for TP2 signaling parameters and associated conformance testing at the September Meeting. The work must consider and provide tradeoff information among component cost, test cost, and power penalties.
Reduction in power was mentioned by Lew today as another outcome to consider.
  • Proposed agenda was approved (see below).
  • Previous notes were approved.
Progress, technical discussions
Intel presentation (presented by Martin)
  • Simulations.
  • Considered penalties vs. laser packaging and laser chip performance.
    • Note - laser driver is another variable...
    • Packaging model was simple 4th-order Bessel Thomson (BT4) filter. Reflections not considered.
    • Laser chip modeling based on rate equations.
      • Parameters chosen to result in various RO frequencies.
  • Other setup for this presentation.
    • Single Cambridge fiber. Intel sampled a few fibers and found similar trends. 300 meter (no scaling).
    • Rx front-end BT4 filter @ 7.5 GHz.
    • Equalizer taps at T/2 spacing.
      • FFE 5 taps.
      • Feedback taps (for DFE): 2.
    • Reference penalty for all slides is lower right cell on slide 8 (perfect/perfect).
    • RIN not included in this setup.
  • Discussion
    • Laser currents seem higher than normal.
    • Observed that unfiltered eyes did not show classic overshoot and ringing, that they were highly damped.
      • Martin thoughts that trends would be the same, however.
    • RIN should be included in future modeling. As eyes close, effect of RIN increases.
    • Curious why some results of slide 13 are better than PIE-D.
    • FFE results on slide 15 seem pessimistic. May be due to small number of taps (only 2.5 bits), but some tap limit is appropriate for practical reasons.
    • Addition of DFE makes a very large improvement. Appears that it may be required for 300 meters?
    • Results are 1E-9 BER, but Martin expects trends would be the same with small dB offset.
    • Package seemed to have more penalty impact than laser.
      • Could be due to "step size" relative to data rate (8G to 4G more significant in ratio and relative to data rate than 11G to 6G).
      • Note - could also be due to steep 4th order function of package model (I recall lasers are more typically 2nd order above ROF).
    • What is overall budget and relative to what?
      • With DFE, all combinations with fiber are less than 5 dB (relative to "perfect" Tx, back to back).
      • Perhaps penalties should be compared to -L, which is approximated by the good/good cell. If this is done, then the worst penalty on slide 13 is 1.1 dB. Said differently, 4 dB back to back penalty relative to -L (slide 8 bad/bad vs. slide 8 good/good) results in 1.1 dB EDC penalty at end of channel (see slide 13, same cells).
        • Very encouraging results.
      • These penalties are data dependent and do not include RIN or other jitter.
    • Very interesting observation is that penalty impact of TP2 is same whether back to back or with fiber.
      • May help in developing a test that does not need fiber (such as -L test in 802.3ae).
  • Cost drivers
    • Large number of variables possible.
    • Need cost trends to direct simulation (and experimental) efforts.
    • Cost trends will be vendor dependent.
      • Laser chip yield.
      • Packaging costs.
      • Laser driver costs and yields.
      • Others?
  • TP2 test metrics
    • Mask test only would be nice, but concern that it does adequately constrain all degradations.
      • Some degradations will not be correctable (by EDC) whereas some would be.
      • Therefore penalty impacts will be different. Mask cannot distinguish.
      • This will be more true if specs become more relaxed.
    • Re-stated that low test cost/complexity is important - test cost cannot increase more than other savings.
    • Folks need to consider appropriate test metrics now.
      • Simulation work (i.e., Intel work) should calculate test metrics to see how well they track and predict penalties.
Next work
  • Questions for Intel
    • How were EDC coefficients were determined and how BER was calculated (see hanberg_1_0304.pdf)?
    • Where does fiber used line up in the Cambridge set? What cumulative percentage does it represent of the installed base?
    • Why are some results in slide 13 better than ideal PIE-D?
  • Modeling should consider.
    • RIN.
    • Lasers with overshoot and ringing.
    • More realistic electrical drive modeling to include reflections.
    • Other tap combinations.
Next call (8/26, 9 AM Pacific time)
  • Cost vs. performance studies (Opnext, Bookham)?
  • Other presentations encouraged.
    • Cost trends
    • Penalties and test metrics vs. parameters (and try to relate to cost trends)
  • Clarification of budget - how much can be available for TP2-related degradations?
  • TP2 test metrics.
Tom Lindsay
ClariPhy Communications
phone: (425) 775-7013
cell: (206) 790-3240
----- Original Message -----
Sent: Tuesday, August 17, 2004 8:48 PM
Subject: [10GMMF] Reminder for TP2 call, 8/19/04

Meeting details (same numbers as before):
  • Date: Thurs, 8/19/04 (regular day/time)
  • Time: 9:00 AM
  • Duration: 1:00 goal, 1:30 max
  • Number: 401-694-1515
  • Access code: 421721#
  • Approve agenda
  • Approve previous minutes
  • Presentations/results?
    • Cost vs. performance studies (Intel, Opnext, and Bookham?)
    • Non-linear laser effects (Agilent?)
    • Others?
  • TP2 test metrics
Tom Lindsay
ClariPhy Communications
phone: (425) 775-7013
cell: (206) 790-3240
----- Original Message -----
Sent: Monday, August 09, 2004 4:18 PM
Subject: [10GMMF] Minutes/notes for TP2 call, 8/6/04

All - here are my notes for last Friday's TP2 call. Please send corrections.
  • Martin Lobel, Intel
  • Jan Peeters Weem, Intel
  • Jim Heckroth, Silicon Labs
  • Abhijit Shanbhag, Scintera
  • Andre Van Schyndel, Bookham
  • Piers Dawe, Agilent
  • Tom Lindsay, Norm Swenson, Bharath Jagannathan; ClariPhy
  • Brent Whitlock; RSoft
  • Nick Weiner, PhyWorks
  • Sudeep Bhoja, Big Bear
  • Ed Cornejo; Opnext
  • Lars Thon; Aeluros
  • Others?
  • Proposed agenda was approved (see below).
  • Previous minutes were approved with the correction on attendance (see below).
  • Project objective statement was approved (see below). There was some discussion on whether to mention interactions with other groups, but it was felt to be sufficiently implied.
  • We decided to not have specific liaison reports from TP3 or channel modeling groups; Given the common attendance by many, if those groups need to represent a relevant point or view to this group, they are encouraged to do so, and vice versa.
  • Dates/location for interim meetings still TBA.
  • Call dates: weekly, Thursdays, 9 AM Pacific time, starting 8/19.
Progress, technical discussions
  • Refer to slides uploaded with tlindsay email to LRM reflector, 8/6/04 AM.
  • 1st slide, with contribution from John Ewen, shows Tx and components. Need to determine which components have the most impact on cost, then evaluate their impact on budget (penalty): % cost savings/dB penalty metric.
  • Effect of TP2 performance may be dominated by long channels. This is the motivation for considering low cost components, such as TO-cans.
  • This work will be implementation-dependent; depending on design approach, levels of integration, etc.
  • Desirable to have existence proof of running 10G traffic with real hardware using low cost components.
  • Some relaxed spec vs. cost work done, but not yet ready to present.
    • Intel is doing work on packaging, impedance mismatches and laser (rate equations, RO, etc.) variations, including use of 4G components.
    • Bookham is performing internal studies, seeing potential of ~10% cost savings.
    • Others?
  • Even though components may be capable of relaxed specs (e.g., risetime), performance should still be controlled (to some level) over the spectrum required for 10G transmission. That is, there may be hidden costs that offset some of the savings.
  • Volume may provide more potential cost savings than relaxation of specs.
  • Relaxed component specs may enable lower costs sooner, allowing market volume to begin sooner.
  • Non-linear lasers do not exhibit Gaussian impulse response - more work required to evaluate range of laser properties against EDC to assess equalizable and non-equalizable power penalties.
  • In trying to focus your work, please consider what inputs are needed from others.
  • Regardless of cost reduction effort, we must determine set of TP2 test metrics to ensure interoperability.
    • Simple mask test does not account for all potential penalties allowed in the current budgets.
    • Goal is for tests to be simple and not high cost.
    • Test metrics should be used during cost studies.
Next call (8/19, 9 AM Pacific time)
  • Expect results of cost vs. performance studies from Intel, Opnext, and Bookham.
  • Agilent may look into non-linear laser effects.
  • Others encouraged.
  • Work on TP2 test metrics must proceed independently.
Tom Lindsay
ClariPhy communications
phone: (425) 775-7013
cell: (206) 790-3240