************************************************************************* ------------------------------------------------------------------------- Gigabit Ethernet over CAT-5 UTP Cable Interim 802.3z interim meeting Kamran Azadet Bell Laboratories / Lucent Technologies ------------------------------------------------------------------------- Our goal * 1 Gb/s * Full-duplex * Worst Case CAT-5 * 100 meters * Reasonable cost / power and time to market * Reasonable analog requirements ------------------------------------------------------------------------- OUTLINE * Coding scheme : why a 3x3 constellation ? * Digital Implementation * Mixed-mode Implementation * Combining FFE and DFE * Echo and NEXT cancellation * What next ? ------------------------------------------------------------------------- Why a 3x3 constellation ? * Large constellation, low symbol rate (ex : 9x9, 41.6 MBauds, 125 MS/s T/3) * Small constellation, higher symbol rate (ex : 3x3, 83 MBauds, 250 MS/s T/3) (figure shows 81 point and 9 point constellations) ------------------------------------------------------------------------- * Trade-off between : high accuracy at lower speed and lower accuracy at higher speed * Reduced size constellation is more robust in terms of immunity to amplitude and phase noise (clock jitter) * Relaxed analog accuracy requirements * Simple symbols (0,+/-1) => no multipliers needed for DFE, Echo / NEXT cancellers * Eases the convergence of channel equalizer ------------------------------------------------------------------------- Digital Implementation simplified block diagram (figure shows transmit/receive digital blocks clk=83MHz, A/D and D/A clk=250MHz) A fully digital system requires 250MHz A/D and D/A with an accuracy of 6 bit ------------------------------------------------------------------------- Mixed-mode implementation Combine analog FFE and digital DFE : symbol rate ADC (figure shows transmit and receive filters clk=83 MHz using polyphase structure, A/D , D/A , digital adaptive DFE, clk=83MHz) ------------------------------------------------------------------------- Combining FFE and DFE * Less noise enhancement than FFE only * Lower overall hardware complexity (FFE requires multipliers - DFE : no multipliers) * Better immunity to R.F. interference Best performance is obtained when both FFE and DFE are adaptive ------------------------------------------------------------------------- Echo and NEXT cancellation * Simulations based on Return Loss measurements of a complete link, hybrid not included -> models were degraded 3dB in order to account for hybrid (Figure shows FFE, Echo/NEXT cancellers and DFE) ------------------------------------------------------------------------- * Simulation platform : NEXT models from Bob Campbell, MSE margin = +4dB is achieved for BER = 10E-10 (figure shows a snapshot of the simulation platform with windows corresponding to FFE taps I and Q, Echo and NEXT cancellers, DFE, the 3x3 constellation and the MSE plot) ------------------------------------------------------------------------- * Simulation platform : NEXT models from SDE, MSE margin = +6dB is achieved for BER = 10E-10 (figure shows a snapshot of the simulation platform with windows corresponding to FFE taps I and Q, Echo and NEXT cancellers, DFE, the 3x3 constellation and the MSE plot) ------------------------------------------------------------------------- What's next ? * Urgent : set an agreement on evaluation criteria and models, define requirements * Optimize parameters to get the lowest cost that meets the requirements * Define PCS / PMA * Prototype ------------------------------------------------------------------------- ************************************************************************* ------------------------------------------------------------------------- Evaluation of proposals * Chose a common set of models for NEXT, Insertion and Return Loss : models should represent an actual cable (measurements) * Agree on a set of performance evaluation criteria and requirements * Share information needed for double checking results ------------------------------------------------------------------------- Choice of models * Models should reflect a physical system * NEXT models should be as close as possible to the worst case envelope * They should be selected based on an objective criteria (they should not be scheme oriented !) ------------------------------------------------------------------------- (plot of SDE curves show large margin from envelope in 0-70MHz range and curves touch the envelope for frequencies from 80-100MHz) ------------------------------------------------------------------------- (plot of Bob Campbell NEXT show curves touching the envelope curve uniformly along the frequency range 0-100MHz) ------------------------------------------------------------------------- Margins from worst case envelope SDE's NEXT models in the 0-70MHz band in the 0-100MHz band n1 -> +8.3 dB n1 -> +7.4 dB n2 -> +10.5 dB n2 -> +8.1 dB n3 -> +8.4 dB n3 -> +7.5 dB n4 -> +11.6 dB n4 -> +10 dB = 9.6 dB = 8.2 dB Bob Campbell's NEXT models in the 0-70MHz band in the 0-100MHz band n1 -> +5.5 dB n1 -> +5.0 dB n2 -> +4.7 dB n2 -> +5.8 dB n3 -> +4.0 dB n3 -> +4.4 dB n4 -> +6.9 dB n4 -> +6.2 dB = 5.2 dB = 5.3 dB ------------------------------------------------------------------------- Evaluation criteria * Simulations based on Full-duplex, 100m, UTP CAT-5 Worst Case Link models * Simulations should be bit accurate, include analog functions and timing recovery * Description of PHY control and adaptation algorithm * Detailed estimate of silicon area, power consumption (technology ?). For digital blocks : transistor count is an objective measure of complexity * Susceptibility to Noise * EMC compliance