Budget- and Penalty-oriented EDC. System Performance Evaluation

Transcription

1 Budget- and Penalty-oriented EDC System Performance Evaluation Lars E. Ton Aeluros, Inc. IEEE P802.3aq 10GBASE-LRM Task Force Portland, Oregon,

2 Outline Budget- and penalty-oriented system performance evaluation. Performance model: Penalty attributable to DMD before and after equalization. DFE Adaptation Figure-Of-Merit (DFE-AFOM) for cannels Results: Cambridge Fibers at L=220m and L=300m ISI Power Penalty for unequalized fiber EDC idealized penalty (unequalized ISI+noise enancement) EDC implementation penalty 2

3 10Gbase-LRM Link Budget Link budget is loosely based on 10Gbase-LR Carve out un-needed penalty and loss allocations, and assign te slack in te budget to DMD/EDC as a maximum allowable penalty. Current tinking is tat tere is about 6dB left in te budget to spend on EDC. EDC idealized penalty: ~4.5dB EDC implementation loss: ~1.5dB Circuit implementation loss Static and Dynamic Mis-adaptation loss Implementation complexity limitations (e.g. FFE and FBE number of taps and teir precision). CDR static and dynamic sampling error It is likely tat te db split is not entirely accurate. 3

4 Performance Modeling Implications Te budget-based approac implies a comparison wit an ideal fiber (zero differential mode delay) and a launc power of 1 (arbitrary unit). d=+/-1 Pulse Sape Cambridge MMF model Te objective of te model is to establis te penalties associated wit correction and residual of te ISI of te non-ideal fiber. Ideal MMF model 4

5 Budget-Based Power Penalty Model c ( t) = N 1 k = 0 ~ c ( k)* δ ( t t k ) pulse ( t) = N 1 k = 0 ~ c ( k)* p( t t k ) d=+/-1 Pulse Sape Cambridge MMF model MMSE-DFE Optimizer Penalty Calculation Ideal MMF model MMSE-DFE Optimizer p t T t ( / ( ) K e 2σ 0 2 = ) 2 5

6 Model Input: Back-to-Back Pulse Responses %? Agilent spreadseet model uses Gaussian pulse wit Trise(20-80)=47.1ps, translating into sigma=0.4183, or pw50=100ps. 6

7 Power Penalty Calculation Te power penalty of te EDC unavoidably depends on te implementation, for example LE or DFE Nffe and Nfbe (number of taps) Implementation losses Idealized models (PIE-L and PIE-D) provide guidance and correspond to unbounded-lengt equalizers wit zero implementation losses. Specific lengts considered for implementation are also of great interest. Tx or (better) back-to-back pulse sape also matters! 7

8 Power Penalty Reference Model n(t) d=+/-1 Pulse Sape Cambridge MMF model FFE Slicer n(t) FBE Pulse Sape Ideal MMF model Slicer Noise is included only in te sense tat we want to evaluate ow muc te noise is enanced by te EDC. 8

9 Power Penalty : Residual ISI Fiber/launc #163 (nearly te worst case) at L=220m 9

10 Power Penalty : Noise Enancement Fiber/launc #163 (nearly te worst case) at L=220m 10

11 Power Penalty Calculations Te ideal fiber as by definition a penalty of 0dB and an rx pulse wit cursor amplitude 1. A DMD-impaired fiber wit no compensation as a specific ISI penalty relative to te ideal fiber: After MMSE-DFE, te penalty relative to te ideal fiber is comprised of two factors: Residual ISI Noise Enancement Note: FFE(n) is scaled so tat cursor=1. ` DP DP DMD ISI DP 2 = = 2 = = NE = cursor - cursor 1- precursor 1 FFE pulse pulse ( n) 2 1 ( k) - max - 1 precursor pulse 1 N -1 k= 0 1 pulse ( k) 11

12 Cambridge Fiber Model : Pulse responses Ampl Cambridge pulses, L=220m, Gaussian, pw50=100ps 12 Time [ps]

13 Power Penalty : Before and After MMSE-DFE Penalty[dBo] Budget limit =~ -6dBo DFE-AFOM also sown Fiber Number (sorted) 13

14 DFE Adaptation Figure of Merit Fiber/launc #163 (nearly te worst case) at L=220m 14

15 Cambridge Fiber Model : Pulse responses Ampl Cambridge pulses, L=300m, Gaussian, pw50=100ps 15 Time [ps]

16 Power Penalty : Before and After MMSE-DFE Penalty[dBo] Budget limit =~ -6dBo DFE-AFOM also sown Fiber Number (sorted) 16

17 Summary Budget-oriented analysis quantifies effect of EDC witout detailed BER estimation. Sorting EDC penalty measures according to te raw ISI penalty provides insigt into distribution. Consideration of specific implementation coices are an important complement to idealized metrics. Results indicate tat at least 4dBo of budget is required to cover te Cambridge Model fibers, wit ballpark coices of Nffe and Nfbe, net of remaining EDC implementation losses. 17