Update on technical feasibility for PAM modulation Gary Nicholl, Chris Fludger Cisco IEEE 80.3 NG00GE PMD Study Group March 0
PAM Architecture Overview [Gary Nicholl] PAM Link Modeling Analysis [Chris Fludger] Strategy and Assumptions Analysis (RX Sensitivity, RX BW, RIN, ER, CD, DGD, etc ) Summary PAM MPI (Multi-Path Interference) Analysis [Chris Fludger] Approach Results for single reflected path Extension to multiple reflected paths Implications on connector requirements for different system configurations
3
This work investigates the technical feasibility of PAM modulation schemes for a single laser 00G SMF PMD PAM8 and PAM6 are investigated nicholl_0_0 showed that PAM potentially provides substantial cost savings over current 00G-LR4, primarily due to reduction on optics component count / mfg complexity Goals: A substantial lower cost solution for 00G (that is potentially scalable to 400G) Single laser. Uncooled operation is desirable. Wavelength: 30nm Loss Budget: 4dB Link Length: 500m to km 4
5
Presented in nicholl_0_0 Substantial cost savings (in range of 0.5x to 0.5x), compared to mature 00GBASE-LR4 (blue line above) 6
7
Numerical simulations using VPI simulator More complex scenarios Analytical calculations Deeper understanding Effects examined and short summary Chromatic dispersion : All formats > km DGD : All formats > km RIN : Stringent requirements for higher order PAM Thermal noise / Sensitivity : M-PAM is worse than simply faster ASK Electrical bandwidth : Higher requirements for M-PAM Extinction ratio : Impairment same for M-PAM 8
4 Lasers levels @ 8Gbaud Laser Rx Modulator Modulator Modulator Modulator Link Lasers Laser Modulator Rx 4 levels @ 8Gbaud 8 levels @ 8.6Gbaud Modulator Link 4 levels @ 56 Gbaud 8 levels @ 37 Gbaud Laser Laser Modulator Rx 6 levels @ 8 Gbaud Link 9
Rx Tx Link Wavelength available : 95.56, 300.05, 304.58, 309.4 nm Dispersion : -.85 to +0.95 ps/nm/km For km, there is no CD penalty Loss ~0.43 db/km Connectors/Splices :.5dB ( db in IEEE spec. for 0km) Total loss for km fibre =.4 db 0
Channel dominated by thermal noise No RIN included 4 N T kbt R L N th f e.g. 0pA/ Hz Extinction ratio can be included as power penalty later M-ASK levels are linear after square-law optical detector Noise is evenly distributed on rails Baud-rate is constant at 5Gbaud More levels means more overall capacity (per wavelength)
log(ber) 0pA/ Hz) - VPI simulator : levels levels 4 levels 8 levels 6 levels 3 levels -3-4 -5-6 -7-8 -9-0 - - -5-4 -3 - - -0-9 -8-7 -6-5 -4-3 - - -0-9 -8-7 -6-5 -4-3 - - 0 Received power (dbm) Agreement between analytical and VPI model for M= Levels (M) 4 8 6 3 Penalty (db) 0-4.8-8.5 -.8-4.9 Penalty[ db] 0log0 M at low BER 3
log(ber) 5pA/ Hz) - -3 laser : 4 levels : 56 Gbaud laser : 8 levels : 37.3Gbaud laser : 6 levels : 8Gbaud lasers : 4 levels : 8 Gbaud lasers : 8 levels : 8.6Gbaud 4 lasers : levels : 8 Gbaud -4-5 -6-7 -8-9 -0 - - Baseline - -0-9 -8-7 -6-5 -4-3 - - -0-9 -8-7 -6-5 -4-3 - - 0 Received power (dbm) All scenarios carry Gbit/s data Received power is at the photo-detector (demux etc not included) 4
5
Thermal noise 4 N T kbt R L N th f Noise from RIN depends upon the level (m=0...m-) e.g. 5pA/ Hz In a binary signal, the s have RIN, 0 s have no RIN (for infinite extinction) Extinction ratio is ignored in this analysis (optimistic) m=3 RIN I m RIN f m= e.g. -30 db/hz m= m=0 Decision thresholds at receiver should ideally be optimised. 6
Power Penalty (db) 3.5 levels : Penalty @e-5 4 levels : Penalty @e-5 8 levels : Penalty @e-5 Dotted lines are with a fixed decision threshold Solid lines are optimised threshold 5 Gbaud 6 levels : Penalty @e-5.5 0.5 0-60 -55-50 -45-40 -35-30 -5-0 e-5 BER assumes FEC Laser RIN (db/hz) Levels 4 8 6 For e-5-6 -35-43 -49 7
Power Penalty (db) 3.5 levels : Penalty @e- 4 levels : Penalty @e- 8 levels : Penalty @e- Dotted lines are with a fixed decision threshold Solid lines are optimised threshold 5 Gbaud 6 levels : Penalty @e-.5 0.5 0-60 -55-50 -45-40 -35-30 -5-0 For e- BER Laser RIN (db/hz) Levels 4 8 6 For e- -30-4 -48-55 8
9
log(ber) - 8 levels 6 levels -3 levels : CD 0ps/nm levels : CD 00ps/nm levels : CD 50ps/nm -4-5 4 levels levels : CD 300ps/nm 4 levels : CD 0ps/nm 4 levels : CD 0ps/nm 4 levels : CD 60ps/nm 4 levels : CD 00ps/nm -6-7 -8 levels 4 levels : CD 0ps/nm 8 levels : CD 0ps/nm 8 levels : CD 0ps/nm 8 levels : CD 0ps/nm 8 levels : CD 40ps/nm 8 levels : CD 50ps/nm -9-0 - - 6 levels :CD 0ps/nm 6 levels :CD 0ps/nm 6 levels :CD 30ps/nm 6 levels :CD 40ps/nm Responsivity = 0.8 A/W - -0-9 -8-7 -6-5 -4-3 - - -0-9 -8-7 -6-5 -4-3 - Thermal - 0noise = 5pA/ Hz Extinction ratio 99 db Received power (dbm) Tx filter : 5GHz st order Bessel Rx Filter : GHz 4 th order Bessel Higher order formats are significantly more sensitive to CD 0
Power penalty (db) 3.5 5.8 Gbaud.5 levels 4 levels 8 levels 6 levels Levels CD ( db penalty @e-5 BER) 80 4 00 8 50 6 30 0.5 0 0 50 00 50 00 50 CD (ps/nm) Higher order formats are significantly more sensitive to CD Not an issue for links in the 500m to km range
Single DGD element at 45 degrees to signal Ideal waveforms Infinite extinction ratio 5 pa/ Hz thermal noise No RIN Optimised receiver bandwidth 3
Power penalty (db) 3.5.5 levels 4 levels 8 levels 6 levels 0.5 0 0 5 0 5 0 5 30 DGD (ps) 4 8 6 DGD (ps) 8 8 4 4
Power penalty (db) 3.5.5 levels 4 levels 8 levels 6 levels 0.5 0 0 5 0 5 0 5 30 DGD (ps) Pete Anslow s data show 0.5 db penalty at ps rather than 7p 4 8 6 DGD (ps) 7 9 6 3 5
Lasers Levels Baud rate Required BW (GHz) Sensitivity @e-5 BER RIN for 0.5dB penalty @e-5 CD limited distance (km) Max DGD (ps) @e-5 DGD limited distance (km) 4 56 44-0.5-35 7.3 5.5 8.7 8 37.3 3-7.8-43 8.4 5.5 8.7 6 8 5-5. -49 9. 3.7 3.9 4 8 - -35 9.8.0 34.7 8 8.6 6-9.4-43 33.3. 35.0 4 8 0-6.8-6 53.7 6.6 78. DGD max S L PMD 3.75 (conservative) 0.5ps/ km (v.bad fibre) All laser solutions have ~5 ps peak DGD tolerance All laser solutions have ~0 ps peak DGD tolerance km (~.5ps DGD) seems possible with all formats 6
7
log(ber) - -3-4 -5-6 -7-8 -9-0 - - levels 4 levels 8 levels 8 levels : 6 GHz 8 levels : 8 GHz 8 levels : 0 GHz 8 levels : GHz 8 levels : 7 GHz 4 levels : 6 GHz 4 levels : 8 GHz 4 levels : 0 GHz 4 levels : GHz levels : 6 GHz levels : 8 GHz levels : 0 GHz levels : GHz levels : 3 GHz levels : GHz Responsivity = 0.8 A/W Thermal noise = 5pA/ Hz Extinction ratio 99 db Tx filter : 5GHz st order Bessel Rx Filter : GHz 4 th order Bessel - -0-9 -8-7 -6-5 -4-3 - - -0-9 -8-7 -6-5 -4-3 - - 0 Received power (dbm) Higher order modulation schemes are more sensitive to receiver bandwidth The same ISI causes a more significant closure in the sub-level eyes 8
Optimum Rx BW PAM 7.5 GHz (0.7x baudrate) PAM8 GHz (0.85x baudrate) PAM6 3. GHz (0.9x baudrate) 5 GHz 6 GHz 3 GHz Electrical Bandwidth (4 th order Bessel Filter ) 9
30
log(ber) - 4 levels 8 levels Responsivity = 0.8 A/W Thermal noise = 5pA/ Hz Extinction ratio 99 db Tx filter : 5GHz st order Bessel Rx Filter : GHz 4 th order Bessel -3-4 -5-6 -7-8 -9-0 - - levels levels : Extinction 8dB levels : Extinction 0dB levels : Extinction db levels : Extinction 4dB levels : Extinction 6dB 4 levels : Extinction 8dB 4 levels : Extinction 0dB 4 levels : Extinction db 4 levels : Extinction 4dB 4 levels : Extinction 6dB 8 levels : Extinction 8dB 8 levels : Extinction 0dB 8 levels : Extinction db 8 levels : Extinction 4dB 8 levels : Extinction 6dB - -0-9 -8-7 -6-5 -4-3 - - -0-9 -8-7 -6-5 -4-3 - - 0 Received power (dbm) Higher order formats have same sensitivity to extinction ratio Relatively small penalties for extinction ratios of -4 db 3
Lasers Levels Baud rate Required BW (GHz) Sensitivity @e-5 BER RIN for 0.5dB penalty @e-5 CD limited distance (km) Max DGD (ps) @e-5 DGD limited distance (km) 4 56 44-0.5-35 7.3 5.5 8.7 8 37.3 3-7.8-43 8.4 5.5 8.7 6 8 5-5. -49 9. 3.7 3.9 4 8 - -35 9.8.0 34.7 8 8.6 6-9.4-43 33.3. 35.0 4 8 0-6.8-6 53.7 6.6 78. All M-PAM options are feasible for <km transmission RIN is a critical parameter Higher receiver bandwidths are required for higher order modulations (~0.7x for PAM, ~0.86x for PAM8 and ~ 0.9x for PAM6) FEC is required 3
33
Beat product of each rail with a delayed version Depends upon polarisation alignment phase alignment of reflection In general we will have to budget for the worst case 34
Power Penalty (db) 4 3.5 3.5 PAM- @e-5 (VPI) PAM- @e- (VPI) PAM-4 @e-5 (VPI) PAM-4 @e- (VPI) PAM-8 @e-5 (VPI) PAM-8 @e- (VPI) PAM-6 @e-5 (VPI) PAM-6 @e- (VPI).5 0.5 0 5 0 5 30 35 40 45 50 MPI (db) Simulations assume interferometric noise looks Gaussian rather than eye closure Correct statistics for a single reflection will be top-hat shaped. Multiple refelections will tend to a Gaussian. See differences between models in Effects of Phase-to-Intensity Noise Conversion by Multiple Reflections on Gigabit-per-second DFB Laser Transmission Systems, J.L. Gimlett et al., JLT Vol.7, No.6, June 989 and Performance Implications of Component Crosstalk in Transparent Lightwave Networks, E.L. Goldstein et al., PTL Vol.6, No.5, May 994. 35
Power Penalty (db) 4 PAM- @e- (VPI) PAM-4 @e- (VPI) 3.5 PAM-8 @e- (VPI) PAM-6 @e- (VPI) 3 PAM- (J.King) PAM-4 (J.King) PAM-8 (J.King).5 PAM-6 (J.King).5 0.5 0 5 0 5 30 35 40 45 50 MPI (db) Estimates seem more conservative Based on eye closure penalty : P=-0log0(-4 0 EdB/0 ) It is difficult to say which is correct without measurements 36
With multiple reflection sources, the statistics tend towards a Gaussian distribution (Central limit theorem) Solution Paths to Limit Interferometric Noise Induced Performance Degradation in ASK/Direct Detection Lightwave Networks, P.J. Legg et al., JLT Vol.4, No.9, Sept 996 37
Potential Application Scenarios (taken from OIF000.63):. BB w/ no intermediate connectors. BB w/ intermediate connectors (one patch panel) 3. BB w/ 4 intermediate connectors (two patch panels) 4. BB w/0 intermediate connectors (inter-building, x patch panel / building) Tx/Rx + 0 connectors Tx/Rx + 4 connectors Tx/Rx + 0 connectors Source:OIF000.63 (Note: Not a likely NG 00G PMD application) 38
39
Tx R R R3 R4 R5 R6 Rx Main Path R6-R5 R6-R4 R6-R3 R6-R R6-R 5 Reflected Paths [ N(N-)/ ] R5-R4 R5-R3 R5-R R5-R ~ db higher than single Reflection?? R-R 40
Increase in effective reflection coeff. (db) 0.00 8.00 6.00 No loss between Graph for when all reflections are equal 4.00.00 0.00 8.00 0.5 db loss between 6.00 4.00.00 0.00 0 4 6 8 0 4 No. Reflection points All reflections equal, then eff reflection = single reflection + 0log ((N(N-)/) One reflection dominates, then eff reflection = single reflection + 0log(N-) 4
PAM8 PP ~ 0.5dB PAM6 PP > 4dB PAM8 PP ~ 0.5dB PAM6 PP ~ 0.8 db If all reflections are -7dB For points we have -7-7 = -54 db For 6 points we have -54 + = -4 db For points we have -54 + 8 = -36 db 4
R R R 3 R 4 R 5 R 6 Fix R=R6 = -7 db return loss Fix Path Penalty at 0.dB What performance is required from R-R5, assuming all connectors are equal? 0. db penalty @e-5 BER Connector Requirement for 0. db penalty @e-5 BER 4 connectors + TX/RX return loss 0 connectors + TX/RX return loss PAM- -8-8.3 -.6 PAM-4-37 -3.4-7.7 PAM-8-4 -6. -30.3 PAM-6-47 -3. -35. 43
There are four types of polish used on fiber connectors Physical Contact (PC)* Cable Connectivity per Corning App Note Super Physical Contact (SPC)* Ultra Physical Contact (UPC)* Angled Physical Contact (APC)* Example: Commercially Available SMF MPO Patch, 55dB RL (UPC) *Note: Some references assign the P to mean polish rather than physical http://awapps.commscope.com/catalog/systimax/ product_details.aspx?id=3396&tab= Corning Source = Considerations for Optical Fiber Termination, AEN 89, Revision ; http://catalog.corning.com/corningcablesystems/media/resource_ Documents/application_engineering_notes_rl/AEN089.pdf -6 db appears to be a practical worst case for non APC connector 44
Higher order PAM modulations are more sensitive to MPI Initial analysis indicates that while this is an effect that needs to be considered, it does not appear to be a show stopper Even restricting the MPI penalty to 0.dB still only requires a fairly practical connector return loss of ~ -6 db (i.e. non APC) for PAM8. PAM6 is a little more challenging requiring a connector return loss of ~ -3dB. This initial analysis is thought to be conservative, and additional work is needed to further refine the results 45
Thank you.
47
Receive voltage (V) DAC v d t MZ Rx v rx t cos vd V t /electrode length.00 0.86 e.g. 8 levels For M levels and symbol voltage d k = {0...M-}: 0.7 0.57 vd t V acos d M k 0.43 0.9 0.4 0.00-0. 0 0. 0.4 0.6 0.8. Normalised MZ drive (V/Vpi) 48
DAC v d t MZ Rx v rx t cos vd V t /electrode length 49
50
- - 0 3 4 5 For M levels, the Symbol Error ratio is M P e M P 5
- - 0 For a single rail: N x N P e dx erfc N Symbol Error ratio : P e M M erfc M P N av Responsivity + Average power Bit Error ratio P e log M M M erfc M P N av Noise 5
53
I d - - 0 3 4 5 We can calculate the total noise on each rail TOTAL m N th RIN th f I m m I d RIN f I d I m P av m M Both thermal noise and RIN noise change at same rate with f Penalty will not depend upon baud rate 54
Lowest outer rail Middle rails Highest outer rail I 0 I m I M - - 0 - - 0 Decision threshold depends upon noise of each rail: Bit error ratio: P b m M M log log M M erfc erfc I TOTAL I m d TOTAL Id m Im m - - 0 I d I d I d I d m=0 m=...m- m=m- I d m P b m M m I m 0 m P b m m I m m I d m m I m m Not on highest rail m m I m Not on lowest rail 55