From experimental tests to From experimental tests to link
Transcription
From experimental tests to From experimental tests to link
From experimental tests to link engineering: an operator perspective ECOC 2014 Orange Labs & Networks JeanJean-Luc Augé Tests protocol Determination of the linear and nonlinear parameters: Transponder back to back Anl parameter Introduction Why testing transponders? • Performance comparison of different vendors solution, beyond the marketing story. However, transponder performance is only a part of the system performance! • Build an engineering expertise independent of vendor design tools: - “what if ?” analysis, network evolution anticipation, specific use cases… - Design verification - Alien wavelengths engineering: there is little support from vendors on alien engineering and performance prediction : with good in house expertise, the operator can save on field trials. From experiment to simulations: We will focus on the exploitation of experimental results with the GNLI model • Determination of the non linear parameter Anl • Discussion on the relative importance of Anl and OSNR sensibility Test bed Orange Lab generic testbed: not vendor specific, up to 20x100km DCU free transmission 35 x 100G@ 28Gbd 35 x 100G@ 28Gbd 100 km 100G under evaluation 100 km 100 km 100 km 100 km 100 km 100 km 100 km 100 km 100 km 100 km OSA OSA 100 km Tx DGE DGE WSS 100G channels comb 100 km 100 km 100 km 100 km 100 km 100 km 100 km 100 km Rx Transponder under test in its shelf and OADM termination Client tester ASE Vendor system manager BER / Q reading Vendor independent post FEC errors checking Stress the system to noise and non linear limits: non linear measurements • @ different amplifier powers (1dB steps) • @ different OSNR (noise loading in 1dB steps) • Record the resulting BER (Q) / OSNR vs Pch abacus Results exploitation with the GNLI model Consider the NLI noise as gaussian and additive PNLI = Anl ⋅ Pch3 or ε PNLI = anl ⋅ N span Pch3 SNR = Pch PASE + PNLI P. Poggiolini et al., “Analytical modeling of non-linear propagation in uncompensated optical transmission links”, PTL 2011 Anl & ε are determined from experimental results Anl = f ( Pchamplifier , OSNR OSA , QvendorTRX ) amplifier tuning Anl (dB) for 10, 15 & 20 spans 10 spans Pch=0.2dBm Pch=1.2dBm Pch=2.2dBm Pch=3.3dBm 2 2 1,1dB 0 1,9dB -2 -4 0 5 10 relative OSNR (dB) Anl (dB) 15 spans Anl vs span count 1,0dB relative Anl (dB) 20 spans measured +noise loading 0 ε ~ 1,4 -2 vs GNLI in Nyquist WDM: ε~1 -4 10 11 12 N spans (dB) Anl uncertainty and parameters correction: see backup slides SNRlin ∆SNRdB ≈ − 1 ⋅ ∆AnldB OSNRlin Mo.4.3. modeling & system design session: P. Poggiolini & al, Impact of Low-OSNR Operation on the Performance of Advanced Coherent Optical Transmission Systems 13 Tests results exploitation: what parameters can we trust? -Anl measurement uncertainty -little Anl variation between transponders and in the line (mixed transponders) ⇒ Is Anl a good performance indicator to compare transponders ? Anl vs Non Linear Threshold (NLT) and amplifier tuning 20 spans, Q vs Pch at different OSNR levels Q (dB) 11,0 Q for 21dB SL 10,0 Q for 24dB SL 9,0 PchNLT is defined as the optimum channel power at a fixed number of spans N 8,0 PchNLT dependency with OSNR 7,0 6,0 5,0 -3 -2 -1 0 1 2 3 PchNLT = 4 1 2 ⋅ Anl ⋅ OSNR0 5 Pch (dBm) ⇒ OSNR0 (B2B transponder response) is as much as Anl contributing to the NLT ⇒ What about amplifiers tuning? Amplifiers are not tuned to PchNLT but to the power required for maximum reach at a given (fixed) span loss and can be derived from a simple calculation: We introduce the following variables and consider anl dependency with span count • • • • Anl = anl ⋅ N ε (OSNR0,Q0) defined as the system minimum Q for the operator: Q0 = FEClimit+operatormargin. Q0 is also used to define the transponder NLT. ∆margin are the system margins OSNRT is the system OSNR target: OSNRT=OSNR0+∆margin span ASE contribution Aspan = NF ⋅ SL ⋅ hυ∆υ We look at the power engineering Pch required to achieve maximum reach: 2+ε N max = Pch 2 +ε ε∈ ~[1 - 1,5] 1 1 ⋅ 2anl ⋅ OSNR03 (Aspan ⋅ ∆ m arg in )2 ⇒ weak dependency of the reach with anl OSNR0ε −1 ε = ⋅ (Aspan ⋅ ∆ m arg in ) 2anl ⇒ weak dependency of the power settings with OSNR sensibility Illustration over an example: 100G SDFEC vs HDFEC Example of two imaginary transponders with • • • HDFEC & SDFEC back top back Q (dB) 12 11 10 • Q B2B HDFEC 9 8 ⇒ Same Anl but different NLT ⇒ Performance difference almost equal to the OSNR difference ! ⇒ Anl is only part of the story… HDFEC limit Q B2B SDFEC 7 SDFEC limit 6 12 13 14 15 16 OSNR (dB) 17 18 19 the same back to back Q/OSNR response same Anl and ε = 1,3 (not withstanding baud rate difference) but different FEC limit (OSNR sensitivity) : 8,5dB Q (HDFEC) & 6,5dB Q (SDFEC). we consider operation at 1,5dB Q margin (just a use case) N spans NLT (given by PchNLT) maximum span loss for a given reach => not an engineering comparison Large NLT improvement ( ½ ∆OSNRlimit) fixed 20 spans reach (variable SL) Q (dB) line amplifiers tuning (Pch) (previous calculation) maximum reach for the same span loss => real engineering comparison Similar Pch: can be mixed with the same engineering! fixed 21dB span loss (variable reach) Q HDFEC 21dB SL 11,0 Q (dB) 11,0 Q HDFEC 20 spans 10,0 Q SDFEC 30 spans Q SDFEC 24dB SL 10,0 Q B2B HDFEC 9,0 HDFEC limit Q B2B SDFEC 8,0 SDFEC limit HDFEC limit 1dB ∆P vs 2dB Q 7,0 6,0 almost same power 8,0 requirement 7,0 SDFEC limit 6,0 5,0 5,0 -3 -2 -1 Q SDFEC 20 spans, 24dB SL 9,0 0 1 2 Pch (dBm) 3 4 5 -3 -2 -1 0 1 2 Pch (dBm) 3 4 5 wrap up & discussion A set of (Anl, ε) constants may be determined by experiments and used to compare transponders from different vendors in the same conditions Is it a key differentiator? considering: • • similar anl between modulation formats (BPSK, QPSK, 16QAM…) at same baud rate anl baud rate dependency but today transponders share similar baud rate (from 28Gbd to 32Gbd for 100Gb/s) and are mixed over the same line, averaging the differences (XPM from all transponders) Reach & power settings: N 2 +ε Pch 1 1 = ⋅ 2anl ⋅ OSNR03 (Aspan ⋅ ∆ m arg in )2 2 +ε ε −1 OSNR0 ε = ⋅ (Aspan ⋅ ∆ m arg in ) 2anl N3 = if ε ~> 1 (Nyquist WDM?) 1 1 ⋅ 2anl ⋅ OSNR03 (Aspan ⋅ ∆ m arg in )2 Pch3 = 1 ⋅ Aspan ⋅ ∆ m arg in 2anl no Pch dependency with OSNR0 (transponder B2B): ⇒ Same power settings for all 100G QPSK / 200G 16QAM transponders in the network Strong dependency of maximum reach to OSNR0, whereas Anl is a weaker contributor ⇒ The linear back to back response is a powerful indicator to compare the same type of transponders from different vendors Not withstanding different fiber types or transponder with significant baud rate difference Non linear post compensation (back propagation) could also change this statement BACKUP Anl uncertainty Pch correction Can we trust the results? Q factor Chose a unique value for Anl to cover all noise and power conditions for a given span count Fit over N spans ⇒ good fit with |∆Q|<0.2dB for 1dB Anl spread Q response over N spans and fit ⇒ a quick uncertainty calculation confirms that the need for Anl precision is moderate B2B 13,5 Pch=0.2dBm Pch=1.2dBm linearly: Pch=2.2dBm dSNR SNR dAnl = − 1 ⋅ SNR OSNR Anl Pch=3.3dBm B2B fit 0.2dBm NLI fit 1.2dBm NLI fit At non linear threshold (dSNR/dP = 0), SNR/OSNR = 0,66, hence in dB (at low dAnl/Anl values): 2.2dBm NLI fit 5 3.3dBm NLI fit 10 15,7 21,4 OSNR ∆SNR ≈ −0.33 ⋅ ∆Anl ∆Q(dB) ≈ −0.33 ⋅ ∆Anl (dB) In return, measurement points at SNR~OSNR (low Pch or span #) are less precise for the determination of Anl: for example ∆Anl > 3∆Q below the non linear threshold. Weight each contributions (Pch, Q, OSNR measurement triplet) by its SNR/OSNR ratio and do not use data with OSNR/SNR < 1,3 (reminder OSNR/SNR=1,5 at NLT) ⇒ measure Anl at high power (better accuracy) ⇒ Use the model below NLT (less impact of Anl errors) Pch correction: Rx noise loading & power mode amplifiers We use power mode amplifiers in our labs PTOT = Psig+PASE = Cte, whereas in the field, smart gain mode amplifiers are used: ASE noise build up power and may contribute to excess NL We use Rx noise loading : little impact with power mode amplifiers, unlike if we were using gain mode amplifiers Pch include in line ASE noise and NLI: correction is needed • B is the signal bandwidth ratio to 0,1nm (~3) because NLI noise is mostly contained in signal bandwidth • B’ is calculated for a 90 channels amplifier bandwidth with only 60 channels in use (=90/60*0,4/0,1) Pch = PTOTamplifier / 60 Psig = Pch − B'⋅PASEligne − B ⋅ PNLI Consider Pch for NLI contribution because ASE and NLI noise contribute PNLI = Anl ⋅ Pch3 OSNRTOTAL = SNR = Mo.4.3. modeling & system design session: P. Poggiolini & al, Impact of Low-OSNR Operation on the Performance of Advanced Coherent Optical Transmission Systems Psig + B ⋅ PNLI The measured OSNR with OSA include some PNLI noise! PASEtotal Psig PASEtotal + Anl ⋅ Pch3 correcting factor at low Q factor and low line OSNR Anl = 1 Pch 2 1 1 ⋅ − SNR OSNRTOTAL B' 1 + B + SNR OSNR ligne