Università di Parma Parametric-Gain Approach to the Analysis of DPSK Dispersion-Managed Systems A. Bononi, P. Serena, A. Orlandini, and N. Rossi Dipartimento di Ingegneria dell’Informazione, Università di Parma Viale degli Usberti, 181A, 43100 Parma, Italy e-mail: [email protected] Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 1/21 Università di Parma Milan Parma Rome Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 2/21 Outline Università di Parma Introduction State of the Art: BER tools in DPSK transmission The PG Approach: Key Assumptions Tools Results Conclusions Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 3/21 Introduction Università di Parma Amplified spontaneous emission (ASE) noise from optical amplifiers makes the propagating field intensity time-dependent even in constant-envelope modulation formats such as DPSK. Random intensity fluctuations, through self-phase modulation (SPM), cause nonlinear phase noise [1], which is the dominant impairment in single-channel DPSK. Most existing analytical models focus on the statistics of the nonlinear phase noise. [1] J. Gordon et al., Opt. Lett., vol. 15, pp. 1351-1353, Dec. 1990. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 4/21 State of the Art Università di Parma K.-Po Ho [2] computed the probability density function (PDF) of nonlinear phase noise and derived a BER expression for DPSK systems with optical delay demodulation. Very elegant work, but: model assumes zero chromatic dispersion (GVD) does not account for the impact of practical optical/electrical filters on both signal and ASE Tx SPM only Matched filter [2] K.-Po Ho, JOSAB, vol. 20, pp. 1875-1879, Sept. 2003. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 5/21 State of the Art Università di Parma Wang and Kahn [3] computed the exact BER for DPSK (but provided no algorithm details) using Forestieri’s Karhunen-Loeve (KL) method [4] for quadratic receivers in Gaussian noise : Model accounts for impact of practical optical/electrical filters on both signal and ASE ....but ignores nonlinearity: it concentrates on GVD only. Tx OBPF LPF no SPM [3] J. Wang et al., JLT, vol. 22, pp. 362-371, Feb. 2004. [4] E. Forestieri, JLT, vol. 18, pp. 1493-1503, Nov. 2000. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 6/21 The PG Approach Università di Parma Also our group [5] computed the BER for DPSK using Forestieri’s KL method. Our model: besides accounting for impact of practical optical/electrical filters also accounts for the interplay of GVD and nonlinearity, including the signal-ASE nonlinear interaction using the tools developed in the study of parametric gain (PG) is tailored to dispersion-managed (DM) long-haul systems N Tx OBPF LPF [5] P. Serena et al., JLT, vol. 24, pp. 2026-2037, May 2006. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 7/21 DPSK DM System Università di Parma DPSK RX N in-line Tx OBPF post pre D A LPF Dispersion Map KL method requires Gaussian field statistics at receiver (RX), after optical filter Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 8/21 Why Gaussian Field? Università di Parma At zero dispersion, PDF of ASE RX field before OBPF is strongly non-Gaussian [2] …but with some dispersion, PDF contours become elliptical Gaussian PDF Im[E] Im[E] 0412 D= 3 ps/nm/km in-line Re[E] D Re[E] Din =0 Single span OSNR= 25 dB/0.1nm FNL = 0.15p rad [2] K.-Po Ho, JOSAB, vol. 20, pp. 1875-1879, Sept. 2003. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 9/21 Why Gaussian Field? Università di Parma Even at zero dispersion...PDF of ASE RX field AFTER OBPF Gaussianizes [6] before OBPF Iafter OBPF, Bo=10 GHz Red: Monte Carlo (MC) Blue: Multicanonical MC (MMC) OSNR=10.8 dB/0.1 nm, FNL=0.2p, ASE BW BM=80 GHz [6] A. Orlandini et al., ECOC’06, Sept. 2006. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 10/21 Why Gaussian Field? Università di Parma Reason is that a white ASE over band BM remains white after SPM h(t) w(t) SPM n(t) OBPF n( t ) w( )h( t )d If optical filter bandwidth Bo << BM, n(t) is the sum of many comparable-size independent samples Central Limit Theorem Xi’an, Oct. 23, 2006 Gaussian whatever the input noise distribution A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 11/21 Università di Parma Having shown the plausibility of the Gaussian assumption for the RX field, it is now enough to evaluate its power spectral density (PSD) to get all the needed information, to be passed to the KL BER routine. A linearization of the dispersion-managed nonlinear Schroedinger equation (DM-NLSE) around the signal provides the desired PSDs, according to the theory of parametric gain. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 12/21 Università di Parma Linear PG Model Linearized NLSE CW Rx ASE is Gaussian CW t t Small perturbation [7] C. Lorattanasane et al., JQE, July 1997 [8] A. Carena et al., PTL, Apr. 1997 [9] M. Midrio et al., JOSA B, Nov. 1998 DM, finite N spans [5] P. Serena et al., JLT, vol. 24, pp. 2026-2037, May 2006. Xi’an, Oct. 23, 2006 DM, infinite spans A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 13/21 Linear PG Model Università di Parma No pre-, post-comp. Red : quadrature ASE » Blue: in-phase ASE Parametric Gain = Gain (dB) over white-ASE case due to Parametric interaction signal-ASE Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 14/21 Limits of Linear PG Model Università di Parma linear PG model (dashed) versus Monte-Carlo BPM simulation (solid) FNL= 0.55 p rad, D=8 ps/nm/km, Din=0 /0.1 nm Xi’an, Oct. 23, 2006 /0.1 nm A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 15/21 Università di Parma @ PG doubling strengths for 10 Gb/s NRZ end-line OSNR (dB/0.1nm) 1.4 21 F [rad/p] NL 1.2 DM systems with Din=0. ( N>>1 spans) 19 17 15 1 0.8 For fixed OSNR (e.g. 15dB) in region well below red PG-doubling curve: Linear PG model holds ASE ~ Gaussian 0.6 0.4 0.2 00 0.2 0.4 0.6 0.8 Map strength S ( DR2 ) 1 [10] P.Serena et al., JLT, vol. 23, pp. 2352-2363, Aug. 2005. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 16/21 Our BER Algorithm Università di Parma Steps of our semi-analytical BER evaluation algorithm: 1. Rx DPSK signal obtained by noiseless BPM propagation (includes ISI from DM line) 2. ASE at RX assumed Gaussian. PSD obtained either from linear PG model (small FNL) or estimated off-line from Monte-Carlo BPM simulations (large FNL). Reference FNL for PSD computation suitably decreased from peak value to average value for increasing transmission fiber dispersion (map strength). 3. Data from steps 1, 2 passed to Forestieri’s KL BER evaluation algorithm, suitably adapted to DPSK. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 17/21 Results Università di Parma Check with experimental results [H. Kim et al., PTL, Feb. ’03] 10 Gb/s single-channel system, 6100 km NZDSF NRZ RZ-33% Theory Exp. Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 18/21 Results Università di Parma R=10 Gb/s single-channel, 20100 km, D=8 ps/nm/km, Din=0. OSNR=11 dB/0.1 nm, Bo=1.8R Noiseless optimized Dpre, Dpost NRZ-DPSK 1E-9 BER 1E-4 1E-2 NRZ-OOK RZ-DPSK 50% Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 19/21 Results Università di Parma 10 Gb/s single-channel system, 20100 km, Din=0. Bo=1.8R . Noiseless optimized Dpre, Dpost. DPSK-NRZ DPSK-RZ (50%) @ D=8 ps/nm/km PG no PG ΦNL=0.5p ΦNL=0.5p ΦNL=0.3p ΦNL=0.1p Strength ( DR2) Xi’an, Oct. 23, 2006 ΦNL=0.3p Strength ( DR2) A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 20/21 Conclusions Università di Parma Novel semi-analytical method for BER estimation in DPSK DM optical systems. The striking difference between OOK and DPSK is that in DPSK PG impairs the system at much lower nonlinear phases, when the linear PG model still holds. Hence for penalties up to ~3 dB one can use the analytic ASE PSDs from the linear PG model instead of the time-consuming off-line MC PSD estimation. Hence our mehod provides a fast and effective tool in the optimization of maps for DPSK DM systems. More information on our work: www.tlc.unipr.it Xi’an, Oct. 23, 2006 A. Bononi, China-Italy Workshop Photon. Commun. & Sens. 21/21