1. Electronic Dispersion Compensation for optical communication systems NEC Labs America July 12 th, 2006

2. Techniques for EDC FFE/DFE FFE/DFE Single Side Band (SSB) with Microstripe Single Side Band (SSB) with Microstripe MLSE MLSE Volterra Series Volterra Series Pre-distortion (including line-coding) Pre-distortion (including line-coding) Hybrids Hybrids

3. FFE and DFE Standard techniques for equalization of linear channels Standard techniques for equalization of linear channels Easiest to implement, but worst performance compared to other techniques in linear systems Easiest to implement, but worst performance compared to other techniques in linear systems Performance further degraded with nonlinear systems such as the optical channel Performance further degraded with nonlinear systems such as the optical channel DL (ps/nm) OSNR (dB) CriterionAuthorRemarks 224022.7BER=1e-9 Watts (UCL) NRZ, 5-tap FFE + 2- stage DFE, optimized for min. BER Experiments:

4. SSB transmission SSB preserves phase information, making the whole system linear SSB preserves phase information, making the whole system linear Compensation achieved by using Microstrip lines Compensation achieved by using Microstrip lines DL (ps/nm) OSNR (dB) CriterionAuthorRemarks 2040(High)?BER=1e-10 Watts (UCL) NRZ, Optical filter to generate SSB 5440(High)?BER=1e-9Sieben (TRLabs, Canada) NRZ, MZ Modulator to generate SSB Experiments:

5. MLSE Theoretically optimal detector (with infinite states and continuous sampling). Practically, 2 Sa/b is good enough. Memory depends on amount of Dispersion Theoretically optimal detector (with infinite states and continuous sampling). Practically, 2 Sa/b is good enough. Memory depends on amount of Dispersion Need correct noise statistics, but usually use Gaussian approximation Need correct noise statistics, but usually use Gaussian approximation DL (ps/nm) OSNR (dB) CriterionAuthorRemarks 357014BER=1e-3 Napoli (Italy) NRZ, 2 Sa/b, Memory=4, Gaussian noise stat. 350013BER=3e-4 Elbers (CoreOptics) ODB, 2 Sa/b, Memory=2, ? Noise stat. 300018BER=1e-3 Farbert (CoreOptics NRZ, 2 Sa/b, Memory=2, ? Noise stat. Experiments:

6. Volterra Series Taylor series approximation of nonlinear systems with memory Taylor series approximation of nonlinear systems with memory FFE/DFE with taps that are coefficients of nonlinear terms FFE/DFE with taps that are coefficients of nonlinear terms Performance close to MLSE, but hard to optimize tap coefficients Performance close to MLSE, but hard to optimize tap coefficients DL (ps/nm) OSNR (dB ) CriterionAuthorRemarks 459013BER=1e-3 Xia (Kiel, Germany) ODB, 4-tap FFE + 1- stage DFE+ 2 nd order nonlinear terms, tap weights optimized for MMSE Experiments:

7. Pre-distortion (1): Pulse Shaping Modulation formats such as duobinary are line coding techniques that can be thought of as a kind of Pre-distortion Modulation formats such as duobinary are line coding techniques that can be thought of as a kind of Pre-distortion DL (ps/nm) CriterionAuthor 3315BER=1e-10 Price (Alcatel) 3500BER=1e-10 Kuwano ( NTT Japan) 3366BER=1e-10 Pierre (Alcatel) Note: Dispersion values correspond to no power penalty compared to back-to-back Note: Dispersion values correspond to no power penalty compared to back-to-back All Duobinary simulations/experiments show optimal system performance at certain values of dispersion, probably because duobinary spectrum partially cancel the dispersion effect All Duobinary simulations/experiments show optimal system performance at certain values of dispersion, probably because duobinary spectrum partially cancel the dispersion effect Simulations show more complicated line coding further enhance dispersion tolerance (Forestieri & Prati, JLT Nov. 2001) Simulations show more complicated line coding further enhance dispersion tolerance (Forestieri & Prati, JLT Nov. 2001) Experiments:

8. Predistortion (2): Inverting the dispersive Channel EDC at transmitter to exactly cancel the channel effects (analogous to Pre-chirping). The non-linearity of photo-detector becomes a non- issue EDC at transmitter to exactly cancel the channel effects (analogous to Pre-chirping). The non-linearity of photo-detector becomes a non- issue Requires knowledge of the channel. Not a problem for CD since its relatively static. For PMD, might not be practical since it requires regular channel feedback Requires knowledge of the channel. Not a problem for CD since its relatively static. For PMD, might not be practical since it requires regular channel feedback DL (ps/nm) OSNR (dB) CriterionAuthor 6000010BER=1e-3 McNicol (Nortel) 8243310.5BER=3e-3 McGhan (Nortel) Experiments:

9. Hybrids: Pre-distortion and electronic processing at receiver Most common form of hybrid investigated is Duobinary modulation with FFE/DFE/MLSE in IM/DD channels. With Duobinary modulation, FFE/DFE at receiver didn’t help while MLSE still provides benefits. Most common form of hybrid investigated is Duobinary modulation with FFE/DFE/MLSE in IM/DD channels. With Duobinary modulation, FFE/DFE at receiver didn’t help while MLSE still provides benefits. In coherent systems with nonlinear effects, Savory (UCL, Elec. Lett. March 2006) showed through simulations that half pre-distortion – half receiver EDC optimizes performanceIn coherent systems with nonlinear effects, Savory (UCL, Elec. Lett. March 2006) showed through simulations that half pre-distortion – half receiver EDC optimizes performance DL (ps/nm) OSNRCriterionAuthorRemarks 350013BER=3e-4 Elbers (CoreOptics) ODB, MLSE with 2 Sa/b, Memory=2, ? Noise stat. 431213BER=1e-3 Gnauck (Bell Labs) ODB, MLSE with 2 Sa/b, Memory=2, ? Noise stat. Experiments: