Neural network-based electrical dispersion pre-compensation for a 56 Gb/s PAM-4 over an 80 km fiber in intensity-modulation and direct-detection systems.

A neural network (NN)-based electrical dispersion pre-compensation (pre-EDC) scheme in intensity-modulation and direct-detection (IM/DD) systems is proposed and experimentally demonstrated in this Letter. The scheme enables 56 Gbit/s four-level pulse amplitude modulation (PAM-4) generation at a transmitter over an 80 km single-mode fiber (SMF) transmission in the C-band. The NN is utilized to better fit nonlinear phase-amplitude transformation due to the chromatic dispersion (CD) in IM/DD systems, in place of the existing Gerchberg-Saxton (GS) iterative algorithm and linear GS-based finite impulse response (GS-FIR) non-iterative compensation schemes. The experimental results show that the measured bit error ratio (BER) can be reduced to below the 7% hard-decision forward error correction (HD-FEC) threshold of 3.8 × 10 with 0 dBm receiver optical power (ROP) by the NN-based non-iterative pre-EDC scheme, which also saves up to 81% of computational complexity compared to the GS-based scheme. The results indicate that our proposed scheme is promising for the CD pre-compensation at the transmitter.