Chromatic dispersion compensation via an all-optical perceptron

Abstract

We experimentally demonstrate the use of a feed-forward photonic neural network (PNN) for chromatic dispersion compensation in fiber transmission within IM-DD protocols. The PNN device is constituted by an 8-channel alloptical delayed complex perceptron integrated on a Silicon-On-Insulator platform. The PNN device is inserted after the transmitter and before the fiber, thus acting as a pre-compensator. The training is performed via a Particle Swarm Optimizer and aims to provide an open eye diagram at the end-of-line receiver. We observe a 5-order of magnitude Bit Error Rate reduction for -7 dBm of power at the receiver between bare and equalized transmission for 10 Gbps Non-Return-to-Zero signals in a 125 km fiber link (average excess loss of 15 dB). We also perform a study on the minimum number of channels in the PNN needed for full equalization. Overall, the experimental results validate our solution for channel equalization via a PNN with negligible latency and a power consumption of 250 mW on average.