Convolutional Neural Network Based Equalization for 112-Gbit/s High Speed Optical Link

In this paper, we proposed a convolutional neural network (CNN) based equalizer with 2 convolutional layers, 2 full-connected layers and 1 output layers. We experimentally demonstrate 112-Gbit/s pulse amplitude modulation (PAM) transmission based on a 30-GHz MZM over 2/10-km standard single mode fiber (SSMF) at 1550
$nm$
using the proposed equalizer. For PAM-4 signal, it improves the sensitivity by 2 dB when the bit-error-rate (BER) is 3.8e-3. For PAM-6 signal, BER can be lower than 2.5e-2 after being processing by the CNN-based equalizer.