Improved convolutional neural network-based symbol synchronization scheme in MB-OFDM UWBoF system

In this paper, a symbol synchronization scheme based on improved Convolutional Neural Networks (CNN) for the multiband orthogonal frequency division multiplexing ultrawideband over-fiber (MB-OFDM UWBoF) system is proposed. CNN is used to learn the shape features of the synchronization metric peak to determine the synchronization point more accurately. Simulation results demonstrate that the accuracy curve of the proposed improved CNN structure is about five epochs faster than the traditional CNN after training when reaching the ideal accuracy. When the False Detection Probability (FDP) is 0.1, the optical receiver sensitivity of the system with the synchronization of improved CNN symbol is about 1.1 dB higher than that of the traditional synchronization scheme, and about 0.5 dB higher than the symbol synchronization scheme of traditional CNN. After 75 km transmit over fiber at 128 QAM, the system reaches the HD-FEC threshold of 3.8×10^-3 when the received optical power is -10.5 dBm.