Direct decoding of nonlinear OFDM-QAM signals using convolutional neural network.

Nonlinear Fourier transform, as a technique that has a great potential to overcome the capacity limit in fibre optical communication system, faces speed and accuracy bottlenecks in practice. Machine learning using convolutional neural networks shows great potential in NFT-based applications. We have developed a convolutional neural network for decoding information in NFT-based communication and numerically demonstrated its performance in comparison to a fast NFT algorithm. The comparison indicates the potential of conventional neural network to replace NFT calculations for decoding of information.