Neural Networks for PAPR Reduction in Optical OFDM Signal Transmission

In recent years, with the improvement of computing performance, the application of neural networks in communication field has been widely studied. In this paper, we propose and apply neural networks (NNs) to improve the transmission performance of optical orthogonal frequency division multiplexed (O-OFDM) signals. A pair of NNs is added in transmitter and receiver respectively, to modify and restore the QAM constellation. The main target for NNs is to reduce the peak to average power ratio (PAPR) of the O-OFDM signals and the bit error rate (BER) of the received signals. The proposed scheme is verified by the simulations. And a transmission of 18.8 Gb/s, 16-QAM optical OFDM signals is successfully demonstrated and analyzed for the cases with and without NNs. The experimental results show that, the PAPR and BER can be improved significantly (~4 dB for BER) after using NNs, since PAPR is reduced and more importantly the channel distortion and noises are effectively compensated by NNs via learning procedure.