Noise Equalization Scheme Based on Probabilistic Shaping and 1D Convolutional Bidirectional Long Short-Term Memory Neural Network for Dual-Polarization NFDM System with Hermite-Gaussian Subcarriers

In order to effectively reduce the impact of the amplified spontaneous emission (ASE) noise and processing noise for the dual-polarization nonlinear frequency division multiplexing system with Hermite-Gaussian subcarriers (DP-HG-NFDM), this paper proposes a noise equalization scheme based on probabilistic shaping (PS) and one-dimensional convolutional-bidirectional long short-term memory neural network (1D CNN-BiLSTM). Simulation results demonstrate that under the condition of equal information entropy, compared with the uniform 32QAM scheme, the 64QAM scheme based on PS and 1D CNN-BiLSTM can achieve a Q-factor gain of 2.98 dB and a spectral efficiency (SE) improvement of 0.39 bit/s/Hz (0.195 bit/s/Hz per polarization) in transmission scenarios with different launch power and a distance of 960 km. Furthermore, under the condition of meeting the 7% hard-decision forward error correction (HD-FEC) threshold, the transmission distance of this scheme is extended by 400 km.