A Fully Connected Neural Network Approach to Mitigate Fiber Nonlinear Effects in 200G DP-16-QAM Transmission System

Complexity Reduction over Bi-RNN-Based Nonlinearity Mitigation in Dual-Pol Fiber-Optic Communications via a CRNN-Based Approach

“…The two RX models in Fig. 1 were considered for equalization using the proposed CRNN model, as well as MLP, CNN+MLP, bi-RNN, bi-GRU, bi-LSTM, and CNN+bi-LSTM neural structures with similar architectures to [23], [24], [26] [25], [40], [27], respectively. The following subsections discuss the performance versus the complexity of these neural network structures in each RX model, and discuss an overall comparison between the two RX types.…”

“…Table . 2 presents the complexity of the implemented neural network models as the number of FLOPs they incur in the inference mode per output symbol. These values are obtained analytically according to the formulas discussed in Section 3 for the recurrent DBP-2StPS MLP [23] CNN+MLP [24] bi-RNN [26] bi-GRU [25] bi-LSTM [40] CNN(no-stride)+bi-LSTM [27] Proposed CRNN DBP-80StPS…”

“…A number of neural network structures have been investigated in the literature in this regard [23][24][25][26][27][28][29][30][31]. As universal approximators, multilayer perceptrons (MLPs) are among the first adopted models in this field.…”

“…As universal approximators, multilayer perceptrons (MLPs) are among the first adopted models in this field. A number of MLP-based adopted models for equalization are discussed in [23,[32][33][34]. However, as MLPs try to find the correlation among each pair of the samples in data repeatedly in each layer using fully-connected layers, they are prone to over-fitting due to the large number of trainable parameters.…”

POL-MUX System for Noncoherent Optical Networks

“…Nevertheless, it was not employed in optical data networks and real a comeback occurred with the development of coherent high-speed networks with direct detection system, digital signal processing and multistate modulation formats. Many groups were interested in this topic [14][15][16][17][18], e.g., Bermani et al studied synchronous demodulation of coherent 16-QAM with feedforward carrier recovery [19], El-Nahal et al described coherent 16 quadrature amplitude modulation (16QAM) Optical Communication Systems [20], Gnauck et al examined spectrally-efficient long-haul WDM transmission using 224-Gb/s polarizationmultiplexed 16-QAM [21]. Finally, direct detection systems with digital signal processing and X-QAM modulations were deployed into real high-speed transport optical networks, with high tolerance to negative dispersions effects [22].…”