Neural Networks-Based Equalizers for Coherent Optical Transmission: Caveats and Pitfalls

We propose a deep learning-based phase retrieval method to accurately reconstruct the optical field of a singlesideband minimum-phase signal from the directly detected intensity waveform. Our method relies on a fully convolutional Neural Network (NN) model to realize non-iterative and robust phase retrieval. The NN is trained so that it performs fullfield reconstruction and jointly compensates for transmission impairments. Compared to the recently proposed Kramers-Kronig (KK) receiver, our method avoids the distortions introduced by the nonlinear operations involved in the KK phase-retrieval algorithm and hence does not require digital upsampling. We validate the proposed phase-retrieval method by means of extensive numerical simulations in relevant system settings, and we compare the performance of the proposed scheme with the conventional KK receiver operated with a 4-fold digital upsampling. The results show that the 7% hard-decision forward error correction (HD-FEC) threshold at BER 3.8e-3 can be achieved with up to 2.8 dB lower carrier-to-signal power ratio (CSPR) value and 1.8 dB better receiver sensitivity compared to the conventional 4-fold upsampled KK receiver. We also present a comparative analysis of the complexity of the proposed scheme with that of the KK receiver, showing that the proposed scheme can achieve the 7% HD-FEC threshold with 1.6 dB lower CSPR, 0.4 dB better receiver sensitivity, and 36% lower complexity.

Section: Resultsmentioning

confidence: 99%

Deep Learning-Based Phase Retrieval Scheme for Minimum-Phase Signal Recovery

Orsuti

Antonelli

Chiuso

et al. 2023

“…Finally, the P-NN has also been investigated in a WDM multi-channel optical transmission system. Neural networks have the fitting and prediction capability, therefore, the NN nonlinear equalization using the pseudorandom bit sequence (PRBS) as the training data may show an overfitting effect [66]- [68]. In this work, independent random binary sequences were utilized to assess the performance of the NN.…”

Section: Discussionmentioning

confidence: 99%

Intra-Channel Nonlinearity Mitigation in Optical Fiber Transmission Systems Using Perturbation-Based Neural Network

Ding

Liu

et al. 2022

In this work, a perturbation-based neural network (P-NN) scheme with an embedded bidirectional long short-term memory (biLSTM) layer is investigated to compensate for the Kerr fiber nonlinearity in optical fiber communication systems. Numerical simulations have been carried out in a 32-Gbaud dualpolarization 16-ary quadrature amplitude modulation (DP-16QAM) transmission system. It is shown that this P-NN equalizer can achieve signal-to-noise ratio improvements of ~1.37 dB and ~0.80 dB, compared to the use of a linear equalizer and a single step per span (StPS) digital back propagation (DBP) scheme, respectively. The P-NN equalizer requires lower computational complexity and can effectively compensate for intra-channel nonlinearity. Meanwhile, the performance of P-NN is more robust to the distortion caused by equalization enhanced phase noise (EEPN). Furthermore, it is also found that there exists a tradeoff between the choice of modulation format and the nonlinear equalization schemes for a given transmission distance.

“…two benchmarks: (ii) one for the complexity provided by the CDC and (ii) one for nonlinear mitigation given by DBP, where we used the implementation described in [33]. Our primary goal was to assess the complexity of NN with respect to CDC, while guaranteeing a level of nonlinear compensation comparable to the one of the widely used DBP 8 .…”

Section: Assessment Of Performance Of Neural Network Based Equalizers...mentioning

confidence: 99%

Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation

Freire

Napoli

Costa

et al. 2023

In this paper, a new methodology is proposed that allows for the low-complexity development of neural network (NN) based equalizers for the mitigation of impairments in highspeed coherent optical transmission systems. In this work, we provide a comprehensive description and comparison of various deep model compression approaches that have been applied to feed-forward and recurrent NN designs. Additionally, we evaluate the influence these strategies have on the performance of each NN equalizer. Quantization, weight clustering, pruning, and other cutting-edge strategies for model compression are taken into consideration. In this work, we propose and evaluate a Bayesian optimization-assisted compression, in which the hyperparameters of the compression are chosen to simultaneously reduce complexity and improve performance. Next, this paper presents four distinct metrics (RMpS, BoP, NABS, and NLGs) that are discussed here that can be used to evaluate the amount of computing complexity required by various compression algorithms. These measurements can serve as a benchmark for evaluating the relative effectiveness of various NN equalizers when compression approaches are used. In conclusion, the trade-off between the complexity of each compression approach and its performance is evaluated by utilizing both simulated and experimental data in order to complete the analysis. By utilizing optimal compression approaches, we show that it is possible to design an NN-based equalizer that is simpler to implement and has better performance than the conventional digital back-propagation (DBP) equalizer with only one step per span. This is accomplished by reducing the number of multipliers used in the NN equalizer after applying the weighted clustering and pruning algorithms. Furthermore, we demonstrate that an equalizer based on NN can also achieve superior performance while still maintaining the same degree of complexity as the full electronic chromatic dispersion compensation block. We conclude our analysis by highlighting open questions and existing challenges, as well as possible future research directions.