Advanced Convolutional Neural Networks for Nonlinearity Mitigation in Long-Haul WDM Transmission Systems

Abstract

Practical implementation of digital signal processing for mitigation of transmission impairments in optical communication systems requires reduction of the complexity of the underlying algorithms. Here, we investigate the application of convolutional neural networks for compensating nonlinear signal distortions in a 3200 km fiber-optic 11x400-Gb/s WDM PDM-16QAM transmission link with a focus on the optimization of the corresponding algorithmic complexity. We propose a design that includes original initialisation of the weights of the layers by a filter predefined through the training a single-layer convolutional neural network. Furthermore, we use an enhanced activation function that takes into account nonlinear interactions between neighbouring symbols. To increase learning efficiency, we apply a layer-wise training scheme followed by joint optimization of all weights applying additional training to all of them together in the large multi-layer network. We examine application of the proposed convolutional neural network for the nonlinearity compensation using only one sample per symbol and evaluate complexity and performance of the proposed technique. Index Terms-Convolutional neural networks. Nonlinearity mitigation in fiber-optic links.

Channel Arrangement Design in Lumped Amplified WDM Transmission over NZ-DSF Link with Nonlinearity Mitigation Using Optical Phase Conjugation

“…The WDM channel spacing was set to 100, 200, or 400 GHz. In this simulation, only five channels were simulated due to computational resources, but in reality, 10 channels could be implemented using OPC with a complementary spectral inversion configuration [5]. The transmission link consisted of 12×80-km NZ-DSFs and inline erbium-doped fiber amplifiers (EDFAs).…”

“…Thus, conventional optical transmission systems are operated with the optimal fiberinput power balanced between OSNR-improvement and signal distortion due to fiber nonlinearities. Although techniques for digitally mitigating fiber nonlinearity, such as digital back propagation [3], the Volterra equalizer [4], and the neural-network-based equalizer [5], have been developed, the deployment of novel transponders with these functions is required.…”

Perturbation Theory-Aided Learned Digital Back-Propagation Scheme for Optical Fiber Nonlinearity Compensation

“…Recently, a CNN structure has been employed to replace the DNN structure [27]. The CNNbased LDBP considers a block of symbols as the input, while the output is the equalized symbol corresponding to the center of the input block.…”

“…This approach has been also validated through experiments [25], [26]. More recently, a convolutional neural network (CNN) has been considered to replace the DNN structure in LDBP [27]. To summarize, the LDBP technique accomplishes the linear steps through the weight matrices operation in DNN or CNN, and the nonlinear steps through the nonlinear activation functions.…”

“…However, the differences between the proposed PA-LDBP and ESSFM in [24] and [27] are evident. Specifically, in [24], the focus is on the performance gain of LDBP that adopts ESSFM over the conventional DBP with ESSFM; while in [27], the aim is to improve the performance at each nonlinear step, and only 1 span per step is considered. In this work, we show that the benefits of our proposed PA-LDBP are fourfold:…”