Fast and Accurate Optical Fiber Channel Modeling Using Generative Adversarial Network
Hang Yang,Zekun Niu,Shilin Xiao,Jiafei Fang,Zhiyang Liu,David Faninsin,Lilin Yi
DOI: https://doi.org/10.1109/JLT.2020.3037905
2022-01-18
Abstract:In this work, a new data-driven fiber channel modeling method, generative adversarial network (GAN) is investigated to learn the distribution of fiber channel transfer function. Our investigation focuses on joint channel effects of attenuation, chromic dispersion, self-phase modulation (SPM), and amplified spontaneous emission (ASE) noise. To achieve the success of GAN for channel modeling, we modify the loss function, design the condition vector of input and address the mode collapse for the long-haul transmission. The effective architecture, parameters, and training skills of GAN are also displayed in the paper. The results show that the proposed method can learn the accurate transfer function of the fiber channel. The transmission distance of modeling can be up to 1000 km and can be extended to arbitrary distance theoretically. Moreover, GAN shows robust generalization abilities under different optical launch powers, modulation formats, and input signal distributions. Comparing the complexity of GAN with the split-step Fourier method (SSFM), the total multiplication number is only 2% of SSFM and the running time is less than 0.1 seconds for 1000-km transmission, versus 400 seconds using the SSFM under the same hardware and software conditions, which highlights the remarkable reduction in complexity of the fiber channel modeling.
Information Theory,Machine Learning,Signal Processing