Opt Lett. 2018 Aug 1;43(15):3542-3545. doi: 10.1364/OL.43.003542.
We propose a novel radial basis function neural network (RBF-NN)-based nonlinear equalizer (NLE) for the intensity modulation/direct detection (IM/DD) transmission. After optimizing input characteristics of the RBF-NN, we experimentally demonstrate C-band 4×50 Gb/s four-level pulse-amplitude modulation (PAM-4) transmission over 80 km standard single-mode fiber (SSMF), using 18 GHz direct-modulated lasers and dispersion compensation fiber. We demonstrate that the RBF-NN-based NLE outperforms the commonly used Volterra filter equalizer and recently proposed multilayer perceptron (MLP)-based NLE by about 4.5 and 1.5 dB improvements of receiver sensitivity at the 7% forward error correction threshold, respectively. Furthermore, we identify that both the training stability and fitting ability of the RBF-NN-based NLE are better than those of the MLP-based NLE.
我们提出了一种基于径向基函数神经网络（RBF-NN）的新型非线性均衡器（NLE），用于强度调制/直接检测（IM/DD）传输。在优化 RBF-NN 的输入特性后，我们使用 18GHz 直接调制激光器和色散补偿光纤，在 80km 标准单模光纤（SSMF）上实验演示了 C 波段 4×50 Gb/s 四电平脉冲幅度调制（PAM-4）传输。我们证明，基于 RBF-NN 的 NLE 比常用的 Volterra 滤波器均衡器和最近提出的基于多层感知器（MLP）的 NLE 分别提高了约 4.5dB 和 1.5dB 的接收灵敏度，在 7%前向纠错阈值下。此外，我们发现基于 RBF-NN 的 NLE 的训练稳定性和拟合能力均优于基于 MLP 的 NLE。
