光学学报, 2018, 38 (9): 0906002, 网络出版: 2019-05-09
基于广义回归神经网络的CO-OFDM系统非线性均衡 下载: 1156次
Nonlinear Equalizer Based on General Regression Neural Network in Coherent Optical OFDM System
光通信 相干光正交频分复用 大线宽 高阶正交幅度调制 广义回归神经网络 非线性均衡 optical communications coherent optical orthogonal frequency division mul large laser linewidth high-order quadrature amplitude modulation general regression neural network nonlinear equalization
摘要
针对高阶正交幅度调制和大线宽相干光正交频分复用(CO-OFDM)系统,提出了一种基于广义回归神经网络(GRNN)的非线性均衡算法。将接收端进行相位噪声恢复之后的批量数据作为训练数据样本,通过训练学习得到GRNN的唯一参数平滑因子,然后对测试数据进行非线性均衡。对传输速率为50 Gb/s,传输距离为100 km的CO-OFDM系统进行了仿真验证。仿真结果表明,在大线宽和高阶调制下,GRNN非线性均衡算法对系统非线性损伤的补偿效果优于相应反向传播神经网络(BPNN)非线性均衡算法,且其训练运行时间远小于BPNN。GRNN非线性均衡算法能极大促进CO-OFDM系统在中长距离光纤传输中的应用。
Abstract
A nonlinear equalization algorithm is proposed based on the general regression neural network (GRNN) in the coherent optical orthogonal frequency division multiplexing (CO-OFDM) system with high-order quadrature amplitude modulation and large laser linewidth. After phase recovery at the receiver, the training data is chosen to carry out the training and studying in the GRNN. In the process, the smoothing factor, the only parameter, can be decided in the GRNN. Then, for the detecting data at the receiver, the nonlinear equalization is performed by the GRNN. The numerical simulations have been completed by the proposed GRNN nonlinear equalization algorithm in the CO-OFDM system with a transmission rate of 50 Gb/s and a transmission distance of 100 km. Compared with the back propagation neural network nonlinear equalization (BPNN-NLE) algorithm, under lager laser linewidth and high-order quadrature amplitude modulation (QAM), the proposed method has a better nonlinear equalization performance and a shorter time of training running, which will greatly promote the application of CO-OFDM system in the fiber transmission with long and medium distance.
吴金达, 卢瑾, 任宏亮, 覃亚丽, 郭淑琴, 胡卫生. 基于广义回归神经网络的CO-OFDM系统非线性均衡[J]. 光学学报, 2018, 38(9): 0906002. Jinda Wu, Jin Lu, Hongliang Ren, Yali Qin, Shuqin Guo, Weisheng Hu. Nonlinear Equalizer Based on General Regression Neural Network in Coherent Optical OFDM System[J]. Acta Optica Sinica, 2018, 38(9): 0906002.