Reduction of Nonlinear Intersubcarrier Intermixing in Coherent Optical OFDM by a Fast Newton-Based Support Vector Machine Nonlinear Equalizer
Reduction of Nonlinear Intersubcarrier Intermixing in Coherent Optical OFDM by a Fast Newton-Based Support Vector Machine Nonlinear Equalizer
复制标题
DOI：
10.1109/jlt.2017.2678511
复制
发表时间：
2017-06
影响因子：
4.7
通讯作者：
E. Giacoumidis;S. Mhatli;M. Stephens;A. Tsokanos;Jinlong Wei;M. McCarthy;N. Doran;Andrew D. Ellis
中科院分区：
文献类型：
--
作者：
E. Giacoumidis;S. Mhatli;M. Stephens;A. Tsokanos;Jinlong Wei;M. McCarthy;N. Doran;Andrew D. Ellis
首次在传输距离为2000 km的40 Gb/s 16正交幅度调制相干光正交频分复用系统中实验验证了一种基于牛顿的快速支持向量机（N-SVM）非线性均衡器（NLE）。结果表明，N-SVM-NLE扩展了2 dB的基准相比，基于Volterra的NLE的最佳发射光功率。N-SVM的性能改进是由于其能够处理确定性光纤引起的非线性效应以及非线性与随机噪声之间的相互作用（例如，偏振模色散）。N-SVM比基于Volterra的NLE更能容忍子载波间非线性串扰效应，特别是当同时应用于所有子载波时。与传统的SVM相比，该算法降低了分类器的复杂度，降低了计算量和执行时间。对于4的低C参数（与复杂度相关的惩罚参数），N-SVM需要1.6 s的执行时间来有效地减轻非线性。与传统的SVM相比，N-SVM的计算量降低了2.6倍。
A fast Newton-based support vector machine (N-SVM) nonlinear equalizer (NLE) is experimentally demonstrated, for the first time, in 40 Gb/s 16-quadrature amplitude modulated coherent optical orthogonal frequency division multiplexing at 2000 km of transmission. It is shown that N-SVM-NLE extends the optimum launched optical power by 2 dB compared to the benchmark Volterra-based NLE. The performance improvement by N-SVM is due to its ability of tackling both deterministic fiber-induced nonlinear effects and the interaction between nonlinearities and stochastic noises (e.g., polarization-mode dispersion). An N-SVM is more tolerant to intersubcarrier nonlinear crosstalk effects than Volterra-based NLE, especially when applied across all subcarriers simultaneously. In contrast to the conventional SVM, the proposed algorithm is of reduced classifier complexity offering lower computational load and execution time. For a low C-parameter of 4 (a penalty parameter related to complexity), an execution time of 1.6 s is required for N-SVM to effectively mitigate nonlinearities. Compared to conventional SVM, the computational load of N-SVM is ∼6 times lower.