基于机器学习的相干光通信系统传输性能评估  

Transmission Performance Evaluation of Coherent Optical Communication System based on Machine Learning

作　　者：陈中盛 高冠军[1] 刘鸿飞[2] CHEN Zhong-sheng;GAO Guan-jun;LIU Hong-fei(State Key Laboratory of Information Photonics and Optical Communication,Beijing University of Posts and Telecommunications,Beijing 100876,China;Key Laboratory of FAST,National Astronomical Observatories,CAS,Beijing 100101,China)

机构地区：[1]北京邮电大学信息光子学与光通信国家重点实验室,北京100876 [2]中国科学院国家天文台FAST重点实验室,北京100101

出　　处：《光通信研究》2021年第4期15-20,共6页Study on Optical Communications

基　　金：国家自然科学基金资助项目(U1831110);北京邮电大学行动计划资助项目(2019XD-A15);信息光子学与光通信国家重点实验室自主研究课题资助项目(IPOC2020ZZ02)。

摘　　要：针对传统相干光通信系统中光纤链路传输性能评估方法理论计算复杂且计算时间较长的问题,文章采用机器学习技术对调制格式和输入光功率等参数各异信道的相干光传输系统传输性能进行关联式建模和精确评估。建模和评估结果表明,该机器学习模型能够代替复杂的理论计算,并且能够较为准确地给出预测值,理论计算值与模型预测Q值整体的平均误差为0.27 dB,理论计算值与模型预测光信噪比(OSNR)值整体的平均误差为0.33 dB;采用机器学习模型的计算时间稳定在4.4 s左右,而采用理论计算的时间随着干扰信道个数的增加计算时间逐渐增加。In view of the complexity of theoretical calculation and long calculation time of traditional evaluation method of optical fiber link transmission performance in coherent optical communication system,this paper uses machine learning technology to build correlation model and accurately evaluate the transmission performance of coherent optical transmission system with different modulation format and input optical power parameters.Through the training model,the results show that the machine learning model can replace the complex theoretical calculation,and can more accurately predict the predicted value,The overall average error between the theoretical value and the predicted Q value is 0.27 dB.The overall average error between the theoretical value and the predicted Optical Signal to Noise Ratio(OSNR)value is 0.33 dB.The calculation time of machine learning model is about 4.4 s,while the theoretical calculation time increases with the increase of the number of interference channels.

关 键 词：机器学习 相干光通信 传输性能评估 

分 类 号：TN914[电子电信—通信与信息系统]