Research On Nonlinear Equalization Technology For Coherent Optical Communication System With High-order Modulation Format

The emergence of cloud computing,artificial intelligence,mobile Internet,has brought us into the "big data era" of "Internet of Everything".The increasing demand for super-large data storage,transmission,sharing and other services has promoted the explosive growth of network traffic.Therefore,modern communication networks are evolving towards higher transmission rate,larger transmission capacity and better transmission quality.The optical fiber communication system has the advantages of low attenuation,strong anti-interference ability,and large transmission capacity.After decades of exploration and breakthrough,optical fiber communication system has become the cornerstone of global interconnection and the backbone of modern communication network.Coherent optical communication technology,which combines high-order modulation formats,coherent detection technology,and digital signal processing technology,can simultaneously achieve high-spectral efficiency,long-haul,and large-capacity transmission.Coherent optical fiber communication technology is an important technology to deal with the traffic crisis of modern communication networks.However,for high-speed coherent optical communication system,nonlinear impairments not only severely limit the transmission distance but also critically limit the achievable capacity.Therefore,the study of nonlinear equalization(NLE)technology for coherent optical communication system is crucial.This dissertation focuses on the research of nonlinear equalization technology for single-carrier polarization division multiplexing coherent optical communication system with high-order modulation format.The main research contents include:nonlinear decision algorithm,nonlinear equalization scheme based on neural network,nonlinear equalization scheme based on perturbation theory and regression algorithm.The main research contents and innovations are as follows:1.Mixture-of-Gaussian clustering-based nonlinear decision technique for M-QAM modulation format signalIn order to solve the problem that the traditional decision algorithm based on maximum likelihood estimation(MLE)can not effectively deal with the nonlinear distortion signal,a decision technique using mixture-of-Gaussian(MoG)clustering algorithms is proposed.A majorization method for decision-directed least mean square(DD-LMS)using the MoG clustering-based decision algorithm,MoG-LMS,is also proposed.In the MoG-LMS algorithm,we use the MoG clustering-based decision to replace the traditional MLE-based decision in the decision module,and the mean vector obtained by the MoG clustering-based decision algorithm replaces the standard constellation point for the error computation.We construct a single-carrier polarization division multiplexed(PDM)16-QAM coherent optical communication system experimental platform to verify the performances of the proposed decision algorithm.The experimental results show that,the proposed decision technique is insensitive to the system's nonlinear impairments and achieves a more accurate classification of constellation points to effectively improve nonlinear tolerance of the system.2.Nonlinear equalization scheme by feature engineering based-deep neural network for coherent optical communication systemBased on the research for the nonlinear equalization technology based on neural network in coherent optical communication system,aiming at the problem that the nonlinear equalization performance of neural network is limited due to the lack of input data characteristics,a feature engineering-based deep neural network(FE-DNN)nonlinear equalization scheme is proposed.Feature engineering and weighted loss training mechanism are introduced to improve performance of DNN.The effectiveness of the proposed FE-DNN nonlinear equalizer is verified by a PDM-64-QAM coherent optical communication system.The experimental results show that feature engineering and weighted loss training mechanism can effectively improve the performance of DNN.And the maximum 1.07 dB Q factor improvement is obtained.3.Bi-directional gated recurrent unit neural network based nonlinear equalization schemeA bi-directional gated recurrent unit(Bi-GRU)neural network is proposed for nonlinear equalization(NLE)in coherent optical communication systems not only to handle inter-symbol interference(ISI)among preceding and succeeding symbols caused by chromatic dispersion(CD)and fiber nonlinear effect,but also to compensate the nonlinear impairments caused by devices and fiber transmission links.Experimental verification of the proposed nonlinear equalization scheme is demonstrated in a PDM-64-QAM coherent optical communication system.Results show that,the Q factors can exceed the HD-FEC limit of 8.53 dB(according to 3.8×10-3 in BER)with the aid of Bi-GRU NLE,when the launched optical power is in the range of-3 dBm to 3 dBm.The optimum launch optical power can be extended by 2 dB after nonlinear equalization.4.Bi-directional long short-term memory neural network-conditional random field-based nonlinear equalization schemeBased on the research for the nonlinear equalization technology based on recurrent neural network in coherent optical communication system,a nonlinear equalizer based on bi-directional long short-term memory neural network-conditional random field(Bi-LSTM-CRF)is proposed and experimentally demonstrated for compensation of nonlinear distortions in a PDM-64-QAM coherent optical communication system.Experimental results show that,the Q factors can exceed the FEC limit of 9.8 dB(according to 1.0×10-3 in BER)with the aid of Bi-LSTM-CRF NLE,when the launched optical power is in the range of-3 dBm to 3 dBm.The optimum launch optical power can be extended by 2 dB.5.Regression algorithm-based perturbative nonlinear equalization scheme for coherent optical communication systemBased on the research for nonlinear equalization technology based on perturbation theory in coherent optical communication system,regression algorithm-based perturbative nonlinear equalization schemes are proposed.Without prior knowledge of the transmission link parameters,the intra-channel cross-phase modulation and intrachannel four-wave mixing triplets calculated according to the received signal sequence are used as input features for regression models.The nonlinear impairment of the signal after the transmission is predicted by regression models,and the predicted nonlinear impairment is removed from the received symbol data to realize the nonlinear equalization.The effectiveness of the proposed nonlinear equalization schemes is verified by a PDM-64-QAM coherent optical communication system.Experimental results show that,the BER can achieve lower than 1.0×10-3 after support vector regression-based perturbative nonlinear equalization when the launched optical power is 1 dBm.The optimum launch optical power can be extended by 2 dB.