High-speed Optical Signal Modulation Format Identification And OSNR Monitoring Based On Machine Learning

The next generation optical networks are required to allocate bandwidth,modulation format,data rate,etc.adaptively according to the time-varying channel condition,network resource efficiencies and capacity demand from customers,thus maximizing the efficiency of network resources.Therefore,the key characteristics of the receiver is to identify the bandwidth,data rate,modulation format,signal power,etc.Among them,the OSNR monitoring technology can detect the quality of the transmitted signal and provide a basis for signal detection.Modulation format recognition technology can automatically identify the modulation format of the received signal,and provide information to the DSP in the coherent receiver to automatically configure the modulation format dependent algorithm in the receiver to obtain the best reception performance.This thesis researches two modulation format recognition schemes and an OSNR monitoring scheme.The main tasks are organized as follows:1.This thesis demonstrates the use of amplitude variance in combination with only one time of 4th power transformation and fast Fourier transform(FFT)for modulation format identification in digital coherent receiver.The incoming signals are firstly classified into two main categories,PDM-mPSK and mQAM,based on the amplitude variance.Then,the sub-categories of PDM-mPSK or PDM-mQAM are further identified by utilizing the logical regression algorithm on a two-dimensional(2-D)plane which is constructed by using the mean and maximum value after 4th power spectrum density.The feasibility is firstly verified in the PDM-QPSK/-8PSK/-16QAM/-32 QAM system via numerical simulation and experiment.The results show that the minimum required OSNR value(in the case of 100% accuracy rate)is lower than its corresponding 7% FEC limit.In addition,the results show that the proposed method has a certain tolerance to linewidth and nonlinearity.2.The method based on 4th power spectral density and peak value of 8th power spectrum combined with SVM is investigated.We first project the peak value of 4th power spectrum density and peak value of 8th power spectrum density onto a 2D plane,and then identify them using SVM.For the unidentifiable 16 QAM and 64 QAM,we classify them by the amplitude partition and the mean of 4th power spectrum density.The simulation results show that this method can identify the mainstream modulation formats(PDMQPSK/-8PSK/-16PSK/-16QAM/-32QAM/-64QAM)in coherent optical communication systems,and achieve the minimum required for 100% recognition rate in the OSNR value lower than its corresponding 7% FEC.3.An OSNR monitoring scheme using the signal power spectral density and deep neural networks are demonstrated.The features of signal amplitude,2th,4th,and 8th power spectrum density are all correlative with the OSNR of signals.By utilizing the DNN to extract those OSNR depended specific features,the signal OSNR value can be estimated.Simulation results for 28 Gbaud PDM-QPSK/-8PSK/-8QAM/-16 QAM signals demonstrate OSNR monitoring with mean estimation standard errors(SEs)of 0.1dB,0.09 dB,0.33 dB,0.46 dB in back-to-back system and 0.43 dB,0.34 dB,0.66 dB,0.79 dB in 2000 km,1040km,1040 km,800km single mode fiber transmission system with input optical power of 4dBm,4dBm,3dBm and 3dBm,respectively.