Research On Optical Performance Monitoring Technology Based On Deep Learning Algorithm

In the information age,mobile Internet,big data,and live video are rapidly developing.The demand for high information transmission rates continues to rise.Optical fiber communication systems continue to develop toward high transmission rates,high channel capacity,and long transmission distances.The coherent optical fiber communication system has the characteristic of high receiving sensitivity,which can support the use of high-order modulation formats with high spectral efficiency and digital signal processing technology,so it is widely used in high-speed,long-distance fiber transmission.In a coherent optical fiber communication system,the receiver must perform chromatic dispersion equalization after coherent reception,firstly.Most chromatic dispersion equalization algorithms are based on chromatic dispersion monitoring,which will affect non-linear monitoring.Therefore,in the next generation of dynamic reconfigurable optical networks,it is of great significance to perform fast and low-cost chromatic dispersion monitoring.Linear damage such as chromatic dispersion can be compensated by digital signal processing technology,so nonlinear noise has become one of the main factors affecting system performance.Nonlinear noise and ASE noise are aliased with each other,so monitor OSNR and nonlinear noise are very important in optical transmission systems.The changes of optical performance indicators such as chromatic dispersion,OSNR,and non-linear noise can be reflected in data,such as spectrum and time domain signals.Deep learning has strong feature extraction capabilities,and can capture relevant feature information from these types of data.Therefore,this thesis uses deep learning as the key technology to study the three optical performance indicators of chromatic dispersion,OSNR,and nonlinear noise.The research contents and results are as follows:1.Aiming at the limitations of traditional dispersion monitoring algorithms,a chromatic dispersion monitoring algorithm based on deep neural networks with ultra-low sampling rates is proposed,which can reduce the receiver's requirements for ADC high sampling rates.In order to verify the feasibility of chromatic dispersion monitoring algorithm based on deep neural network with ultra-low sampling rate,a 28 GBaud 5-channel optical fiber communication transmission simulation system was set up.The results show that the maximum MAE of the dispersion measured under the QPSK and 16QAM modulation formats is less than 35 ps/nm and 75 ps/nm,respectively.It also proves that the monitoring results are not affected by the polarization state and are robust to ASE noise and nonlinear noise.In addition,an experimental platform of 20 GBaud QPSK optical fiber transmission system is established,which proves the feasibility of the method in practical optical fiber communication systems.2.Aiming at the monitoring of nonlinear noise and OSNR under high nonlinear conditions,a joint OSNR and nonlinear noise power monitoring algorithm based on deep neural network is proposed.This method uses deep neural networks to monitor OSNR and non-linear noise power at the same time.It can learn the effect of two kinds of noise on the frequency spectrum.It does not need to train two deep neural networks,thus reducing the redundancy of neural network parameters.The feasibility of this method is proved by the simulation of 28 GBaud 5-channel fiber optic communication transmission system.OSNR monitoring results show that when the OSNR in the fiber link changes from 12 dB to 30 dB,the maximum value of the OSNR monitoring MAE of QPSK and 16QAM does not exceed 0.5 dB.The results of nonlinear noise power monitoring show that the maximum MAE of the nonlinear noise power of QPSK and 16QAM does not exceed 1 dB.At the same time,the robustness of OSNR and non-linear noise power monitoring algorithm based on deep neural network to non-linear noise power is proved to be robust to ASE noise.In summary,this thesis studies chromatic dispersion monitoring and joint monitoring of OSNR and nonlinear noise based on deep learning.A chromatic dispersion monitoring algorithm based on deep neural networks is proposed with an extremely low sampling rate,which can effectively reduce the receiving cost.A joint OSNR and non-linear noise power monitoring algorithm based on deep neural network is proposed to reduce the redundancy of neural network parameters.The results of this thesis are of great significance for optical performance monitoring in coherent fiber communications.