Research On Dispersion Estimation Algorithm In Long-haul And High Speed Coherent Optical Communication System Under Large DGD

With the explosive growth of network data traffic,optical fiber communication has developed towards the trend of ultra-high speed,ultra-long distance,ultra-large capacity,and ultra-low loss.This is how to control the transition from transmitter to fiber channel to receiver.The physical damage introduced during a transmission puts forward higher requirements.Common physical damage includes loss,dispersion,and nonlinearity.This article focuses on dispersion effects.In order to better compensate for the degradation of the transmission performance of the system and the impact on the quality of the optical signal,the dispersion of the system needs to be accurately estimated at the receiving end,and a variety of dispersion estimation algorithms have been born.Based on the Fractional Fourier Transform(Fr FT)idea,this paper develops a dispersion estimation algorithm,combines VPI and Matlab to build a long-distance polarization-multiplexed coherent optical communication joint simulation system,and explores the algorithm's conditions for large differential group delay(DGD).This paper proposes a method to improve the dispersion estimation accuracy of the algorithm under the condition of random large DGD using machine learning.The main results achieved include:(1)In the case of a fixed fiber length,when the DGD changes randomly within 50?200 ps,the BP neural network is used to learn and train the signal that has been transmitted over a long distance and has been estimated by the Fr FT dispersion estimation algorithm.When the transmission distance is 1000 km,in the QPSK and 16QAM modulation formats,the absolute error of the dispersion estimation algorithm is reduced from 10⁴orders before training to 5 ps after training,and the relative error is controlled within 10^-3orders..(2)In the case where the fiber transmission distance varies from 500?1500 km,when the DGD changes randomly within 50?130 ps,the BP neural network is used to perform a signal transmitted over a long distance and subjected to dispersion estimation based on the Fr FT dispersion estimation algorithm Learning and training,when the fiber length randomly changes in this range,the absolute error of the dispersion estimation algorithm is reduced from 10⁴orders before training to 100 ps after training under the two modulation formats of QPSK and 16QAM,and the relative error is controlled within Within 10^-2orders of magnitude.It was also found that as the fiber length further increases,the effectiveness of the BP neural network in optimizing the error of the dispersion estimation algorithm will decrease.The above method combines the machine learning and the dispersion estimation algorithm,and preliminarily solves the problem of the influence of large random DGD on the dispersion estimation algorithm which cannot be solved by the traditional method.It is expected to provide a new idea for further improving the accuracy of dispersion estimation algorithm and upgrading the signal processing algorithm in DSP.