Research On Digital Reproduction Simulation Technology Of Optical Communication System Based On Deep Learning

In the information age of increasing transmission capacity,optical fiber communication technology is booming,and optical communication system has also undergone tremendous innovation.With the development of complex optical modulation,dynamic optical transmission and other technologies,the simulation of ultra-high speed,large capacity,long distance new optical transmission system has important research value.It can effectively and flexibly simulate the actual optical transmission process,which is helpful for scientific researchers and operation and maintenance personnel to carry out transmission rehearsal,numerical simulation and theoretical analysis.Usually,a complete system simulation can collect the data information of a signal at different stages from the sender to the receiver,which can better understand the changes of the signal after passing through different modules.Although the technology in the simulation system is quite mature,the traditional modeling ideas based on mathematical and physical methods often strictly rely on expert experience,and are only effective for static ideal scenarios with complete system parameters,and each module is designed separately,usually with different assumptions and objectives.Therefore,it is difficult to determine the global optimality of the system by building the system with different modules.In addition,a hypothetical mathematical model is often embedded in the design to represent the channel.The assumed model may not be able to correctly reflect the real time-varying dynamic optical transmission process,thus affecting the system performance.Based on intensity modulation/direct detection and coherent optical communication technology,this paper proposes a digital modeling method based on deep learning BiLSTM neural network for optical communication systems with different modulation formats.This method greatly reduces the dependence on expert experience and mathematical knowledge,and can simulate the channel accurately and flexibly.The main innovations of this paper are as follows:Firstly,for low cost and short distance OOK system,the fiber channel model based on deep learning neural network has good performance in time domain amplitude and phase,frequency domain spectrum and eye diagram,and in different transmission distance,transmitted optical power,dispersion coefficient,up-sampling rate and other system configurations.All of the normalized mean square error is less than 0.002.Secondly,for PM-QPSK system with long distance and more complex channel model,the amplitude and phase characteristics of polarized light signal are successfully fitted by neural network based on waveform and constellation analysis.The decision coefficient is greater than 0.98 and the cosine similarity is as high as 0.99.