The Research On Forward Error-Correcting Code Based On Deep Learning In Optical Communication

As the main communication method in modern infrastructure construction,optical fiber communication supports the development of intelligent and digital information technology.Faced with the continuous development of high-bandwidth applications such as big data,cloud computing,online teaching,and remote conferences,the pressure on the capacity of high-speed optical transmission networks is increasing.Therefore,on how to further increase the channel capacity and improve the transmission efficiency,higher requirements are put forward for the optical fiber communication system.Forward Error Correction(FEC)coding,as a kind of channel coding technology,can effectively reduce the influence of error information on the receiving end,and improve the transmission efficiency of the system.At the same time,the deep learning(DL)technology is used to optimize the forward error correction code decoding method of the system,further improve the system performance,and reduce the complexity.This thesis studies the forward error correction code in the optical communication,conducts an in-depth study on the Low-Density Parity Check(LDPC)technology based on deep learning.To improve the transmission performance of the system,two types of deep-learning-based LDPC decoding algorithms are proposed,and an intelligent Log Likelihood Ratio(LLR)estimation algorithm based on neural network is proposed:(1)In order to solve the problem of high iterative complexity of the iterative decoding algorithm in LDPC codes,LDPC decoding algorithms based on Deep Neural Network(DNN),Convolutional Neural Network(CNN)and Recurrent Neural Network(RNN)are proposed respectively.The advantage of the proposed algorithms that only needs hard decision codeword input without LLR operation,which reduces the computational complexity of the channel receiver.The simulation results show that the proposed algorithm can achieve better performance than the hard-decision decoding algorithm under a certain code length,and better than the softdecision iterative decoding algorithm when the channel conditions are poor.Among them,the best deep neural network decoder can bring up to 1dB of gain.(2)In order to solve the problem that the network is difficult to converge as the codeword length increases,by analyzing the characteristics of the LDPC iterative decoding algorithm,an LDPC decoding algorithm based on a graph model network is proposed,and a neural network is established by using the Tanner graph structure of the LDPC code.Simulations are carried out for different cases of Gaussian white noise channel and fiber channel.The simulation results show that,compared with the iterative decoding algorithm of the same complexity,the proposed algorithm can bring about 0.6dB gain under the condition of low SNR,and about 0.2dB under the condition of high SNR.(3)In order to solve the problem that the channel LLR calculation method is complex,and the estimation algorithm error is large in the softdecision decoding algorithm,an intelligent LLR estimation algorithm based on neural network is proposed in this paper.A neural network trained with a large amount of channel data is used to implement the calculation of LLR.The performance is verified by the LDPC decode algorithm.Compared with the accurate calculation method of LLR,the computational complexity of proposed method can be effectively reduced,and compared with the commonly used LLR estimation method,when the frame error rate is high,it can bring 0.2dB sensitivity improvement,has higher accuracy and can improve the communication efficiency.