| With the rapid development of data centers,the transmission rate of optical networks in data centers has been increasing from 100Gb/s to 400Gb/s or even higher.The continuous improvement of transmission rate requires higher and higher transmission technology.Forward error correction(FEC)coding technology can guarantee high speed and reliable transmission.Low density parity check(LDPC)code is a FEC technique with sparse check matrix,which has excellent characteristics of approaching Shannon limit,and has become a hot research area.At present,there are many LDPC decoding methods such as Belief Propagation(BP)algorithm、Log-likelihood Ratios-Belief Propagation(LLR-BP)algorithm、Min-Sum(MS)Algorithm,etc.,but they also have the issues of existing algorithms,that is,the contradiction among the complexity of hardware implementation,algorithm complexity and algorithm precision,so how to solve the above problems is cruical.Based on LDPC coding and decoding system,this paper mainly studies the decoding algorithm of LDPC Then,a dynamic hierarchical min-sum algorithm based on grid search and Monte Carlo criterion,min-sum decoding algorithm based on deep learning,hierarchical min-sum algorithm based on longitudinal and horizontal information are proposed.The main work of this paper is as follows:1、Since the accuracy differences between MS algorithm and LLR-BP algorithm is mainly due to the difference in the accuracy of Log-likelihood Ratio(LLR)calculation formula of check nodes,a dynamic hierarchical min-sum decoding algorithm based on grid search and Monte Carlo criterion is proposed in this paper.According to the grid search algorithm and Monte Carlo criterion,the relationship between the scaling factors and iterations is found,then iterations in the decoding process is layered and shared with the scaling factors,which can dynamically change the modification step of the verification node of LLR and improve the decoding accuracy.Simulation results illustrate that compared with the traditional MS algorithm,the decoding accuracy of the proposed algorithm is improved while keeping the complexity unchanged,and the maximum gain of the algorithm can reach 0.437dB.2、In order to solve reduce the problem that the complexity of deep learning network model increases with the increase of code length,a min-sum decoding algorithm based on deep learning is proposed.The proposed algorithm is designed to learn the ideal adjustment multiple required by MS algorithm to reach the accuracy of LLR-BP algorithm and avoid the whole iterative decoding process of network learning.The proposed algorithm aims at avoiding the whole iterative decoding process of network learning when MS algorithm is to reach the accuracy of LLR-BP algorithm,so as to "reduce the burden" of the network and ensure the performance of the coding and decoding network.At the same time,it only studies the variation characteristics of one parameter in the decoding process,so the requirement of the network is not very high,which increases the universality of the network.Simulation results show that compared with the traditional MS algorithm,the decoding performance of the proposed algorithm is greatly improved,and the maximum gain can reach 0.738dB.3、To avoid the decoding "trap" during LDPC decoding,a hierarchical min-sum algorithm based in longitudinal and horizontal information is proposed.Decoding "trap" refers to a sudden change in the value of some external messages of a node during the decoding iteration,resulting in inaccurate LLR value of the node and an increase in bit error rate.At the same time,the smaller the LLR is,the more likely this decoding "trap" will appear,and then it will be trapped by the decoding"trap".So the number of iterations goes up and the BER goes up.The proposed algorithm is modified again when the LLR value is calculated at the check point.At the same time,the horizontal and vertical information of the check node is combined to make a hard decision,so as to eliminate the decoding "trap",reduce the iterations,improve the performance of the whole coding and decoding system.Simulation results show that the decoding performance of the proposed algorithm is significantly improved compared with the traditional MS algorithm,and the maximum gain of the proposed algorithm can reach 0.733dB. |
