Research On Joint Iterative Detection And Decoding Algorithm For Polar Coded-SCMA System Based On Deep Learning

With the large-scale commercialization of the 5th Generation Mobile Networks(5G),polar code has been adopted by 3rd Generation Partnership Project(3GPP)as a 5G control channel coding technology due to its proven ability to reach Shannon channel capacity in memoryless channels and the advantages of low computational complexity of coding and decoding algorithm,on the other hand,large connection business scenarios,Sparse Code Multiple Access(SCMA)technology is a highly competitive non-orthogonal Muliple Access(NOMA)technology,which is characterized by low complexity of multiuser detection algorithm and high efficiency of forming gain spectrum.Because of its wide applicability in many fields and remarkable optimization effect,deep learning(DL)technology provides a new way for the bit error rate(BER)performance optimization of signal and channel processing algorithms.For the channel decoding algorithm based on neural network and multi-user detection algorithm,how to achieve a good balance between improving the performance of BER and reducing the complexity of operation,and how to design reasonable neural network for different message update rules and other problems need to be solved.In this paper,the Polar Coded Sparse Code Multiple Access(PC-SCMA)system's polar code decoding algorithm and the joint iterative detection decoding(JIDD)algorithm are studied and improved by combining with deep learning technology.The specific research contents are as follows:1.Design of Polar Code Belief Propagation Decoding Algorithm Based on Recurrent Neural NetworkAiming at the problems of excessive weight storage overhead and complex multiplication operation in the existing deep neural network-based polar code decoding algorithms,a recurrent neural network offset min-sum belief propagation(RNN-OMSBP)decoding algorithm is proposed in this paper,combining the characteristics of belief propagation(BP)decoding algorithm and deep learning technology.Firstly,the RNN architecture is used to realize parameter sharing among multiple iterations of the neural network,then use the OMS algorithm and modify the message in the process of iterative approximation algorithm update strategy,a polar code decoding algorithm with low computational complexity,low memory consumption and good decoding performance is implemented.The simulation results show that compared with the traditional BP decoding algorithm,the bit error rate performance is significantly improved,and compared with the DNN-BP decoding algorithm,the computational complexity is significantly reduced under the premise of less performance sacrifice.2.Design of PC-SCMA joint iterative detection decoding algorithm based on deep neural networkThe JIDD algorithm in PC-SCMA system uses the combination of the original message passing algorithm(MPA)and the polar code BP decoding algorithm,so there is a large optimization space in BER performance and operational complexity.In this paper,a joint iterative detection and decoding algorithm based on deep neural network(DNNJIDD)for PC-SCMA systems is proposed,Firstly,weight factors that can be learned offline are added to the MPA algorithm of SCMA system.Combined with the OMS-BP decoding algorithm proposed in this paper,the joint factor graph architecture of the two is expanded into a deep neural network architecture.A JIDD algorithm for PC-SCMA system with significantly improved BER performance is implemented at the cost of a small increase in computational complexity.The simulation results show that the bit error rate performance of DNN-JIDD algorithm is better than that of decoding detection separation scheme and JIDD algorithm.