Research On High-performance LDPC Decoding Algorithm

With the performance approaching Shannon limit,Low-density paritycheck(LDPC)code has been widely applied in various types of wireless communication scenarios,such as the next generation digital broadcast television standard(ATSC 3.0)and the enhanced mobile broadband(e MBB)of the fifth-generation mobile communication system(5G).The error correction performance of LDPC is determined by decoding algorithm,which requires better performance in waterfall area,lower error floor or faster convergence rate to meet the transmission requirements of highthroughput and high-speed.So far,belief propagation(BP)algorithm is one of the best decoding algorithms.However,due to high computational complexity,it is not easy to implement in hardware.The min-sum(MS)algorithm,which reduces complexity of BP algorithm,degrades performance seriously.In order to improve the decoding performance of MS algorithm,normalized min-sum(NMS)algorithm and offset min-sum(OMS)algorithm are proposed,both of which introduce a fixed correction factor.When NMS algorithm and OMS algorithm are applied in LDPC codes of ATSC 3.0,it indicates their performance are better than that of MS algorithm,but there is still room for improvement.In addition,NMS algorithm may suffer error floors for some codes.In this paper,an improved min-sum(SVNMS)algorithm based on generalized mutual information is obtained by simplification,and a criterion to evaluate the offset factor is derived based on the theory of generalized mutual information and density evolution.We proposed an improved minsum(SVOMS)algorithm based on this criterion.The correction factors of these two algorithms vary with the number of iterations.In order to match the actual LLR,corresponding correction factor sequences are obtained offline through Monte Carlo.We run simulation for LDPC codes of ATSC3.0,and the results show that the performance of SVOMS is closest to that of BP algorithm.SVNMS algorithm not only achieves a noticeable gain over NMS algorithm in waterfall region,but also eliminates error floors occurred in NMS algorithm.Although for some codes,the performance of SVNMS algorithm is not as good as that of OMS algorithm,SVNMS algorithm is not sensitive to channel estimation while OMS algorithm is.Both algorithms provide more options for LDPC decoder in ATSC 3.0.High convergence rate is required in many low-latency applications,such as 5G.To improve the convergence rate of decoding,LBP algorithm,RBP algorithm and NWRBP algorithm are proposed.Both RBP algorithm and NWRBP algorithm are based on informed dynamic scheduling,which reasonably adjust the update order of node message based on residual.Their convergence rate is better than that of LBP algorithm.When applied in LDPC codes with medium or short length,RBP algorithm has higher convergence rate in the early iteration time.However,it is a greedy algorithm,so the order of node information update maynot be global optimal.NWRBP algorithm optimizes the global performance by receiving and delivering messages of more nodes to achieve better performance than RBP algorithm.However,its complexity is greatly increased.In this paper,belief propagation algorithm with a joint scheduling based on dynamic message(RLBP)is proposed.In early stage of decoding,it adopts algorithm based on informed dynamic scheduling to accelerate convergence,then LBP algorithm based on non-dynamic scheduling is applied in later iteration time,which achieves excellent decoding performance in the case of lowcomplexity.