一种提高LDPC译码层内并行度的方法

An approach to improve parallelism inside layer in LDPC decoding

普通准循环低密度校验码(QC-LDPC)按照SHIFT结构特性分层时,层内并行度难以提高。提出了一种非均匀抽取分层方法,可在保证层内行(列)重至多为1的前提下,提高层内并行度,并设计了相应的寻址结构和流水结构,进而提高吞吐率。对于实际应用中的2种标准码型——空间数据系统咨询委员会(CCSDS)深空通信(5 632,4 096)码和中国地面数字电视传输(DTMB)标准(7 493,6 096)码,层内并行度对比均匀抽取分别提升41.3%和32.2%。对深空通信码的综合结果表明,对比采用均匀抽取的译码器,使用双相消息传递译码算法时,可在相同资源利用率的情况下提高吞吐率41.3%;使用按列分层译码算法时,译码系统最大吞吐率可提升28.2%。

It is hard to increase parallelism inside layer when converting a generic Quasi Cyclic-Low Density Parity Check（QC-LDPC） code to a SHIFT-structured LDPC code.The paper proposes a non-uniform layered partitioning scheme,which can efficiently increase the parallelism inside layer while maximum row（column） weight of each layer is not more than one.Corresponding shuffle network and pipeline architecture are also designed.In contrast with uniform layered partitioning scheme,the proposed scheme can enhance the parallelism inside layer by 41.3% and 32.2% respectively for two application codes,Consultative Committee for Space Data Systems（CCSDS）（5 632,4 096） code and China Digital Television Terrestrial Multimedia Broadcasting（DTMB）（7 493,6 096） code.Synthesized results of（5 632,4 096） code show that the non-uniform scheme can improve the parallelism by 41.3% comparing to uniform scheme with the same resource utilization while using the Two Phase Message Passing（TPMP） algorithm;and the maximum throughput of the decoding system can be improved by 28.2% while using Shuffled Belief Propagation Decoding（SBP） algorithm.