Sliced Message Passing: High Throughput Overlapped Decoding of High-Rate Low-Density Parity-Check Codes

摘要

The efficient implementation of high-rate high-throughout low-density parity-check (LDPC) code decoding presents challenges to both fully parallel decoding and memory-sharing partially parallel decoding schemes. In this paper, a new decoding scheme, sliced message passing (SMP), is introduced. The key idea is to slice the total set of $N$ variable-to-check messages into equal-sized $p$ chunks, then to perform check-node computation sequentially chunk by chunk. This new decoding scheme can break the sequential tie between the check- and variable-node update stages and thus greatly improve the throughput. The hardware architectures of SMP decoding are introduced. Each check-node processing unit of the proposed register-based architecture has only $c/p$ inputs instead of $c$ inputs. By remapping the variable-node and check-node processing unit decoding blocks, the optimized SMP decoding units can reduce the overall hardware cost, shorten the critical-path delay, and improve the hardware-usage efficiency. An optimized SMP decoder has been further implemented for a 2048-bit (6, 32) LDPC decoder of the emerging IEEE 10 GBase-T standard in an IBM CMOS 90-nm process. Implementation results from synthesis and postlayout simulation have shown the effectiveness of the proposed SMP scheme.