Effective Application of Normalized Min-Sum Decoding for BCH Codes

High-throughput decoding of BCH codes necessitates efficient and parallelizable decoders. However, the algebraic rigidity of BCH codes poses significant challenges to applying parallel belief propagation variants. To address this, we propose a systematic design scheme for constructing parity-check matrices using a heuristic approach. This involves a sequence of binary sum operations and row cyclic shifts on the standard parity-check matrix, aiming to generate a redundant, low-density, and quasi-regular matrix with significantly fewer length-4 cycles. The relationships between frame error rate, rank deficiency of minimum-weight dual-code codewords, and row redundancy are empirically analyzed. For the revised normalized min-sum decoder, we introduce three types of random automorphisms applied to decoder inputs. These are unpacked and aggregated by summing messages after each iteration, achieving a 1-2dB improvement in bit error rate compared to parallelizable counterparts and two orders of magnitude faster convergence in iterations than iterative rivals. Additionally, undetected errors are highlighted as a non-negligible issue for very short BCH codes.