Abstract
We propose a novel soft-aided iterative decoding algorithm for product codes (PCs) named iterative bounded distance decoding with combined reliability (iBDD-CR). This algorithm enhances conventional iterative bounded distance decoding (iBDD) by exploiting channel soft information through a refined reliability combining rule. Derived from the analysis of extrinsic message passing in generalized low-density-parity check (GLDPC) ensembles, iBDD-CR maintains the low-complexity binary data flow of hard-decision decoders while achieving performance gains up to 0.51 dB. This makes it an attractive solution for high-throughput applications like fiber-optic communications.
Introduction
As fiber-optic communications push toward Terabit-per-second (Tbps) speeds, decoder complexity and power consumption have become critical bottlenecks. Standard Soft-Decision (SD) decoders offer superior error correction but require high internal data flow to pass multi-bit soft messages. Conversely, Hard-Decision (HD) decoders using iBDD are fast and efficient but suffer from a significant performance gap and are prone to "miscorrections"—where the decoder incorrectly flips a bit due to a lack of reliability information.
This paper introduces iBDD-CR as a "middle way." It preserves the high-speed, single-bit internal architecture of HD decoders while utilizing channel Log-Likelihood Ratios (LLRs) locally to prevent miscorrections and improve sensitivity.
iBDD-CR Architecture and Methodology
The core innovation of iBDD-CR is the integration of a mathematically rigorous combining rule derived via Bayes' theorem. Unlike previous heuristic weighting methods, iBDD-CR combines the BDD output with the channel LLR to make a more informed final decision at each component decoder.
As shown in the architecture above, the "soft" combining happens locally within each component decoder. The information passed between row and column decoders remains purely binary, preserving the throughput advantages of HD hardware. The reliability estimates for the BDD outputs are calculated using Density Evolution (DE) analysis, treating product codes as part of the Generalized Low-Density Parity-Check (GLDPC) ensemble family.
A key insight used in this design is the "Markov Chain Trick": the authors observed that the BDD output depends only on the sign of the channel LLR, not its magnitude. This simplification allows the optimal combining rule to be implemented efficiently via small Look-Up Tables (LUTs).
Performance Analysis and Results
The algorithm was evaluated using simulated AWGN channels and specific product codes built from BCH component codes (e.g., (255,231,3) and (511,484,3)). The results demonstrate that iBDD-CR significantly outperforms standard iBDD and even theoretical "genie-aided" iBDD models that simply prevent miscorrections.
Key findings include:
- Coding Gains: iBDD-CR provides a gain of 0.36 dB for binary transmission and up to 0.51 dB for 256-QAM systems using Bit-Interleaved Coded Modulation (BICM).
- Theoretical Consistency: As seen in the performance plots, the DE thresholds closely match simulation results, confirming the validity of the GLDPC ensemble analysis.
- Hardware Readiness: The algorithm is robust to LLR quantization. Using 3-bit or 4-bit quantization results in a negligible performance loss (less than 0.07 dB), making it suitable for practical ASIC/FPGA implementations.
Conclusion
iBDD-CR successfully bridges the gap between hard-decision and soft-decision decoding for product codes. By utilizing a refined combining rule and DE-based reliability estimates, it achieves significant sensitivity gains without increasing the internal data flow complexity. While it requires additional memory for channel LLRs and LUTs, its high-throughput potential makes it a compelling candidate for next-generation optical networks. Future work will focus on full hardware implementation and extending the framework to other product-like structures such as staircase or braided codes.
Summary updated