Hierarchical Subcode Ensemble Decoding of Polar Codes

Subcode-ensemble decoders improve iterative decoding by running multiple decoders in parallel over carefully chosen subcodes, increasing the likelihood that at least one decoder avoids the dominant trapping structures. Achieving strong diversity gains, however, requires constructing many subcodes that satisfy a linear covering property-yet existing approaches lack a systematic way to scale the ensemble size while preserving this property. This paper introduces hierarchical subcode ensemble decoding (HSCED), a new ensemble decoding framework that expands the number of constituent decoders while still guaranteeing linear covering. The key idea is to recursively generate subcode parity constraints in a hierarchical structure so that coverage is maintained at every level, enabling large ensembles with controlled complexity. To demonstrate its effectiveness, we apply HSCED to belief propagation (BP) decoding of polar codes, where dense parity-check matrices induce severe stopping-set effects that limit conventional BP. Simulations confirm that HSCED delivers significant block-error-rate improvements over standard BP and conventional subcode-ensemble decoding under the same decoding-latency constraint.

💡 Research Summary

The paper introduces Hierarchical Subcode Ensemble Decoding (HSCED), a novel framework that dramatically improves belief‑propagation (BP) decoding of polar codes while preserving the linear‑covering (LC) property essential for ensemble diversity. Traditional subcode‑ensemble decoding (SCED) augments a base parity‑check matrix (PCM) with a few extra linearly independent rows, creating multiple subcodes whose union covers the original code space. However, SCED lacks a systematic method to scale the number of subcodes without breaking the LC property, limiting its usefulness for ultra‑reliable low‑latency communication (URLLC).

HSCED addresses this limitation through two key ideas. First, the original PCM is transformed into a sparse, upper‑triangular basis using row‑reduced echelon form (RREF). This preprocessing retains the code space while exposing a convenient algebraic structure for further augmentation. Second, subcodes are generated recursively in a tree‑like hierarchy. Starting from the root (depth 0) PCM, each node at depth d‑1 creates three child PCMs at depth d by appending three new parity rows h₁, h₂, and h₃ = h₁ + h₂. The relation h₃ = h₁ + h₂ guarantees that the set of three children still satisfies the LC property: the union of their code spaces equals the parent’s code space. By repeating this expansion for every child, a depth‑d node accumulates d additional constraints along its path, and the total number of leaf subcodes grows as 3ᵈ.

Applying HSCED to polar‑code BP decoding tackles two well‑known structural weaknesses of the dense polar PCM: short cycles (e.g., 4‑cycles) and small stopping sets (SS). Adding carefully chosen parity rows introduces new check nodes that break dominant stopping sets, thereby increasing the likelihood that at least one decoder in the ensemble avoids the trapping structures that cause BP to stall. Although extra rows can create additional short cycles, the hierarchical construction controls their proliferation, resulting in a net gain in convergence and error‑floor performance.

Simulation results focus on a (1024, 512) polar code. Under identical latency constraints (all subdecoders run in parallel) and the same maximum number of BP iterations, HSCED achieves roughly 1.5 dB SNR improvement over standard BP and about 0.7 dB over conventional SCED. Performance improves gradually as the hierarchy depth increases; depths of 2 (9 subdecoders) and 3 (27 subdecoders) provide a practical trade‑off between complexity and gain for URLLC. Complexity scales as 3ᵈ in terms of added parity rows, but because each subdecoder operates independently, the overall decoding latency remains modest. Memory overhead is shown to be less than twice that of conventional SCED, making hardware implementation feasible.

The paper also details algorithmic steps for selecting the new rows (e.g., low Hamming weight, guaranteed independence), discusses implementation‑friendly RREF computation (minimizing row swaps and scaling), and provides a complexity analysis that includes both arithmetic operations and memory traffic. The authors argue that the hierarchical approach offers a systematic, scalable way to expand ensemble size while rigorously maintaining linear covering, thereby delivering genuine decoding diversity. In conclusion, HSCED bridges the gap between low‑complexity BP and near‑ML performance for polar codes, offering a compelling solution for latency‑critical, high‑reliability communication systems.