arXiv:2610.01068
Learned Parallel Bit-Flipping Sequential Belief Propagation Decoding of Quantum LDPC Codes
Mohsen Moradi, Taejoon Kim, Remi A. Chou
Abstract
Quantum low-density parity-check (QLDPC) codes are promising candidates for low-overhead fault-tolerant quantum computation, but their practical use requires fast, low-complexity, and reliable decoders. Belief propagation (BP) is attractive because of its local message-passing structure, yet standard flooding BP often suffers from convergence failures on QLDPC codes due to short cycles, degeneracy, and symmetric decoding trajectories. Learned sequential BP improves convergence by using a reinforcement-learning policy to choose variable-node update orders, but a single learned trajectory can still be sensitive to unfavorable local Pauli decisions. We propose a learned parallel bit-flipping sequential BP decoder with a new quaternary score combining syndrome gain, a quantized Pauli log-likelihood penalty, and Q-table lookahead at hypothetical post-flip states. Our decoder first runs learned sequential BP for a fixed number of iterations. If the syndrome is not satisfied, it constructs quaternary bit-flipping candidates, each corresponding to changing the current Pauli decision of one qubit. The selected candidates initialize independent learned sequential BP continuations from the same decoder state. Since these continuations are independent, they can be executed in parallel, so testing several candidates mainly increases parallel hardware resources rather than sequential decoding latency. Simulations on representative QLDPC codes over the depolarizing channel show that the proposed decoder improves the reliability of learned sequential BP while having a parallel low-latency structure.
AI Summary
The paper addresses convergence failures and decoding depth in belief-propagation (BP) decoding of CSS quantum LDPC codes under independent depolarizing noise. It extends reinforcement-learning-based sequential BP (RL-S), which uses an offline-trained Q-table to select variable-node update orders, with a one-shot branching stage. After an initial RL-S run, an unsatisfied syndrome triggers evaluation of single-qubit Pauli changes. Candidates are ranked using residual-syndrome reduction, a quantized local log-likelihood penalty, and Q-table values at hypothetical post-change states. The selected candidates initialize independent continuations from the saved BP state, with finite-magnitude message injection biasing each branch toward its selected change. Simulations use $K=25$ branches. With sufficient parallel hardware, the continuation depth is $T_{\rm br}$ rather than $K T_{\rm br}$, giving total depth $T_0+T_{\rm br}$, apart from candidate-selection and state-management overhead. Simulations on three finite-length codes report that depth 200 matches RL-S with a budget of 1000 on the $[[180,10,15\leq d\leq18]]$ A5 and $[[288,12,18]]$ bivariate bicycle codes. Reported block-error-rate reductions are approximately tenfold against BP, BPGD, and BP-OSD-0 for A5 at $p=0.04$, and greater than tenfold against BP and BPGD for the 288-qubit code at $p=0.05$. For the $[[144,12,12]]$ code, depth 20 matches RL-S at 100 and BP at 1000. The evidence is numerical; the latency argument assumes parallel branch resources rather than a measured hardware implementation. Detailed complexity analysis, hardware studies, and systematic score-parameter optimization are explicitly left for future work.
Error found: The reported decoding depths establish only an idealized parallel-stage comparison, not an end-to-end latency reduction including candidate selection and branch-management overhead. The A5 results describe T_0 and T_br as sequential updates although the algorithm uses iterations associated with VN sweeps, leaving the units of the claimed depth comparison ambiguous. The numerical experiments omit the number and values of the reliability-penalty thresholds needed to reproduce candidate selection.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.