A Highly Accurate Fast Decoding Framework for QLDPC codes Accelerated by Noise Perturbation and Ensemble Decoding

Quantum Low-Density Parity-Check (QLDPC) codes are essential for fault-tolerant quantum computing, but their decoding remains a significant bottleneck. Traditional decoders like Belief Propagation (BP) often struggle with the degenerate error structures inherent in these codes, leading to poor performance or high latency. The authors seek a decoding framework that achieves high accuracy while significantly reducing the computational overhead associated with iterative decoding.

The authors propose Noise Assisted Ensemble Decoding (NAED). Instead of relying on iterative message-passing, NAED constructs an ensemble of Tanner forests—acyclic subgraphs of the original Tanner graph. For each forest, the decoder performs exact inference using a lightweight dynamic programming algorithm. To ensure the ensemble explores diverse and informative decoding paths, the authors introduce a synthetic soft information metric derived from channel statistics and syndrome measurements. This metric is further perturbed by controlled noise, which generates varied column orderings for the Tanner forest construction across the ensemble.

Benchmark results demonstrate that NAED achieves decoding performance comparable to or better than state-of-the-art solutions like BP+OSD0. Crucially, because NAED uses exact inference on acyclic structures rather than iterative message-passing, it provides orders-of-magnitude improvements in per-round decoding speed under circuit-level noise. The authors show that their synthetic soft information-driven approach effectively prioritizes critical error-supporting columns, making it a robust alternative to BP-based initialization.