Reinforced Ordered Statistics Decoding

• Reinforced OSD is a robust decoding strategy that enhances classical OSD by integrating modified BP, neural reliability refinement, and statistical rules to approach maximum likelihood performance.
• It refines the reliability ranking using iterative decoders and model-based adjustments, which control the test-error-pattern search and reduce computational complexity.
• The method employs dynamic path control, candidate pruning, and early termination rules to limit exhaustive reprocessing while maintaining strong error-correction capabilities.

Searching arXiv for recent and foundational papers on reinforced ordered statistics decoding and related OSD enhancements. In the papers considered here, reinforced ordered statistics decoding denotes OSD schemes that are augmented so that the reliability ordering, the test-error-pattern (TEP) traversal, or the stopping logic is no longer determined by classical OSD alone. The reinforcement may come from a modified belief-propagation front end with an offset parameter (Zhang et al., 2023), from neural models that refine reliabilities using the soft-output trajectory of normalized min-sum or neural min-sum decoding (Li et al., 2023, Li et al., 2024, Li et al., 29 Sep 2025), or from statistically derived stopping and discarding rules based on distance distributions (Yue et al., 2020). Across these variants, the shared objective is near-maximum-likelihood behavior for short block codes with substantially lower average complexity than higher-order or exhaustive OSD (Zhang et al., 2023, Li et al., 29 Sep 2025).

1. Classical OSD substrate

Classical order-$p$ OSD begins by sorting coordinates according to reliability, forming a size-$K$ most-reliable basis and a size-$(N-K)$ least-reliable basis, and then applying Gaussian elimination to a permuted parity-check or generator matrix to obtain systematic form (Li et al., 29 Sep 2025). The literature surveyed here uses both the language of most reliable positions (MRPs) and most-reliable basis (MRB) for the $K$ coordinates that anchor reprocessing (Zhang et al., 2023, Li et al., 29 Sep 2025). After the permutation and elimination steps, OSD flips all TEPs of Hamming weight at most $p$ in the MRB, re-encodes each candidate, and selects the codeword minimizing a weighted Hamming metric such as

$\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$

or, in the generator-matrix formulation,

$$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$

This is the algorithmic core that reinforced OSD retains (Zhang et al., 2023, Li et al., 2023).

The limitations of standard OSD are also consistent across the cited papers. The number of TEPs grows combinatorially in $p$ and $K$, Gaussian elimination and repeated syndrome checks can dominate the cost, and OSD by itself does not exploit the full soft-information trajectory produced by an iterative decoder (Li et al., 29 Sep 2025). In embedded hybrid decoders, this makes OSD the throughput bottleneck unless the MRB is improved or the TEP search is curtailed (Li et al., 2023, Li et al., 2024).

2. Reinforcement by iterative front ends

One reinforcement strategy is to modify the iterative decoder that precedes OSD. In the modified-BP front end for short LDPC codes, the variable-to-check update and posterior log-likelihood ratio are changed by introducing an offset parameter $\beta\in(0,1]$: $K$0 The check-to-variable update remains

$K$1

By choosing $K$2, the modified front end tempers the influence of correlated extrinsic messages at early iterations and produces a refined reliability vector $K$3 for the subsequent OSD stage (Zhang et al., 2023).

The same work ties the number of front-end iterations to Tanner-graph girth through

$K$4

with the explicit rationale that excessive BP iterations on a loopy graph may harm reliability (Zhang et al., 2023). After $K$5 iterations, the coordinates are sorted by $K$6, the first $K$7 positions define the MRPs, and OSD is applied to the corresponding information basis. The central performance claim is specific: if modified BP reduces the probability that the number of bit errors in the MRPs exceeds $K$8, then order-$K$9 mBP-OSD can approximately match order-$(N-K)$0 OSD in block error rate (Zhang et al., 2023).

A second family of front ends uses normalized min-sum or neural min-sum decoding. In the hybrid NMS-OSD architecture, the NMS decoder first attempts convergence; if it succeeds and, where used, passes an undetected-error check, decoding stops, otherwise reinforced OSD is invoked (Li et al., 29 Sep 2025). This two-stage structure is also present in Li and Yu’s NMS-DIA-OSD pipeline, where OSD is activated upon failure of a neural min-sum decoder (Li et al., 2023). A plausible implication is that reinforced OSD is not a replacement for iterative decoding but a relayed or serially coupled reprocessing stage.

3. Reliability refinement as the principal reinforcement mechanism

A major shift from classical OSD is that the reliability ordering is no longer restricted to raw channel magnitudes or the last-iteration LLRs. In Li and Yu’s formulation, for bit $(N-K)$1 the neural model consumes the time series

$(N-K)$2

where

$(N-K)$3

A small 1-D convolutional network then outputs a scalar reliability score

$(N-K)$4

with the interpretation that larger $(N-K)$5 indicates higher reliability (Li et al., 2023). The reported architecture has three Conv1D layers followed by flatten and a dense scalar output, and training minimizes a bitwise cross-entropy loss on NMS failures (Li et al., 2023).

In the later neural-model-augmented hybrid NMS-OSD decoder, the decoding information aggregation model operates on the full posterior-LLR trajectory

$(N-K)$6

The DIA CNN uses two 1-D convolutions across the iteration axis, a flatten layer, and a dense output layer producing refined reliabilities $(N-K)$7, which are injected into the OSD sorting stage through

$(N-K)$8

The training loss is mean-squared error against an ideal reliability target (Li et al., 29 Sep 2025). This explicitly addresses one stated limitation of standard OSD: it does not exploit the full soft-information trajectory available from an iterative decoder (Li et al., 29 Sep 2025).

A related but architecturally simpler variant replaces the convolutional model with a four-layer fully connected network: $(N-K)$9 Here the refined reliabilities are used to order bits from least to most reliable before constructing the MRB, and the network is trained off-line on failed NMS decodes with an $K$0-regularized mean-squared-error loss whose target reliabilities are extracted from a maximum-likelihood oracle (Li et al., 2024). Taken together, these constructions suggest that, within reinforced OSD, the dominant design problem is not the re-encoding rule itself but the quality of the reliability ranking supplied to OSD.

4. Decoding-path design and candidate-space restriction

A second reinforcement axis concerns the order in which TEPs are explored. In Li and Yu’s adaptive OSD, the MRB is divided into $K$1 contiguous segments of widths $K$2, and a TEP block is indexed by an order pattern

$K$3

During a query phase, the hit count

$K$4

is accumulated and used as a priority score; the top $K$5 patterns form a fixed decoding path for the test phase (Li et al., 2023). Complexity can then be controlled either by choosing $K$6 or by imposing segment-wise constraints such as

$K$7

which prune high-weight blocks (Li et al., 2023).

The fixed-path construction in the 2024 short-LDPC work follows the same logic with different notation. The refined reliabilities $K$8 are sorted, the MRB is partitioned into contiguous segments, and a path length $K$9 is selected so that only the most plausible TEP blocks—ordered by past hit-rate—are visited in series or partially in parallel (Li et al., 2024). This replaces the exhaustive sweep over all orders up to $p$0 with a historically informed path.

The 2025 hybrid NMS-OSD paper develops an adaptive decoding path initialized by the arranged list of the most a priori likely tests algorithm. A buffer of size $p$1 is initialized with counters $p$2; when OSD corrects with a buffered pattern, the corresponding counter is decremented, and the buffer is periodically re-sorted so that the front $p$3 TEPs define the current decoding path (Li et al., 29 Sep 2025). The stated effect is to pull useful TEPs forward and tailor the search path to the empirical error statistics (Li et al., 29 Sep 2025).

A separate but related mechanism is auxiliary candidate pruning. Before final re-encoding metric evaluation, Li and Yu discard any candidate $p$4 whose disagreement with the anchoring hard decision on a small subset of very reliable LRB positions exceeds a threshold: $p$5 The typical choice reported is $p$6 and $p$7, and the empirical claim is that this halves the final list size with negligible FER loss (Li et al., 2023).

5. Early termination and statistical stopping rules

Reinforcement also appears as decision logic that prevents traversal of the full TEP list. In the sliding-window-assisted short-LDPC decoder, the fixed path is scanned through a window of width $p$8. For each window position, the minimum weighted distance of each block is computed,

$p$9

the resulting vector is sorted and passed to a small early-termination network

$\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$0

and traversal stops when

$\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$1

If the condition fails, the window slides by one block (Li et al., 2024). The user-set soft margin $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$2 explicitly trades performance against complexity: as $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$3, early termination is effectively disabled; as $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$4, complexity is reduced at the price of some performance loss (Li et al., 2024).

The later hybrid NMS-OSD formulation adopts a closely related sliding-window assisted model. The path of length $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$5 is broken into blocks of size $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$6, a window of width $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$7 is formed, and for block $\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$8 the minimum distance

$\sum_{i=1}^N \mathbf 1\{\hat c_i\neq c_i\}\,|y_i|$9

is computed. A two-layer neural model predicts stop versus continue from the vector of current block minima, and termination occurs when the margin $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$0 exceeds a threshold $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$1 (Li et al., 29 Sep 2025). The stated bounds are that the worst case examines all $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$2 TEPs, while the best case examines only $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$3 TEPs (Li et al., 29 Sep 2025).

A non-neural reinforcement line derives stopping and discarding rules from the statistical distributions of Hamming distance and weighted Hamming distance in OSD. The 2020 revisit to OSD models these distances as two-component mixtures linked to the number of MRB errors and the code weight enumerator, and then defines hard and soft stopping rules and discarding rules: HISR, HGSR, HDR, SISR, SGSR, and SDR (Yue et al., 2020). For example, the hard individual stopping rule tests the posterior success probability

$$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$4

while the soft analogues use weighted distances instead of Hamming distances (Yue et al., 2020). This is a conceptually different form of reinforcement: the OSD list is not improved by learned reliabilities, but by probabilistic rules that stop or discard search branches when their posterior utility is sufficiently low.

6. Reported performance, complexity, and scope

The reported gains are concentrated in short-code regimes where classical OSD is effective but expensive. For the short LDPC $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$5 TU KL code, pure BP at $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$6 dB has BLER $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$7, order-3 OSD has BLER $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$8, and order-2 mBP-OSD matches the order-3 OSD BLER; at high SNR the average CPU time per codeword is approximately $$\mathbf c_{\min} = \arg\min_{\mathbf c_{\mathbf e}} \sum_{i=1}^N |\ell_i''|\, [c_{\mathbf e,i}\oplus y_i].$$9 ms for BP, $p$0 ms for order-2 mBP-OSD, and $p$1 ms for order-3 OSD (Zhang et al., 2023). For the $p$2 CCSDS LDPC code, order-3 mBP-OSD tracks order-4 OSD down to BLER $p$3, with average decoding times at $p$4 dB of approximately $p$5 ms for BP, $p$6 ms for order-3 mBP-OSD, and $p$7 ms for order-4 OSD (Zhang et al., 2023).

In Li and Yu’s neural enhancement of OSD for the LDPC $p$8 code at $p$9, classical order-2 OSD with path length $K$0 incurs an approximately $K$1 dB loss versus ML, whereas the reinforced N–D–O$K$2 configuration is about $K$3 dB closer to ML and yields about $K$4 dB gain over N–O$K$5; with a slightly longer path, N–D–O$K$6–S approaches within $K$7 dB of ML at $K$8 (Li et al., 2023). The same paper attributes part of the reduction to a factor-of-two decrease in surviving candidates after pruning (Li et al., 2023).

For the 2024 short-LDPC decoder, the LDPC CCSDS $K$9 code at $\beta\in(0,1]$0 dB yields $\beta\in(0,1]$1 for N–O$\beta\in(0,1]$2 versus $\beta\in(0,1]$3 for PB-OSD, while using about $\beta\in(0,1]$4K TEPs on average rather than about $\beta\in(0,1]$5K (Li et al., 2024). The same source states that the full hybrid NMS + DIA + OSDsw is within $\beta\in(0,1]$6 dB of the learned-decimation NBP-D$\beta\in(0,1]$7 at $\beta\in(0,1]$8 but at no more than $\beta\in(0,1]$9 of its complexity (Li et al., 2024).

The 2025 hybrid NMS-OSD framework broadens the scope from LDPC to BCH and RS codes. For LDPC $K$00, NMS+DDS1 closes to within $K$01 dB of ML at $K$02, DDS2 overlaps the ML bound up to about $K$03, and at $K$04 the approximate SNR requirements are $K$05 dB for NMS$K$06 alone, $K$07 dB for NBP-D$K$08, $K$09 dB for NMS+DDS1, and $K$10 dB for MRB$K$11 order-4 OSD (Li et al., 29 Sep 2025). At the same FER, the reported 6-bit-quantized complexity is about $K$12 BOPS for NMS$K$13+MRB$K$14 OSD and about $K$15 for NMS$K$16+DDS2, while the throughput advantage arises because only a few percent of frames invoke the serial OSD stage (Li et al., 29 Sep 2025).

The statistical-rule literature reports a different but compatible complexity profile. For an eBCH $K$17 code with order-1 OSD, HISR at $K$18 reduces the TEP count from $K$19 to about $K$20 with less than $K$21 dB extra loss, and HDR can reduce the TEP count to $K$22-$K$23 at high SNR with less than $K$24 dB loss; more generally, the six hard and soft stopping or discarding rules reduce average TEPs by up to $K$25-$K$26 at sub-dB penalty (Yue et al., 2020). This reinforces a central point across the literature: the practical value of reinforced OSD lies less in changing the optimality criterion than in reducing how often expensive reprocessing is invoked and how far the TEP traversal must proceed.

Reinforcement mode	Representative realization	Reported purpose
Offset-damped iterative front end	modified BP with $K$27 and $K$28 (Zhang et al., 2023)	reduce errors in MRPs before OSD
Learned reliability refinement	DIA CNN or dense network on NMS trajectory (Li et al., 2023, Li et al., 2024, Li et al., 29 Sep 2025)	improve MRB/MRP construction
Path control	adaptive or fixed decoding paths, ALMLT-based updates (Li et al., 2023, Li et al., 2024, Li et al., 29 Sep 2025)	prioritize historically useful TEP blocks
Termination and pruning	sliding-window NN, auxiliary pruning, stopping/discarding rules (Li et al., 2024, Li et al., 29 Sep 2025, Yue et al., 2020)	reduce TEP count, latency, or re-encodings

Two misconceptions are explicitly contradicted by the cited work. First, reinforced OSD is not restricted to neural models: offset-damped BP reinforcement and statistically derived stopping/discarding rules are both non-neural (Zhang et al., 2023, Yue et al., 2020). Second, reinforcement does not simply mean increasing OSD order: one of the central claims in the mBP-OSD literature is that order-$K$29 decoding with enhanced reliabilities can achieve the performance of order-$K$30 OSD (Zhang et al., 2023). A plausible implication is that reinforced OSD is best understood as a family of reliability-management and search-management strategies layered on top of the classical OSD substrate, rather than as a single decoder architecture.