On Parameter Optimization of Product Codes for Iterative Bounded Distance Decoding with Scaled Reliability

Alex Alvarado; Alexandre Graell i Amat; Alireza Sheikh; Gianluigi Liva

REVIEW 3 major objections 4 minor 15 references

On Parameter Optimization of Product Codes for Iterative Bounded Distance Decoding with Scaled Reliability

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Binary product codes with 3-bit-error-correcting BCH component codes offer the best performance-complexity trade-off under iterative bounded distance decoding with scaled reliability (iBDD-SR) for all overheads from 6.25% to 33.33%.

desk verdict A useful, practical design rule (t=3 for iBDD-SR) that is real but rests on a load-bearing, unproven shortening approximation; worth refereeing, not blindly trusting. read the letter →

arxiv 1908.09502 v1 pith:3BK4OXS6 submitted 2019-08-26 cs.IT math.IT

Alireza Sheikh , Alexandre Graell i Amat , Gianluigi Liva , Alex Alvarado This is my paper · ORCID

classification cs.ITmath.IT MSC 94B1594B35

keywords boundeddistancedecodingdensityevolutionproductcodeshigh-throughputfiber-opticsystemslow-complexityiBDD-SRBCHcomponentshortened

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks how to choose the component code of a binary product code when decoding with iBDD-SR, a soft-aided scheme whose complexity and data flow stay close to hard-decision decoding. Using density evolution, it optimizes the BCH component-code parameters (Galois-field size, error-correcting capability $t$, and shortening length) for overheads between 6.25% and 33.33%. The central result is that $t=3$, not $t=4$ or larger, gives the best performance-complexity trade-off for every overhead in this range, and the optimized codes come close to an ideal decoder that never miscorrects. The practical payoff would be a decoder that works near the waterfall of the code at moderate power, which matters for high-throughput fiber-optic links.

What carries the argument

iBDD-SR (iterative bounded distance decoding with scaled reliability) is the mechanism: each component-code BDD output is mapped to $\pm1$ on success and to $0$ on failure, then combined with the channel log-likelihood ratio through a per-iteration scaling factor $w_l$ before being passed to the other dimension. The density evolution for this decoder, which tracks the outbound error probability of the BCH component decoders and relies on the component code's weight enumerator, gives the threshold used for optimization. To handle shortened BCH codes, the paper uses the approximation that the average numbers of input and output errors match those of the non-shortened code after rescaling by the length ratio.

What would settle it

Simulate the shortened BCH code at 33.33% overhead (for example, $v=8$, $t=3$, $s=76$) and measure its output error probability against the length-scaled non-shortened prediction used in the density evolution; if the mismatch is large enough to change which code has the lower threshold, the $t=3$ recommendation fails. Alternatively, compare $t=3$ and $t=4$ PCs at 33.33% overhead at a post-FEC BER of $10^{-12}$ with measured energy per decoded bit: the recommendation fails if $t=4$ meets the target with less energy.

Watch

Extended reading notes

Core claim

The paper's claim, on its own terms, is that for binary product codes decoded with iBDD-SR and composed of BCH component codes, the error-correcting capability $t=3$ is optimal across the full range of overheads $1/i$ for $i=3,\ldots,16$. Density-evolution thresholds are computed by tracking the output error probability of the row and column decoders over iterations; because the weight enumerator of a shortened BCH code is not known in general, the paper adapts the DE to shortened codes by scaling the input error probability by the factor $(2^v-1)/(2^v-1-s)$. Over the search space $v\in\{8,9,10,11,12\}$, $t\in\{3,4\}$, no code with $v=11$ or $v=12$ achieves the best threshold. The optimized $t=3$ codes approach the performance of ideal iBDD (a miscorrection-free genie), and the Gaussian-noise-model WDM QPSK estimate gives optical-reach enhancements up to 5.8% over conventional iBDD.

Load-bearing premise

Shortened BCH component codes behave like their non-shortened parents once the input error rate is scaled by the ratio of the two code lengths.

Editorial extensions

If this is right

Across the 6.25% to 33.33% overhead range, a system designer can pick a $t=3$ BCH component code and stay in the waterfall region close to genie-aided iBDD without searching over $t=4$ or larger $v$.
The search space for optimal component codes can be restricted to $v=8,9,10$; codes over $\mathrm{GF}(2^{11})$ and $\mathrm{GF}(2^{12})$ never give the best threshold.
At 33.33% overhead, the $t=4$ code gives a steeper BER slope but less than 0.05 dB gain over $t=3$ at the waterfall, so $t=3$ saves complexity without a meaningful performance penalty.
For the modeled WDM QPSK link, switching from iBDD to iBDD-SR with the optimized $t=3$ code adds up to 5.8% to the reach (560 km in the 11.11% overhead example) because of the waterfall improvement.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The length-ratio shortening approximation is untested for severe shortening; if it fails, the optimal shortening parameter and possibly $t$ could shift, so a direct DE for shortened codes is the natural next check.
Below the target post-FEC BER of $10^{-15}$, $t=4$ codes with steeper BER slopes may overtake $t=3$; the recommendation should be re-derived for each operating point rather than extrapolated.
Because iBDD-SR treats rows and columns symmetrically, the same $t=3$ parameter optimization plausibly transfers to other product-like constructions such as staircase or braided codes, but that transfer would need its own DE and simulations.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague, here's the short version: this paper gives designers a concrete rule for iBDD-SR product codes — use t=3 BCH components across 6.25% to 33.33% overhead — and backs it with a DE-based parameter search that explains why t=3 wins in the waterfall region. That rule is useful and, as far as the simulations go, plausible. But the DE leans on an unproven shortening approximation, and the paper's own simulations show t=4 codes have steeper BER slopes and catch up at low BER, so the trade-off claim is only as strong as that approximation.

What is actually new: adapting the DE from [7] to shortened BCH component codes. The WE of shortened BCH codes is unknown, so they scale the average input error count by (2^v-1)/(2^v-1-s) and assume the same outbound behavior. That is a reasonable first-order heuristic, and the paper says so. But it is not derived, and it silently carries the DE: every threshold in Table 1 passes through it. The DE in [7] depends on the full weight enumerator; shortening removes codewords whose support intersects the shortened positions, which changes both miscorrection probability and the distribution of output reliabilities that iBDD-SR uses. The paper does not compare the heuristic against the exact WE for any code, nor give a sensitivity analysis. For a design rule meant to be used across a wide OH range, that's a real gap.

What the paper does well: the search is disciplined, the search space is practical (v=8..10, t=3..4), the simulations check the DE predictions for a few OHs, and the reach gain numbers (5.8% over iBDD) are honestly derived via the GN model. The authors also flag the finite-length slope issue themselves — C4 beats C3 at 10^-6 despite a worse threshold — which is more honest than most code papers.

The main soft spot is the load-bearing approximation; it is not fatal, because the qualitative trend (t=3 good) may survive a more careful analysis, but the paper currently overstates "best performance-complexity trade-off" as if it were established. The performance claim is really about the waterfall region; at lower target BERs the t=4 codes come back. That should be in the abstract or conclusion.

This paper is for the fiber-optic FEC community, especially people choosing component codes for iBDD-SR decoders. It will be useful as a reference design table even if the approximation is later tightened. Worth a serious referee. The core idea is sound enough and the result is practically relevant; the referee should ask for better validation of the shortening approximation (exact WE for at least one shortened code, or a Monte Carlo check across more OHs) and a qualification of the waterfall-only claim. I'd accept it for further review, not desk reject.

Referee Report

3 major / 4 minor

Summary. The paper addresses parameter optimization of binary product codes (PCs) decoded by iterative bounded-distance decoding with scaled reliability (iBDD-SR). It proposes to use density evolution (DE) from the authors' prior work, adapted to shortened BCH component codes through an average-error-count approximation, to optimize component-code parameters (v, t, s) for overheads from 6.25% to 33.33%. Table 1 gives optimized parameter sets for iBDD and iBDD-SR, and simulations for three representative overheads show that iBDD-SR approaches the genie-aided ideal iBDD decoder. The paper's central claim is that PCs with 3-bit-error-correcting component codes provide the best performance-complexity trade-off for all overheads considered.

Significance. If the central claim is correct, the paper provides actionable design guidance for high-throughput, energy-efficient fiber-optic FEC: iBDD-SR with t=3 BCH component codes can approach miscorrection-free performance at lower complexity than t=4 alternatives, which is significant given current interest in low-power coherent transceivers. The paper's strengths are its concrete optimized parameter table covering 14 overhead values, its use of an analytic DE tool rather than exhaustive simulation, and its transparent reporting of finite-length BER results that do not fully align with the asymptotic conclusion. The practical impact, however, is contingent on the validity of the Section III shortening approximation and on how the 'performance-complexity trade-off' is quantified, since the simulations themselves show t=4 codes with steeper BER slopes at BER=1e-6.

major comments (3)

[Section III, 'Density Evolution for PC Design'] The shortening approximation p^s_in = ((2^v - 1)/(2^v - 1 - s)) p_in and the corresponding statement that 'the same performance is expected' for shortened BCH codes is introduced as intuition without derivation. This approximation is load-bearing because the DE from [7] depends on the component-code weight enumerator and on the reliability statistics of BDD outputs, both of which change when a code is shortened. Since the exact weight enumerator of shortened BCH codes is admitted to be unknown, the DE thresholds in Table 1—and hence the t=3 recommendation—rest entirely on this approximation. The paper needs either a derivation or a careful validation, for example by comparing DE predictions against exact shortened-code DE where the weight enumerator can be computed numerically for small v, or by a sensitivity analysis showing that the optimized parameter choices are robust to plausible deviations from the approximation.
[Section 4, Figures 2 and 3] The conclusion that t=3 component codes provide the best performance-complexity trade-off for all overheads is not fully supported by the paper's own simulations. For OH=14.29%, Figure 2 shows that the t=4 code C4 has a steeper BER slope and yields a small improvement over the t=3 code C3 at BER=1e-6; for OH=33.33%, Figure 3 shows that C6 (t=4) outperforms C5 (t=3) at BER=1e-6 for both iBDD-SR and iBDD. The paper's defense is that C5 offers the better waterfall performance and much lower complexity, but the abstract and conclusion state the claim without this qualification. To make the claim precise, the authors should state the BER regime and the complexity metric used for the trade-off, and discuss whether the finite-length crossover observed at 1e-6 persists at the target post-FEC BER of 1e-15.
[Section 4, 'Numerical and Simulation Results'] The statement that t=3 gives the best threshold for iBDD-SR and all overheads is made only within a restricted search space: v in {8,9,10,11,12} and t in {3,4}, with v=11 and v=12 effectively excluded after an initial comparison. This is a reasonable practical restriction, but the conclusion should be worded as a statement about the searched parameter space rather than an unrestricted universal claim. In addition, the paper does not provide an explicit quantitative complexity model; complexity is discussed informally via code length and the efficient t=3 BCH implementation of [15]. A quantitative comparison—e.g., estimated decoder area, energy per bit, or number of operations—is needed to substantiate the phrase 'best performance-complexity trade-off'.

minor comments (4)

[Section 4, paragraph after Figure 1] The phrase 'stoping set' is a typo and should read 'stopping set'.
[Section 4, paragraph after Figure 3] The sentence 'C6 gives better better performance in the waterfall region than C5' contains a duplicated adverb and should be corrected.
[Section 3, first paragraph] The clause 'the WE of shortened BCH codes is not know in general' should read 'is not known in general'.
[Section 4, Figure 2 caption discussion] The notation C1 through C6 is introduced in the text but not in a single consolidated list; a table or a sentence listing all code parameters would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the DE-based parameter search is a discrete optimization whose t=3 conclusion is an output, not an input; the shortening approximation is a stated modeling assumption, not a definitional reduction.

full rationale

I find no circularity in this paper. The parameter optimization is carried out with the density evolution for iBDD-SR imported from the authors' prior work [7]. This is a self-citation, but it is not a fitted parameter and its stated assumptions do not include the target result that t=3 is optimal. The DE is a parameter-free asymptotic analysis of the iBDD-SR decoder for GLDPC ensembles, and it is therefore independent support for the optimization methodology. The only new modeling step for shortened BCH component codes is the average-error-count scaling p^s_in = (2^v - 1)/(2^v - 1 - s) * p_in, which the paper explicitly presents as an intuition rather than a derived identity. This is a simplifying assumption, not a circular definition: it does not force the outbound error probability or the threshold ranking. The optimized parameters in Table 1 are produced by a discrete search over (v, t, s), and the t=3 recommendation is an output of that search, not an input. The simulation comparisons against ideal iBDD and conventional iBDD provide external checks rather than fitted predictions. Any concern about the accuracy of the shortening approximation would be a correctness risk, not circular reasoning.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters are fitted in this paper; the design parameters (v, t, s) are searched. The main load-bearing assumption is the shortening approximation, which is ad hoc to this paper and not independently validated. The DE itself is inherited from prior work, including the authors' own [7].

assumptions (4)

domain assumption Density evolution for GLDPC ensembles accurately predicts the threshold of product codes.
Invoked in Section III, relying on [11]. This is the analytical basis for all threshold computations.
domain assumption The DE for iBDD-SR derived in [7] is correct.
The paper adapts this DE to shortened codes. Since [7] is the authors' own prior work, this is an assumption carried forward.
ad hoc to paper The shortening approximation (average number of input/output errors scales as (2^v - 1)/(2^v - 1 - s)) holds for both iBDD and iBDD-SR.
Stated in Section III as 'intuition'. It is used to compute DE thresholds for all optimized codes, but is not derived or validated beyond a few simulation points.
standard math The weight enumerator of non-shortened BCH codes is known and sufficient for the DE.
Mentioned in Section III as a known quantity; the DE relies on it.

how reviews work

0 comments

Cite this review

Pith. "Pith review of On Parameter Optimization of Product Codes for Iterative Bounded Distance Decoding with Scaled Reliability." pith.science (2026). https://pith.science/paper/3BK4OXS6

@misc{pith2026190809502,
  author       = {Pith},
  title        = {Pith review of: On Parameter Optimization of Product Codes for Iterative Bounded Distance Decoding with Scaled Reliability},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3BK4OXS6}},
  note         = {Machine review of arXiv:1908.09502}
}

read the original abstract

We use density evolution to optimize the parameters of binary product codes (PCs) decoded based on the recently introduced iterative bounded distance decoding with scaled reliability. We show that binary PCs with component codes of 3-bit error correcting capability provide the best performance-complexity trade-off.

Figures

Figures reproduced from arXiv: 1908.09502 by the authors.

**Figure 3.** Performance of PC using iBDD, ideal iBDD, and iBDD-SR for OH [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 14 canonical work pages

[7]

Sheikh, A., Graell i Amat, A., Liva, G.: ‘Binary message passing decoding of product-like codes’, 2019, Available from: https://arxiv.org/abs/1902.03575

work page Pith review arXiv 2019
[15]

P., Farhood, A., Hunt, A., et al.: ‘Staircase codes: FEC for 100 Gb/s OTN’, IEEE/OSA J

Smith, B. P., Farhood, A., Hunt, A., et al.: ‘Staircase codes: FEC for 100 Gb/s OTN’, IEEE/OSA J. Lightw. Technol., 2012, 30, pp. 110–117 4

work page 2012
[1]

Commun., 1998, 46, (8), pp

Pyndiah, R.M.: ‘Near-optimum decoding of product codes: block turbo codes’, IEEE Trans. Commun., 1998, 46, (8), pp. 1003–1010

work page 1998
[2]

Fougstedt, C., Larsson-Edefors, P.: ‘Energy-efﬁcient high-throughput VLSI architectures for product-like codes’, IEEE/OSA J. Lightw. Technol. , 2019, 37, (2), pp. 477–485

work page 2019
[3]

Optics, 2016, 18, (6), pp

Agrell, E., Karlsson, M., Chraplyvy, A.R., et al.: ‘Roadmap of optical communications’, J. Optics, 2016, 18, (6), pp. 063002

work page 2016
[4]

M., Kschischang, F

Zhang, L. M., Kschischang, F. R.: ‘Low-complexity soft- decision concatenated LDGM-staircase FEC for high- bit-rate ﬁber-optic communication’,IEEE/OSA J. Lightw. Technol., 2017, 35, (18), pp. 3991–3999

work page 2017
[5]

R.: ‘Low-complexity concatenated LDPC-staircase codes’, IEEE/OSA J

Barakatain, M., Kschischang, F. R.: ‘Low-complexity concatenated LDPC-staircase codes’, IEEE/OSA J. Lightw. Technol., 2018, 36, (12), pp. 2443–2449

work page 2018
[6]

Sheikh, A., Graell i Amat, A., Liva, G.: ‘Iterative bounded distance decoding of product codes with scaled reliability’, Proc. Eur. Conf. Opt. Commun. (ECOC), Rome, Italy, 2018, pp. 1–3

work page 2018

Show all 15 references

[8]

Lei, Y ., Chen, B., Liga, G., et al.: ‘Improved Decoding of Staircase Codes: The Soft-aided Bit- marking (SABM) Algorithm’, 2019, Available from: https://arxiv.org/abs/1902.01178

2019 arXiv
[9]

53rd Annu

Sheikh, A., Graell i Amat, A., Liva, G.: ‘Binary message passing decoding of product codes based on generalized minimum distance decoding’, Proc. 53rd Annu. Conf. Inf. Sciences and Systems (CISS). Baltimore, MD, 2019, pp. 1–5

2019
[10]

Optical Fiber Commun

Fougstedt, C., Sheikh, A., Graell i Amat, A., et al.: ‘Energy-efﬁcient soft-assisted product decoders’, Proc. Optical Fiber Commun. Conf. (OFC), San Diego, CA, 2019, pp. 1–3

2019
[11]

D., Narayanan, K

Jian, Y ., Pﬁster, H. D., Narayanan, K. R.: ‘Approaching capacity at high rates with iterative hard-decision decod- ing’,IEEE Trans. Inf. Theory, 2017, 63, pp. 5752–5773

2017
[12]

M., Kschischang, F

Zhang, L. M., Kschischang, F. R.: ‘Staircase codes with 6% to 33% overhead’, IEEE/OSA J. Lightw. Technol. , 2014, 32, pp. 1999–2002

2014
[13]

Sheikh, A., Graell i Amat, A., Liva, G.: ‘Achievable infor- mation rates for coded modulation with hard decision decoding for coherent ﬁber-optic systems’, IEEE/OSA J. Lightw. Technol., 2017, 35, (23), pp. 5069–5078

2017
[14]

D.: ‘Approaching miscorrection- free performance of product codes with anchor decod- ing’,IEEE Trans

Häger, C., Pﬁster, H. D.: ‘Approaching miscorrection- free performance of product codes with anchor decod- ing’,IEEE Trans. Commun., 2018, 66, pp. 2797–2808

2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [7]

Sheikh, A., Graell i Amat, A., Liva, G.: ‘Binary message passing decoding of product-like codes’, 2019, Available from: https://arxiv.org/abs/1902.03575

work page Pith review arXiv 2019

[2] [15]

P., Farhood, A., Hunt, A., et al.: ‘Staircase codes: FEC for 100 Gb/s OTN’, IEEE/OSA J

Smith, B. P., Farhood, A., Hunt, A., et al.: ‘Staircase codes: FEC for 100 Gb/s OTN’, IEEE/OSA J. Lightw. Technol., 2012, 30, pp. 110–117 4

work page 2012

[3] [1]

Commun., 1998, 46, (8), pp

Pyndiah, R.M.: ‘Near-optimum decoding of product codes: block turbo codes’, IEEE Trans. Commun., 1998, 46, (8), pp. 1003–1010

work page 1998

[4] [2]

Fougstedt, C., Larsson-Edefors, P.: ‘Energy-efﬁcient high-throughput VLSI architectures for product-like codes’, IEEE/OSA J. Lightw. Technol. , 2019, 37, (2), pp. 477–485

work page 2019

[5] [3]

Optics, 2016, 18, (6), pp

Agrell, E., Karlsson, M., Chraplyvy, A.R., et al.: ‘Roadmap of optical communications’, J. Optics, 2016, 18, (6), pp. 063002

work page 2016

[6] [4]

M., Kschischang, F

Zhang, L. M., Kschischang, F. R.: ‘Low-complexity soft- decision concatenated LDGM-staircase FEC for high- bit-rate ﬁber-optic communication’,IEEE/OSA J. Lightw. Technol., 2017, 35, (18), pp. 3991–3999

work page 2017

[7] [5]

R.: ‘Low-complexity concatenated LDPC-staircase codes’, IEEE/OSA J

Barakatain, M., Kschischang, F. R.: ‘Low-complexity concatenated LDPC-staircase codes’, IEEE/OSA J. Lightw. Technol., 2018, 36, (12), pp. 2443–2449

work page 2018

[8] [6]

Sheikh, A., Graell i Amat, A., Liva, G.: ‘Iterative bounded distance decoding of product codes with scaled reliability’, Proc. Eur. Conf. Opt. Commun. (ECOC), Rome, Italy, 2018, pp. 1–3

work page 2018