REVIEW 3 major objections 5 minor 1 cited by
On the High-Rate FDPC Codes: Construction, Encoding, and a Generalization
T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read The paper claims that FDPC codes built from stacked permuted bidiagonal base matrices need only five normalized min-sum iterations to match or beat 5G LDPC and polar BP decoders running fifty iterations.
desk verdict The Section III girth-6 theorem is false — the paper's own Example 2 contains a 4-cycle — so the theoretical foundation collapses, though the simulation evidence may still be worth a corrective review. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the base parity-check matrix $H_{\mathrm{base}}$: a $2t$-row matrix whose columns each contain exactly two ones, arranged in column groups with gaps of 0, 2, 4, ... zeros between the ones. The algorithm forms a submatrix $C$ from the columns beyond position $m_{\mathrm{size}} = 2t(\mathrm{num\_per}+1)$, applies independent random column permutations to $C$, and stacks each permuted copy in a new block of $2t$ rows. The first $m_{\mathrm{size}}$ columns are then overwritten by a bidiagonal matrix $A$, which makes each parity bit the xor of a row of the remaining part with the previous parity bit, so encoding is a simple sequential pass. Finally, columns beyond $m_{\mathrm{size}}$ are deleted to reach the target blocklength $N$. The base-2 variant changes which column groups are kept, giving a base graph with shortest cycle 6 instead of 4.
What would settle it
Compute the girth of the published parity-check matrices for the reported codes, or list their minimum-weight codewords. If the final (256,164) or (1024,844) matrices contain a 4-cycle for the base-2 construction, or if a weight-4 codeword exists, then the assumed minimum distance of the final code is not 6 and the high-SNR error floor would likely be higher than claimed.
Extended reading notes
Core claim
The paper's discovery is that a carefully ordered base parity-check matrix, in which each column has exactly two ones spaced by gaps of 0, 2, 4, ... zeros, can be turned into a high-rate code by stacking random column permutations of its tail and deleting columns to hit a target blocklength. The resulting FDPC codes are claimed to decode to a given FER with far fewer normalized min-sum iterations than standard decoders need for 5G LDPC and polar codes. A systematic encoder works by reshaping the first $m_{\mathrm{size}}$ columns into a bidiagonal matrix $A$, so parity bits are computed one after another with a single bit of state. The generalization to base 2 removes columns whose ones are separated by 2, 6, 10, ... zeros, leaving a base graph whose shortest cycle has length 6, which corresponds to a base code with minimum distance 6.
Load-bearing premise
The construction assumes that the base matrix's good shortest-cycle length survives the random column permutations and column deletions in the final stacked parity-check matrix, even though only the base matrix's cycles are analyzed.
Editorial extensions
If this is right
- At short block lengths of 128 to 256, FDPC codes reach a given FER with 5 normalized min-sum iterations instead of 50, so decoding latency drops by roughly an order of magnitude for the same or better reliability.
- At (1024,844), the 12-iteration FDPC decoder matches CA-SCL polar decoding with list size 8 and 50-iteration 5G LDPC decoding, replacing a high-latency list decoder with a low-latency message-passing decoder.
- The bidiagonal parity structure gives a systematic encoder whose per-parity-bit cost is a constant number of xor operations, making high-rate FDPC codes encodable at linear complexity.
- The base-2 generalization extends the family to base graphs with minimum cycle length 6, promising a minimum distance of 6 in the base code and better high-SNR behavior if the final graph preserves that property.
- For very long blocklengths near 16384, the construction reports FER around $10^{-8}$ with 12 iterations, indicating that the fast-convergence benefit persists at scale.
Reading between the lines
- The paper leaves implicit that the final stacked parity-check graph may contain 4-cycles even when the base graph has shortest cycle 6, because the permutations and column deletions are not analyzed; checking the final girth is a direct test of whether the true minimum distance is actually 6.
- Because the decoder's fast convergence is demonstrated mainly down to FER $10^{-3}$ for the short codes, the paper leaves open what happens at lower error rates; a small stopping set in the final graph could produce an error floor above the promised operating point.
- The construction could be tested with the random permutations replaced by deterministic cyclic shifts, allowing a search over permutation choices to maximize the final graph's girth and potentially improve minimum distance while preserving the encoder.
- The claimed latency advantage could be quantified more directly by measuring wall-clock time or hardware cycle count of the 5-iteration FDPC decoder against the 50-iteration LDPC and pipelined polar BP decoders, rather than by iteration count alone.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes constructions of Fair-Density Parity-Check (FDPC) codes, a low-complexity systematic encoder based on a bidiagonal submatrix, a generalized Base-II construction aimed at minimum cycle length 6, and numerical comparisons with 5G LDPC and polar codes. The central advertised result is that FDPC codes achieve comparable or better FER/BER with far fewer decoding iterations than the 5G benchmark codes. The paper includes concrete construction algorithms, a GitHub repository with code, and simulations for several block lengths and rates.
Significance. If the performance claims hold, the paper would demonstrate an attractive high-rate code family with fast message-passing convergence, a useful practical property for low-latency applications. The encoding scheme is concrete and the authors provide reproducible code, which is a strength. However, the structural foundation of the paper is not sound as written: the claimed girth-6 property of the Base-II matrices is false, and no girth or minimum-distance analysis is provided for the final parity-check matrices actually used in the simulations. The simulation-based performance claims also lack statistical detail. The contributions are potentially valuable, but the paper needs substantial revision before the claims can be accepted.
major comments (3)
- [Section III, Example 2 and the definition of H_base-(t,2)] The claim that H_base-(5,2) has minimum cycle length 6 is false. In the displayed matrix, columns 1, 6, 10, and 11 have supports {1,2}, {6,7}, {1,6}, and {2,7}, respectively; their mod-2 sum is zero, so they form a weight-4 codeword in the base code and a 4-cycle in the Tanner graph. For every t≥5, the same supports appear in columns 1, 6, 2t, and 2t+1 of H_base-(t,2), so the advertised d=6 property fails for the entire family. This directly contradicts the statement that the base code has minimum distance 6 and invalidates the Section III generalization as stated.
- [Algorithm 1, final parity-check matrix] The paper does not establish any cycle or minimum-distance property for the final parity-check matrix H obtained after overwriting the first m_size columns with the bidiagonal encoder form (lines 13–17), appending permuted copies of the base submatrix (lines 5–11), and deleting columns (lines 19–21). These operations can create shorter cycles than those present in the base matrix, even when the base matrix has the claimed girth. Since the base-matrix d=6 claim is already false, and since the final H is the object actually decoded in Section IV, the error-floor behavior is unsupported. An analysis of the girth and minimum distance of the final code for the simulated parameters is needed before the performance advantages can be attributed to the construction.
- [Section IV, Figs. 3–8] The headline performance claims—for example, the approximately 0.5 dB gain over 5G LDPC and polar codes at FER 10^-3—rest entirely on plotted FER/BER curves. The manuscript does not report the number of simulated frames, the number of error events at each SNR, or confidence intervals. At FER 10^-3 with steep curves, a small number of observed frames can shift the apparent crossover by several tenths of a dB. Please add simulation statistics or error bars, or soften the claims to match what the plots actually support.
minor comments (5)
- [Section IV-F] The text says 'FDPC(16384,15661)', but the code dimension should be 15660 based on the stated row count of 724 and block length 16384; this appears to be a typo.
- [Section II-A] The statement 'for a given t, blocklength N≤t^2' is written for Base-I only. For Base-II the base matrix has t(t+1)/2 columns, so the blocklength bounds and column-deletion behavior should be described separately for each base.
- [Algorithm 2] The variable 'gap' is used inconsistently: in lines 3–10 the gap counts zeros between the two ones, while line 22 tests 'current_gap+1' as the distance between one positions. Making the definition consistent between the prose and the pseudocode would avoid confusion.
- [Section I] The introduction says the paper will 'discuss the ECP of FDPC codes after puncturing', but the column-removal operation in Algorithm 1 is shortening (it reduces the block length and dimension together), not puncturing in the standard sense. Please clarify the terminology or add the intended puncturing analysis.
- [Fig. 2b caption] The caption states 'minimum cycle of length 6', which is contradicted by Example 2 as explained in Major Comment 1; the figure and text must be updated to match a corrected construction or analysis.
Circularity Check
No significant circularity; simulation claims are measured against external standard codes, with only non-load-bearing self-citations.
full rationale
The central error-correction claims are simulation measurements, not derived from fitted parameters. The construction parameters (t, num_per, N) are chosen to hit target rates and blocklengths, such as t=23, num_per=1, N=256 for FDPC(256,164), and the FER curves are benchmarked against 5G LDPC and polar codes using external references [14], [16], [17]. No fitted FER parameter is relabeled as a prediction. The encoding algorithm is an explicit low-complexity construction based on a bidiagonal [A|B] parity-check matrix, and it does not presuppose the performance it claims to enable. The only self-citations are [13], the prior FDPC base matrix, which the paper explicitly acknowledges as equivalent to a rearrangement of the earlier construction, and [15], a peripheral observation about bidiagonal matrices in PAC codes. Neither is invoked as a uniqueness theorem, nor used to forbid alternative designs, and the present simulations are independent evidence for the ECP claims. Separately, the Section III girth-6 claim for Hbase-(5,2) appears to be a correctness risk, but a false or unproved structural assertion is a mathematical defect rather than a circular derivation, so it does not raise the circularity score.
Assumptions & free parameters
assumptions (3)
- standard math For a binary linear code whose parity-check matrix has columns of weight 2, the code is the cycle space of the graph, and the minimum distance equals the graph girth.
- domain assumption The full parity-check matrix, built by cascading random permutations of the base submatrix and deleting columns, inherits a sufficiently large girth to avoid error floors.
- domain assumption The BI-AWGN channel model and normalized min-sum decoding settings match the 5G LDPC and polar baseline implementations used for comparison.
Cite this review
Pith. "Pith review of On the High-Rate FDPC Codes: Construction, Encoding, and a Generalization." pith.science (2026). https://pith.science/paper/3VP57RZY
@misc{pith2026250611345,
author = {Pith},
title = {Pith review of: On the High-Rate FDPC Codes: Construction, Encoding, and a Generalization},
year = {2026},
howpublished = {\url{https://pith.science/paper/3VP57RZY}},
note = {Machine review of arXiv:2506.11345}
}
read the original abstract
Recently introduced Fair-Density Parity-Check (FDPC) codes, targeting high-rate applications, offer superior error-correction performance (ECP) compared to 5G Low-Density Parity-Check (LDPC) codes, given the same number of message-passing decoding iterations. In this paper, we present a novel construction method for FDPC codes, introduce a generalization of these codes, and propose a low-complexity encoding algorithm. Numerical results demonstrate the fast convergence of the message-passing decoder for FDPC codes.
Figures
Figures from the paper (4 more)
Forward citations
Cited by 1 Pith paper
-
PAC codes with Bounded-Complexity Sequential Decoding: Pareto Distribution and Code Design
Proves that PAC codes with rate profiles below polarized cutoff rates have Pareto-bounded sequential decoding complexity, and demonstrates a code built on this rule that gains over 0.75 dB at FER 1e-5.
Reference graph
Works this paper leans on
-
[1]
Low-density parity-check codes,
R. Gallager, “Low-density parity-check codes,”IRE Transactions on Information Theory, vol. 8, no. 1, pp. 21–28, 1962
1962
-
[2]
Good codes based on very sparse ma- trices,
D. J. MacKay and R. M. Neal, “Good codes based on very sparse ma- trices,” inIMA International Conference on Cryptography and Coding. Springer, 1995, pp. 100–111
work page 1995
-
[3]
NR; multiplexing and channel coding,
3GPP, “NR; multiplexing and channel coding,” Tech. Rep. TS 38.212, Rel. 15, June 2018. [Online]. Available: http://www.3gpp.org/ DynaReport/38-series.htm
work page 2018
-
[4]
On the road to 6G: Visions, requirements, key technologies, and testbeds,
C.-X. Wanget al., “On the road to 6G: Visions, requirements, key technologies, and testbeds,”IEEE Communications Surveys & Tutorials, vol. 25, no. 2, pp. 905–974, 2023
work page 2023
-
[5]
Spatially coupled LDPC codes constructed from protographs,
D. G. M. Mitchell, M. Lentmaier, and D. J. Costello, “Spatially coupled LDPC codes constructed from protographs,”IEEE Transactions on Information Theory, vol. 61, no. 9, pp. 4866–4889, 2015
work page 2015
-
[6]
LDPC block and convolutional codes based on circulant matrices,
R. Tanner, D. Sridhara, A. Sridharan, T. Fuja, and D. Costello, “LDPC block and convolutional codes based on circulant matrices,”IEEE Transactions on Information Theory, vol. 50, no. 12, pp. 2966–2984, 2004
work page 2004
-
[7]
Deriving good LDPC convolutional codes from ldpc block codes,
A. E. Pusane, R. Smarandache, P. O. V ontobel, and D. J. Costello, “Deriving good LDPC convolutional codes from ldpc block codes,” IEEE Transactions on Information Theory, vol. 57, no. 2, pp. 835–857, 2011
work page 2011
-
[8]
Quasi-cyclic LDPC codes based on pre-lifted protographs,
D. G. M. Mitchell, R. Smarandache, and D. J. Costello, “Quasi-cyclic LDPC codes based on pre-lifted protographs,”IEEE Transactions on Information Theory, vol. 60, no. 10, pp. 5856–5874, 2014
2014
Show all 17 references
-
[9]
Spatially coupled sparse codes on graphs: theory and practice,
D. J. Costello, L. Dolecek, T. E. Fuja, J. Kliewer, D. G. Mitchell, and R. Smarandache, “Spatially coupled sparse codes on graphs: theory and practice,”IEEE Communications Magazine, vol. 52, no. 7, pp. 168–176, 2014
2014
-
[10]
Design and analysis of NB QC-LDPC codes over small alphabets,
I. E. Bocharovaet al., “Design and analysis of NB QC-LDPC codes over small alphabets,”IEEE Transactions on Communications, vol. 70, no. 5, pp. 2964–2976, 2022
2022
-
[11]
Channel polarization: A method for constructing capacity- achieving codes for symmetric binary-input memoryless channels,
E. Arıkan, “Channel polarization: A method for constructing capacity- achieving codes for symmetric binary-input memoryless channels,”IEEE Transactions on Information Theory, vol. 55, no. 7, pp. 3051–3073, 2009
2009
-
[12]
List decoding of polar codes,
I. Tal and A. Vardy, “List decoding of polar codes,”IEEE Transactions on Information Theory, vol. 61, no. 5, pp. 2213–2226, 2015
2015
-
[13]
High-rate fair-density parity-check codes,
H. Mahdavifar, “High-rate fair-density parity-check codes,” 2024. [Online]. Available: https://arxiv.org/abs/2402.06814
2024 arXiv
-
[14]
Design of low-density parity check codes for 5G new radio,
T. Richardson and S. Kudekar, “Design of low-density parity check codes for 5G new radio,”IEEE Communications Magazine, vol. 56, no. 3, pp. 28–34, 2018
2018
-
[15]
Polarization-adjusted convolutional (PAC) codes as a concatenation of inner cyclic and outer polar- and Reed-Muller-like codes,
M. Moradi, “Polarization-adjusted convolutional (PAC) codes as a concatenation of inner cyclic and outer polar- and Reed-Muller-like codes,”Finite Fields and Their Applications, vol. 93, p. 102321, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S...
2024
-
[16]
Polar codes: A pipelined implementation,
E. Arıkan, “Polar codes: A pipelined implementation,” inProc. 4th ISBC, vol. 2010, 2010, pp. 11–14
2010
-
[17]
Early stopping criteria for energy-efficient low-latency belief-propagation polar code decoders,
B. Yuan and K. K. Parhi, “Early stopping criteria for energy-efficient low-latency belief-propagation polar code decoders,”IEEE Transactions on Signal Processing, vol. 62, no. 24, pp. 6496–6506, 2014
2014
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.