Improved Rate-versus-Distance Upper Bounds for LDPC Codes

Abstract: LDPC codes play a vital role in coding theory and practical error correction. A central problem in this direction is to understand their rate--distance tradeoff. In this paper, we introduce a new framework for estimating ball sizes in the coset graphs of LDPC codes. The key new object is the coset-weight generating function, which encodes the minimum Hamming weights of all cosets of a linear code. Rather than estimating coset balls directly, we upper-bound this generating function through a local growth analysis for codes spanned by low-weight vectors. This framework sharpens the previous ball-size estimate of Iceland and Samorodnitsky. Combined with a general method of Friedman and Tillich that relates balls in coset graphs to sizes of error-correcting codes, it further improves the upper bounds on the rate of LDPC codes for a significant range of relative distances.

• The paper introduces a coset-weight generating function framework to obtain tighter rate–distance upper bounds for binary LDPC codes.
• It combines analytic estimates with recursive shortening and coset graph techniques to improve bounds, especially for small parity-check weights (w=3,4).
• The work narrows the rate–distance gap for LDPC codes, influencing both theoretical insights and practical high-reliability design constraints.

Improved Rate-versus-Distance Upper Bounds for LDPC Codes

Introduction and Problem Statement

This work investigates the asymptotic rate-versus-distance tradeoff for binary LDPC codes. The rate–distance function $R(\delta)$ is classically characterized for general codes via the Gilbert–Varshamov lower bound and the MRRW linear programming upper bound, but much less is known for LDPC codes, i.e., linear codes with parity-check matrices of bounded row weight (density). Given the high practical and theoretical relevance of LDPC codes, deriving tight upper bounds on the attainable rate as a function of relative minimum distance $\delta$ and parity-check density $w$ remains a central challenge. Existing results provide bounds either through combinatorial ball-packing, linear programming, or recursive shortening techniques. However, there is still a substantial gap between achievable lower bounds and the best upper bounds available for LDPC codes, especially in the regime of moderate-to-large minimum distance and fixed, small $w$.

Prior Work and Methodological Landscape

The first approaches for rate–distance upper bounds for LDPC codes were primarily based on adaptations of classical bounds for general codes. The Plotkin bound shows zero rate for $\delta \ge 0.5$, and the MRRW linear programming bound applies to all linear codes. Improved upper bounds tailored for LDPC codes were obtained by Burshtein et al., and further enhanced using recursive shortening and combinatorial arguments by Ben-Haim and Litsyn. Iceland and Samorodnitsky introduced the use of coset graphs to relate the geometry of LDPC dual stabilizer codes to upper bounds on code ball sizes, sharpening the corresponding rate upper bounds, particularly for codes with small parity-check weights ($w=3,4$).

The techniques in these works reduce the problem to bounding the sizes of Hamming balls in specific coset graphs, utilizing combinatorial and algebraic tools, ultimately leading to explicit bounds such as:

$$R_w(\delta) \leq H(\rho) - \frac{\log_2 e}{8 w^2} \left( \frac{\rho^w}{2} \right)^{w+1}$$

where $\rho = \tfrac{1}{2} - \sqrt{\delta(1-\delta)}$.

Main Contributions

This paper introduces a new analytic-combinatorial framework based on the coset-weight generating function $Q_C(\lambda)$, which aggregates the minimum Hamming weights of all cosets of a linear code $C$. This framework allows for precise local analysis of the growth of $\delta$0 as low-weight parity generators are added, generalizing the coset-graph ball size argument to an analytic generating function setting. The core technical achievement is a tight upper bound for $\delta$1 for codes generated by vectors of small weight:

$\delta$2

for a code $\delta$3 generated by vectors of weight at most $\delta$4 and covering all coordinates. For $\delta$5, sharper bounds are proven reflecting exact local combinatorics.

This bound is then combined with the Friedman–Tillich result connecting coset graph ball sizes to code size, replacing the previous role of Hamming ball sizes with the analytic estimates derived via the generating function. Choosing the exponential generating parameter $\delta$6 appropriately in terms of the target ball-weight $\delta$7, the authors obtain explicit, closed-form upper bounds for the asymptotic LDPC rate–distance tradeoff $\delta$8:

$\delta$9

with further sharpenings for $w$0.

Key Technical Results and Claims

• General Coset-Weight Generating Function Bound: For codes generated by low-weight vectors and covering the full coordinate set, the paper obtains an exponential-type analytic upper bound on $w$1 that holds tightly for all $w$2. The $w$3 case is shown to be essentially sharp by explicit construction.
• Improved Explicit Rate Bounds: Substitution of these analytic bounds into the rate–distance machinery yields strictly improved upper bounds on $w$4 over prior combinatorial methods for all $w$5, and for all $w$6. In the case $w$7, the bound tightens substantially over previous results for high relative distances.
• Stronger Results for Small $w$8: For $w$9, under a threshold regime on $w$0 (equivalently $w$1), the authors provide an improved bound expressible in base-4 entropy functions, verifying strict improvement over Iceland–Samorodnitsky for $w$2.
• Refined Shortening Bound Recursion: The improved explicit bounds can be substituted into recursive shortening arguments, leading to the improved bounds being reflected under iteration, further narrowing the known achievability-region for LDPC codes.

Theoretical and Practical Implications

The analytic approach using the coset-weight generating function represents an important methodological advance. It shifts the focus from direct ball size counting to generating function methods, enabling sharper control over the combinatorics of coset weight distributions in codes generated by low-weight vectors. Particularly, these improvements provide the best-known asymptotic rate upper bounds for binary LDPC codes across a broad range of distances and all $w$3.

From a practical standpoint, new bounds tighten the provable limitations for LDPC code constructions designed for high-reliability channels, especially in regimes where small $w$4 is mandated by either decoding complexity constraints or hardware considerations. These results interface directly with practical LDPC code design, influencing the theoretical performance ceiling of such codes.

On the theoretical side, the work unifies and generalizes several strands of previous research, bringing together coset graphs, analytic generating functions, and recursive shortening in a unified argument. This opens new avenues for further improvements, for instance by targeting the extremal structure of codes attaining the upper bounds on $w$5 or by extending these ideas to non-binary LDPC codes or codes over irregular graphical models.

Open Problems and Future Directions

A key open conjecture formulated is that, among all codes generated by weight-$w$6 vectors and covering all coordinates, the coset-weight generating function is maximized by the direct sum of codes on disjoint supports, i.e., the “partition code.” Verification of this conjecture in general, or tightening the analytic bounds further for $w$7, are promising directions. There is also potential for leveraging algebraic or spectral techniques to further refine upper bounds for $w$8, as well as for extending the analytic machinery to study local testability and list-decoding capacities in LDPC or more general code ensembles.

Conclusion

This work establishes new, strictly improved upper bounds on the achievable rate as a function of distance for LDPC codes, leveraging a novel analytic framework based on coset-weight generating functions. By reducing the asymptotic growth of coset-graph balls to tractable extremal problems and establishing near-tight analytic bounds, the results close a significant gap in the characterization of LDPC code performance. These findings not only enhance our understanding of the combinatorial-geometric constraints on LDPC codes but also offer methodological innovations likely to have broader impact across coding theory and related combinatorial optimization settings.

Reference: "Improved Rate-versus-Distance Upper Bounds for LDPC Codes" (2605.01213)