Confusion-Erasure Bounds of Error-Bounded Decoders under QAM
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.
Tom: I'm Tom.
Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.
Tom: We're looking at the paper: "Confusion-Erasure Bounds of Error-Bounded Decoders under QAM".
Jane: Here's the gist of the paper. Block error rate (BLER) conflates block confusions (undetected errors) and block erasures, which have fundamentally different impacts on system reliability.
Tom: This paper extends the confusion and erasure analysis of error-bounded decoders to square quadrature amplitude modulation (QAM) constellations in the finite blocklength (FBL) regime.
Jane: As 6G moves beyond 5G’s reliability targets, supporting latency- and reliability-sensitive services such as vehicular control and industrial automation at short blocklengths, the accurate characterization of physical (PHY)-layer decoding failures becomes essential.
Tom: Block error rate (BLER) conflates two fundamentally distinct error types: block erasures, in which the receiver detects the failure and can request retransmission, and block confusions (undetected errors), in which the decoder silently commits to an incorrect codeword.
Paper summary: Tom: So we're looking at this paper called "Confusion-Erasure Bounds of Error-Bounded Decoders under QAM" and what it claims is that block error rate, or BLER, mixes up two different types of errors: undetected errors from confusion and erasures where the receiver knows something went wrong.
Jane: That's the core problem they tackle because those two error types have totally different impacts on how reliable a system is, especially for things like self-driving cars or industrial control where one kind of error is way worse than the other.
Tom: The paper extends the confusion and erasure analysis to square QAM constellations in this finite blocklength regime, which is super important because it's what we deal with at short blocklengths in 6G <ref:2609.12631#pg1,paper extends the confusion and erasure analysis>.
Lu: What they do is handle QAM's heterogeneous symbol energies, which means the Euclidean distance between two symbols isn't just one fixed number; it’s a whole distribution of possibilities instead of a single value.
Meng: That sounds messy to model at scale because you can’t just use one distance metric for everything, so they have to find a way to deal with that variability.
Lalam: The paper says they handle this by collapsing that distribution down to the root-mean-square distance and averaging the confusion over it, which is a key technique for bounding those errors.
Jane: So the main thesis is deriving analytical lower and upper bounds on the block confusion rate using this RMS distance approach.
Tom: They base their derivation on relating the squared Euclidean distance to the Hamming distance through a per-symbol squared-distance coefficient called delta, which is defined as two times the average symbol energy times one minus eta over M, as seen in equation thirteen.
Lu: That coefficient delta is crucial because it connects the two metrics; once that's fixed by the constellation and its symbol usage distribution, it becomes a constant for any codebook.
Jane: Then they use this to find a lower bound on confusion rate by only keeping the shell at the maximum achievable minimum distance, which they call dUB H min, which is the farthest least confusable shell.
Tom: And then there's their sensitivity analysis showing that this lower bound on confusion rate is monotonically decreasing and convex in terms of the average symbol energy, E bar s.
Meng: That convexity part is interesting because it tells us how sensitive our confusion bounds are to changes in the signal power we use.
Lalam: The upper bound on confusion rate is piecewise continuous with jump discontinuities at specific energy levels, and they also note that the blocklength n influences these bounds, with dUB H min increasing as n increases.
Jane: So, when they look at numerical results, they confirm that as the signal-to-noise ratio or redundancy goes up, both these confusion rate bounds fall many orders of magnitude below their target error epsilon.
Tom: And what's left is that detectable erasures are the dominant residual error, which is good because we can exploit those with things like hybrid automatic repeat request and network coding at high SNR.
Lu: It’s interesting because the confusion bounds translate directly into erasure-rate bounds where the confusion rate is much smaller than epsilon.
Meng: From a deployment standpoint, if the confusion rate drops that far below epsilon, it means our system can reliably manage those undetected errors by just using standard retransmission techniques rather than needing more complex error correction schemes for every possibility.
Jane: So to sum up this part of the paper, they show that under these conditions, the confusion bounds are much tighter than what's needed for reliability targets.
Tom: Moving on to the conclusion and implications of this paper called "Confusion-Erasure Bounds of Error-Bounded Decoders under QAM," we have to think about how this changes our view on error types in these systems.
Jane: The authors focus on showing that even with the complexity of QAM constellations, we can still get rigorous bounds for both block confusion and erasures in the finite blocklength regime.
Lu: They establish analytical bounds for these rates by using a specific way to represent the symbol-pair distribution induced by the codebook, focusing on the root-mean-square distance.
Tom: The implication is that for safety-critical applications where undetected errors are catastrophic, like vehicular control, we can have much more confidence in our system's reliability when we know exactly how much confusion we are dealing with.
Meng: For me at the startup level, this means if I design a decoder using these bounds, I can tell my customers exactly what the expected confusion rate is before I even start testing on real traffic data.
Lalam: Based on what's in the paper, they prove that as signal power or redundancy increases, the confusion rate drops far below our target epsilon for typical communication systems.
Jane: So in simple terms, this paper gives us mathematical proof that we can design QAM decoders where the undetected error rate is so low it becomes negligible compared to other errors like erasures.
Tom: It confirms that using hybrid automatic repeat request or network coding at high SNR will be enough to handle what's left over after these confusion bounds are met.
Conclusion: Tom: So we've been looking at this paper called "Confusion-Erasure Bounds of Error-Bounded Decoders under QAM." It basically shows how to separate two different types of errors in decoding—undetected confusion and erasures—when you're using QAM constellations in a short blocklength setting.
Jane: That’s the big picture, Tom. So, instead of just saying "we have an error rate," they’re giving us separate numbers for when something gets lost versus when it gets misinterpreted silently. What does that actually mean for someone who isn't deep into the math?
Lu: It means they developed analytical limits—lower and upper bounds—for how often that silent confusion happens, based on how much signal power you use and how long your data packets are. They do this by looking at the average distance between symbols in a square QAM constellation.
Meng: From my side, I'm curious about those bounds themselves. If they’re giving me an upper limit on confusion that drops really low when I increase the signal strength, does that actually help me design a more reliable system faster? Or is it just theoretical math?
Lalam: The paper confirms that when you crank up the signal or add more redundancy, these confusion bounds fall many orders of magnitude below your target error rate, epsilon. What’s left over is mostly detectable erasures.
Tom: That’s the part I find most practical for deployment. If we can prove that the undetected errors are so small—way smaller than what we need to worry about—then we know our main job is just handling the erasures with things like hybrid automatic repeat request or network coding at high signal levels.
Jane: So, essentially, this work gives us a mathematical guarantee that for those QAM systems in the short blocklength regime, you can design your decoder so the silent errors don't hurt your performance much compared to what we already know how to fix. But there’s still a lot of work left on making sure these bounds hold up under every possible real-world condition.
Lu: Exactly, and the paper points toward future work involving even more complex channel models, moving beyond the current analysis into scenarios where things get really messy. We need to see how these bounds behave when the channel itself is changing in unpredictable ways across a large area.
RPTU University Kaiserslautern-Landau · German Research Center for Artificial Intelligence (DFKI)
cs.IT, math.IT
Submitted: 2026-09-11
Updated: 2026-11-03
Comments: Submitted to IEEE ICC 2027
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 92/100
The gist: Block error rate (BLER) conflates block confusions (undetected errors) and block erasures, which have fundamentally different impacts on system reliability.
Key concepts
- Block Confusion
- This is an undetected decoding error where the decoder incorrectly commits to a wrong codeword without signaling an error. It is critical because it can lead to catastrophic failures in safety-critical systems like vehicular control, unlike erasures which are detectable.
- Block Erasure
- This occurs when the receiver detects a failure and can request retransmission of the block. Unlike confusion, erasures are detectable and manageable through techniques like hybrid automatic repeat request (HARQ). The paper focuses on how confusion bounds relate to these exploitable erasure rates.
- QAM Heterogeneity
- Unlike simpler modulation schemes, QAM symbols have multiple possible Euclidean distances between codewords. This makes calculating the confusion rate complex because the distance isn't a single value. The analysis handles this by collapsing this distribution into a root-mean-square (RMS) distance to derive bounds.
- Finite Blocklength (FBL) Regime
- This refers to systems where the data is transmitted in blocks of limited, finite length rather than an infinite stream. This regime is relevant for short packets used in latency-sensitive applications like 6G vehicular communications and requires specialized information-theoretic analysis.
Terminology
Summary
Block error rate (BLER) conflates block confusions (undetected errors) and block erasures, which have fundamentally different impacts on system reliability. p. 0 This paper extends the confusion and erasure analysis of error-bounded decoders to square quadrature amplitude modulation (QAM) constellations in the finite blocklength (FBL) regime. p. 0
Introduction to the Problem
As 6G moves beyond 5G’s reliability targets, supporting latency- and reliability-sensitive services such as vehicular control and industrial automation at short blocklengths, the accurate characterization of physical (PHY)-layer decoding failures becomes essential. Block error rate (BLER) conflates two fundamentally distinct error types: block erasures, in which the receiver detects the failure and can request retransmission, and block confusions (undetected errors), in which the decoder silently commits to an incorrect codeword. p. 0 This distinction is critical in safety-critical applications such as vehicular communications and avionics, where undetected control-command errors can be catastrophic, yet rigorous information-theoretic analysis of block confusion in the FBL regime remains largely absent. p. 0
Handling QAM Heterogeneity
The principal contributions involve handling QAM’s heterogeneous symbol energies—which make the per-pair Euclidean distance a distribution rather than a single value—by deriving analytical lower and upper bounds on the block confusion rate by collapsing this distribution to its root-mean-square (RMS) distance and averaging the pairwise confusion over it. Specifically, unlike the single-hypersphere PSK case of [15], QAM symbols take multiple inter-symbol distances, so the Euclidean distance of a Hamming-distance-d codeword pair is not a single value. p. 0
Derivation of Bounds
The pairwise confusion rate depends only on the Euclidean distance D(x, x′) between codewords. The analysis relies on relating the squared Euclidean distance to the Hamming distance via the per-symbol squared-distance coefficient ∆ in (13). p. 2 This leads to a lower bound on confusion rate Pcon derived by keeping only the shell at the maximum achievable minimum distance dUB H,min—the farthest, least-confusable shell—and dropping the others. p. 2
Sensitivity Analysis of Bounds
The sensitivity analysis shows that the lower bound P LB con is monotonically decreasing and convex in the average symbol energy E¯s. Furthermore, the upper bound P UB con is piecewise continuous in E¯s, with jump discontinuities at specific energy levels i = (√M+1) R 2(ε) 3(√M−1)i. p. 3 The blocklength n also influences the bounds, with dUB H,min being monotonically increasing in n. p. 3
Numerical Confirmation and Conclusion
Numerical results confirm that as the signal-to-noise ratio (SNR) or redundancy increases, the confusion rate bounds fall many orders of magnitude below the target ε. This leaves detectable erasures as the dominant residual error, which is exploitable by hybrid automatic repeat request (HARQ) and network coding at high SNR. p. 4 The confusion bounds translate directly into erasure-rate bounds where Pcon << ε. p. 4 p. 4
References
[1] Y. Polyanskiy, H. p. 0 V. Poor, and S. p. 5 Verdu, “Channel coding rate in the finite blocklength regime,” IEEE Transactions on Information Theory, vol. p. 5 56, no. 5, pp. 2307–2359, 2010
[2] G. Durisi, T. Koch, and P. p. 5 Popovski, “Toward massive, ultrareliable, and low-latency wireless communication with short packets,” Proceedings of the IEEE, vol. p. 5 104, no. 9, pp. 1711–1726, 2016
[3] G. Forney, “Exponential error bounds for erasure, list, and decision feedback schemes,” IEEE Transactions on Information Theory, vol. p. 5 14, no. 2, pp. 206–220, 1968
[4] N. Merhav, “Error exponents of erasure/list decoding revisited via moments of distance enumerators,” IEEE Transactions on Information Theory, vol. p. 5 54, no. 10, pp. 4439–4447, 2008
[5] A. Somekh-Baruch and N. p. 0 Merhav, “Exact random coding exponents for erasure decoding,” IEEE transactions on information theory, vol. p. 5 57, no. 10, pp. 6444–6454, 2011
[6] W. Yang, G. Durisi, T. Koch, and Y. p. 5 Polyanskiy, “Quasi-static multipleantenna fading channels at finite blocklength,” IEEE Transactions on Information Theory, vol. p. 5 60, no. 7, pp. 4232–4265, 2014
[7] A. Collins and Y. p. 5 Polyanskiy, “Coherent multiple-antenna block-fading channels at finite blocklength,” IEEE Transactions on Information Theory, vol. p. 5 65, no. 1, pp. 380–405, 2018
[8] Y. Hu, J. Gross, and A. p. 5 Schmeink, “On the capacity of relaying with finite blocklength,” IEEE Transactions on Vehicular Technology, vol. p. 5 65, no. 3, pp. 1790–1794, 2015
[9] Z. Xiang, W. Yang, Y. Cai, Z. Ding, Y. Song, and Y. p. 5 Zou, “NOMAassisted secure short-packet communications in IoT,” IEEE Wireless Communications, vol. p. 5 27, no. 4, pp. 8–15, 2020
[10] C. Di, D. Proietti, I. p. 5 E. Telatar, T. p. 5 J. Richardson, and R. p. 5 L. Urbanke, “Finite-length analysis of low-density parity-check codes on the binary erasure channel,” IEEE Transactions on Information theory, vol. p. 5 48, no. 6, pp. 1570–1579, 2002
[11] A. F. Dana, R. Gowaikar, R. p. 5 Palanki, B. p. 5 Hassibi, and M. p. 5 Effros, “Capacity of wireless erasure networks,” IEEE Transactions on Information Theory, vol. p. 5 52, no. 3, pp. 789–804
[12] G. Liva, E. Paolini, and M. p. 5 Chiani, “Bounds on the error probability of block codes over the q-ary erasure channel,” IEEE Transactions on Communications, vol. p. 5 61, no. 6, pp. 2156–2165
[13] A. G. i Fabregas and Q. p. 5 Tang, “Coding in the block-erasure channel,” in 2006 Australian Communications Theory Workshop. p. 5 IEEE, 2006, pp. p. 5 19–24
[14] F. Didier, “A new upper bound on the block error probability after decoding over the erasure channel,” IEEE Transactions on Information Theory, vol. p. 5 52, no. 10, pp. 4496–4503
[15] B. Han, Y. Zhu, R. F. Schaefer, G. p. 5 Caire, A. Schmeink, H. p. 5 V. Poor, and H. p. 5 D. Schotten, “Confusions and erasures of error-bounded block decoders with finite blocklength,” arXiv preprint arXiv:2512. p. 516665, 2025
[18] The paper's references are listed in the final section
Conclusion
In this work, by introducing the per-symbol squareddistance coefficient, analytical bounds on the block confusion rate were derived and shown to be monotonically decreasing in both the average symbol energy and the blocklength, at a rate governed by the constellation order. Numerical results confirmed that as the SNR or redundancy increases, the confusion rate falls many orders of magnitude below the reliability target, leaving erasures as the dominant residual failure, which is exploitable by hybrid automatic repeat request (HARQ) and network coding at high SNR. p. 4 The confusion bounds translate directly into erasure-rate bounds where Pcon << ε.
Appendix A
Proof of Theorem 3 shows that the lower bound P LB con is monotonically decreasing and convex in E¯s. The second derivative analysis proves convexity based on Lemma 2. p. 5
Appendix B
Proof of Lemma 5 uses the Plotkin bound to show dUB H,min(n) < (M−1)(n+1)M.
Appendix C
Proof of Theorem 5 demonstrates that the upper bound P UB con is piecewise continuous in E¯s and convex within each continuity interval. The analysis shows P UB con is monotonically decreasing in E¯s within each interval of constant dLB H,min. p. 5
Appendix D
The proof of Theorem 6 establishes that the upper bound P UB con is monotonically decreasing in n for sufficiently large n. The competition between increasing decision radius and suppressed shell population leads to this global decrease. p. 3
Appendix E
The appendix contains proofs for Theorem 3, Lemma 5, and Theorem 5.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
High reliability communication systems can be designed to tolerate undetected errors by ensuring that residual failures are predominantly detectable erasures, as
undetected confusion becomes a vanishing fraction of ε
when SNR or redundancy increases. -
AI-driven communication protocols can exploit the derived bounds on block confusion rate Pcon to optimize decoding thresholds, allowing the system to reliably distinguish between silent codeword errors and detectable channel failures like those flagged by parity-check syndromes.
-
Machine learning models for error correction can be trained using the derived piecewise continuous upper bound PUB con in terms of symbol energy and SNR to determine the optimal constellation order and transmit power required to meet stringent reliability targets.
Sources
Related papers
- The Gaussian Multiuser Channel Capacity Region
- The mathematical landscape of partial information decomposition: A comprehensive review of properties and measures
- Electromagnetic Signal and Information Theory: A Continuous-Aperture Array Perspective
- Analytically Characterized Optimal Power Control for Signal-Level-Integrated Sensing, Computing and Communication in Federated Learning
- Information Spectrum Methods for Multiple-Access Channels--Part I: epsilon-Capacity Regions in the Mixed Gaussian Case
- Seed Recycling: Explicit Codes for Mixed Synchronization Errors Matching the Leading Term of the Gilbert--Varshamov Bound