pith. sign in

arxiv: 2510.16620 · v4 · submitted 2025-10-18 · 💻 cs.IT · cs.AI· cs.CR· cs.LG· eess.SP· math.IT

Feedback Lunch: Learned Feedback Codes for Secure Communications

Yingyao Zhou , Natasha Devroye , Onur G\"unl\"u This is my paper

Pith reviewed 2026-05-18 05:40 UTC · model grok-4.3

classification 💻 cs.IT cs.AIcs.CRcs.LGeess.SPmath.IT
keywords secure communicationschannel feedbackwiretap channellearned codessecret key agreementuniversal hash functionsblock-fading Gaussian channelISAC
0
0 comments X

The pith

Feedback from the receiver lets legitimate parties agree on a shared secret key in channels where secrecy capacity would otherwise be zero.

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

The paper examines reversely-degraded secure-communication channels, such as the block-fading Gaussian wiretap channel, where the eavesdropper holds a permanent advantage and secrecy capacity is zero without feedback. It introduces a seeded modular code design that pairs learned feedback-based codes to ensure reliable communication over the legitimate link with universal hash functions to extract a secret key while controlling leakage. The approach demonstrates that channel-output feedback allows the sender and receiver to overcome the eavesdropper's advantage and establish a shared key. Trade-offs between reliability and leakage are analyzed through this construction. The work points toward code designs that incorporate sensing for secure integrated sensing and communication systems.

Core claim

For reversely-degraded secure-communication channels where secrecy capacity is zero without feedback, a seeded modular code design for the block-fading Gaussian wiretap channel with channel-output feedback combines universal hash functions for security and learned feedback-based codes for reliability. This construction enables the legitimate parties to agree on a secret key, overcoming the security advantage of the eavesdropper.

What carries the argument

Seeded modular code design that integrates learned feedback-based codes for reliability on the main channel with universal hash functions to extract a controllable-leakage secret key.

If this is right

  • Feedback turns zero-secrecy-capacity channels into ones where a shared key can be established.
  • Reliability and leakage can be traded off explicitly through the modular design.
  • The same structure applies to block-fading Gaussian wiretap settings with output feedback.
  • The design motivates further codes for sensing-assisted security in integrated sensing and communication.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar feedback-assisted codes could extend to other wireless scenarios where the eavesdropper channel is stronger.
  • Combining the codes with real-time sensing data might further reduce leakage in dynamic environments.
  • The learned component may allow adaptation to channel variations that fixed codes cannot handle.

Load-bearing premise

The learned feedback-based codes can achieve enough reliability on the legitimate link at the same time that universal hash functions keep the extracted key's leakage controllable.

What would settle it

A simulation or experiment on the block-fading Gaussian wiretap channel that shows information leakage exceeding the controllable threshold after applying the learned feedback codes and hash functions.

Figures

Figures reproduced from arXiv: 2510.16620 by Natasha Devroye, Onur G\"unl\"u, Yingyao Zhou.

Figure 1
Figure 1. Figure 1: Design of modular deep-learned feedback wiretap codes. The security [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Detailed structure of the reliability layer. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: BLER (left) and estimated Iˆ bob, Lˆeve (right) versus blocklength under SNR = SY,f = SZ,f with noiseless feedback. s = [1, 0, 1], [1, 1, 0, 1], and [1, 1, 0, 1, 0] are fixed for q = 3, 4, 5 and shared between the legitimate parties. As shown in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: BLER versus forward SNR at Bob (left) and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Trade-off between the BLER and information leakage. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
read the original abstract

We consider reversely-degraded secure-communication channels, for which the secrecy capacity is zero if there is no channel feedback. Specifically, we focus on a seeded modular code design for the block-fading Gaussian wiretap channel with channel-output feedback, combining universal hash functions for security and learned feedback-based codes for reliability. The trade-off between communication reliability and information leakage is studied, illustrating that feedback enables agreeing on a secret key shared between legitimate parties, overcoming the security advantage of the eavesdropper. Our findings motivate code designs for sensing-assisted secure communications in the context of integrated sensing and communication (ISAC).

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper considers reversely-degraded secure-communication channels (secrecy capacity zero without feedback) and proposes a seeded modular code design for the block-fading Gaussian wiretap channel with channel-output feedback. It combines universal hash functions for security with learned feedback-based codes for reliability on the legitimate link, studies the resulting reliability-leakage trade-off, and claims that feedback enables secret-key agreement that overcomes the eavesdropper's advantage. The work motivates extensions to sensing-assisted secure communications in ISAC settings.

Significance. If the construction is shown to preserve the min-entropy and independence properties required for the hashing step, the result would demonstrate a concrete, practical route to positive secrecy rates in a class of channels where information-theoretic secrecy capacity is otherwise zero. The modular separation of learned reliability codes from information-theoretic security primitives is a clear strength, as is the explicit motivation toward ISAC applications. The approach supplies a falsifiable design template that can be tested numerically and potentially extended to other feedback-assisted wiretap models.

major comments (2)
  1. [§4] §4 (seeded modular construction): the central claim that feedback overcomes the eavesdropper advantage rests on the assertion that the learned feedback codes simultaneously deliver reliable decoding while producing outputs whose min-entropy and statistical independence from the wiretap observation remain sufficient for the universal hash to drive leakage to zero with block length. No analysis, bound, or diagnostic is supplied showing that the neural-network training does not introduce deterministic structure or correlations visible to the stronger eavesdropper link; if such structure exists, the hashing lemma no longer guarantees vanishing leakage and the security claim fails.
  2. [Numerical results] Numerical results section: the reported reliability-leakage curves are presented without error bars, multiple random seeds, or ablation on the learned-code training objective, making it impossible to assess whether the observed leakage remains controllable across realizations or whether the NN occasionally produces low-entropy outputs that would violate the security guarantee.
minor comments (2)
  1. [System Model] The system-model notation for the block-fading Gaussian wiretap channel with feedback is introduced without an explicit equation reference for the reverse-degradation condition; adding a numbered equation would improve clarity.
  2. [Introduction] A short discussion of related work on feedback-assisted wiretap coding (e.g., prior information-theoretic results on key agreement with feedback) is missing from the introduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We respond to each major comment below and outline the revisions we will make to address the concerns raised.

read point-by-point responses

  1. Referee: [§4] §4 (seeded modular construction): the central claim that feedback overcomes the eavesdropper advantage rests on the assertion that the learned feedback codes simultaneously deliver reliable decoding while producing outputs whose min-entropy and statistical independence from the wiretap observation remain sufficient for the universal hash to drive leakage to zero with block length. No analysis, bound, or diagnostic is supplied showing that the neural-network training does not introduce deterministic structure or correlations visible to the stronger eavesdropper link; if such structure exists, the hashing lemma no longer guarantees vanishing leakage and the security claim fails.

    Authors: We appreciate the referee pointing out this critical requirement for the security argument. In our modular construction, the learned feedback code is trained exclusively to ensure reliable decoding at the legitimate receiver using the channel output feedback, while the universal hash function is applied post hoc to achieve information-theoretic security. Because the training does not involve the eavesdropper's channel and the feedback is from the main link, we believe the outputs preserve the necessary randomness properties. Nevertheless, to rigorously support this, we will revise the manuscript to include additional analysis and numerical diagnostics that estimate the min-entropy of the learned code outputs and their statistical independence from the wiretap observations. This will help validate that the hashing step can indeed drive the leakage to zero. revision: yes

  2. Referee: [Numerical results] Numerical results section: the reported reliability-leakage curves are presented without error bars, multiple random seeds, or ablation on the learned-code training objective, making it impossible to assess whether the observed leakage remains controllable across realizations or whether the NN occasionally produces low-entropy outputs that would violate the security guarantee.

    Authors: We concur that enhancing the numerical results with more statistical detail would strengthen the paper. In the revised version, we will add error bars derived from multiple independent runs with different random seeds. Additionally, we will include ablation studies on the training objective to demonstrate the consistency of the reliability-leakage trade-off and to check for any instances of low-entropy outputs that could compromise security. revision: yes

Circularity Check

0 steps flagged

No circularity detected; derivation self-contained

full rationale

The provided abstract and excerpts describe a seeded modular design that combines universal hash functions for security with learned feedback-based codes for reliability on the block-fading Gaussian wiretap channel. The central illustration—that feedback enables secret-key agreement overcoming the eavesdropper advantage—is presented as the outcome of studying the reliability-leakage trade-off, without any equations, fitted parameters, or self-citations that reduce a claimed prediction or uniqueness result to an input by construction. No self-definitional loops, fitted-input predictions, or load-bearing self-citations appear in the visible text, so the derivation chain does not collapse to its own assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the design implicitly assumes that learned codes can be trained to meet both reliability and secrecy targets simultaneously.

pith-pipeline@v0.9.0 · 5641 in / 1064 out tokens · 26724 ms · 2026-05-18T05:40:28.800387+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

  1. [1]

    Bloch and J

    M. Bloch and J. Barros,Physical-Layer Security: From Information Theory to Security Engineering. Cambridge University Press, 2011. 0.5 1.0 1.5 2.0 2.5 Leve (bits) 10 8 10 7 10 6 10 5 10 4 10 3 10 2 10 1 100 BLER SY, f = 1 dB, noiseless FB, q = 4, n = 9 SY, f = 4 dB, noiseless FB, q = 3, n = 6 SY, f = 1 dB, SY, fb = 5 dB, q = 4, n = 9 Fig. 5. Trade-off bet...

    work page 2011

  2. [2]

    An overview of information-theoretic security and privacy: Metrics, limits and applications,

    M. Blochet al., “An overview of information-theoretic security and privacy: Metrics, limits and applications,”IEEE J. Sel. Areas Inf. Theory (JSAC), vol. 2, no. 1, pp. 5–22, 2021

    work page 2021

  3. [3]

    The wire-tap channel,

    A. D. Wyner, “The wire-tap channel,”Bell Sys. Techn. J., vol. 54, no. 8, pp. 1355–1387, 1975

    work page 1975

  4. [4]

    Transmission, identification and common randomness capacities for wire-tape channels with secure feedback from the decoder,

    R. Ahlswede and N. Cai, “Transmission, identification and common randomness capacities for wire-tape channels with secure feedback from the decoder,” inGeneral Theory of Information Transfer and Combinatorics. Springer, 2006, pp. 258–275

    work page 2006

  5. [5]

    Secure integrated sensing and communication,

    O. G ¨unl¨u, M. Bloch, R. F. Schaefer, and A. Yener, “Secure integrated sensing and communication,”IEEE J. Selected in Inf. Theory (JSAIT), vol. 4, pp. 40–53, 2023

    work page 2023

  6. [6]

    Semantic security for the wiretap channel,

    M. Bellare, S. Tessaro, and A. Vardy, “Semantic security for the wiretap channel,” inAdv. in Crypt. (CRYPTO), R. Safavi-Naini and R. Canetti, Eds. Springer Berlin Heidelberg, 2012, pp. 294–311

    work page 2012

  7. [7]

    Short blocklength wiretap channel codes via deep learning: Design and performance evaluation,

    V . Rana and R. A. Chou, “Short blocklength wiretap channel codes via deep learning: Design and performance evaluation,”IEEE Trans. on Commun. (TCOM), vol. 71, no. 3, pp. 1462–1474, 2023

    work page 2023

  8. [8]

    Deep learning-based codes for wiretap fading channels,

    D. Seifert, O. G ¨unl¨u, and R. F. Schaefer, “Deep learning-based codes for wiretap fading channels,” inIEEE Int. Symp. on Personal, Indoor Mobile Radio Commun. (PIMRC), 2025

    work page 2025

  9. [9]

    Joint communications & sensing – Common radio- communications and sensor technology,

    G. Fettweiset al., “Joint communications & sensing – Common radio- communications and sensor technology,”VDE Positionspapier, 2021

    work page 2021

  10. [10]

    LIGHTCODE: Light analytical and neural codes for channels with feedback,

    S. K. Ankireddy, K. Narayanan, and H. Kim, “LIGHTCODE: Light analytical and neural codes for channels with feedback,”IEEE J. Sel. Areas Commun. (JSAC), pp. 1–1, 2025

    work page 2025

  11. [11]

    No free lunch theorem for security and utility in federated learning,

    X. Zhang, H. Gu, L. Fan, K. Chen, and Q. Yang, “No free lunch theorem for security and utility in federated learning,”ACM Trans. on Intell. Syst. and Techno. (TIST), vol. 14, no. 1, pp. 1–35, 2022

    work page 2022

  12. [12]

    S. K. Leung-Yan-Cheong,Multi-User and Wiretape Channels Including Feedback. Stanford University, 1976

    work page 1976

  13. [13]

    A coding scheme for additive noise channels with feedback–i: No bandwidth constraint,

    J. Schalkwijk and T. Kailath, “A coding scheme for additive noise channels with feedback–i: No bandwidth constraint,”IEEE Trans. on Inf. Theory (T-IT), vol. 12, no. 2, pp. 172–182, 1966

    work page 1966

  14. [14]

    Secret communication with feedback,

    D. Gunduz, D. R. Brown, and H. V . Poor, “Secret communication with feedback,” inInt. Symp. Inf. Theory Appl. (ISITA), 2008, pp. 1–6

    work page 2008

  15. [15]

    Universal classes of hash functions,

    J. L. Carter and M. N. Wegman, “Universal classes of hash functions,” inACM Symp. on Theory of Comput. (STOC), 1977, pp. 106–112

    work page 1977

  16. [16]

    Learned codes for broadcast channels with feedback,

    Y . Zhou and N. Devroye, “Learned codes for broadcast channels with feedback,” inICC 2025-IEEE International Conference on Communi- cations. IEEE, 2025, pp. 3803–3808

    work page 2025

  17. [17]

    Mutual information neural estimation,

    M. Belghaziet al., “Mutual information neural estimation,” inPMLR Int. Conf. on Machine Learning (ICML), 2018, pp. 531–540

    work page 2018

  18. [18]

    f-DIME: Mutual information estimation viaf-divergence and data derangements,

    N. A. Letizia, N. Novello, and A. M. Tonello, “f-DIME: Mutual information estimation viaf-divergence and data derangements,”Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 37, pp. 105 114–105 150, 2024

    work page 2024

  19. [19]

    Understanding the limitations of variational mutual information estimators,

    J. Song and S. Ermon, “Understanding the limitations of variational mutual information estimators,” inInternational Conference on Learning Representations, 2020

    work page 2020

  20. [20]

    CLUB: A contrastive log-ratio upper bound of mutual information,

    P. Cheng, W. Hao, S. Dai, J. Liu, Z. Gan, and L. Carin, “CLUB: A contrastive log-ratio upper bound of mutual information,” inInt. Conf. Mach. Learn. (ICML), 2020, pp. 1779–1788

    work page 2020

  21. [21]

    The Gaussian channel with noisy feedback,

    Y .-H. Kim, A. Lapidoth, and T. Weissman, “The Gaussian channel with noisy feedback,” inIEEE Int. Symp. Inf. Theory (ISIT), 2007, pp. 1416– 1420

    work page 2007

  22. [22]

    Concatenated classic and neural (CCN) codes: ConcatenatedAE,

    O. G ¨unl¨u, R. Fritschek, and R. F. Schaefer, “Concatenated classic and neural (CCN) codes: ConcatenatedAE,” inIEEE Wireless Commun. Netw. Conf. (WCNC), 2023, pp. 1–6

    work page 2023