pith. sign in

arxiv: 2604.04413 · v1 · submitted 2026-04-06 · 📡 eess.SP

A Survey on Robust Deep Joint Source-Channel Coding for Semantic Communications

Eunhye Hong , Taewoo Park , Yongjune Kim This is my paper

Pith reviewed 2026-05-10 20:20 UTC · model grok-4.3

classification 📡 eess.SP
keywords semantic communicationsdeep joint source-channel codingrobustnesschannel mismatchadaptive methodsjoint source-channel codingdeep learning
0
0 comments X

The pith

Deep JSCC for semantic communications achieves robustness either through training that anticipates channel changes or through adaptive adjustments at deployment.

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

This survey reviews techniques to make deep joint source-channel coding robust for semantic communications when real channels differ from those used in training. It divides the methods into robust training approaches, which prepare the model in advance, and adaptive approaches that adjust at runtime. The adaptive class is split into selecting semantic features, adapting at the physical layer, and adapting semantic features themselves. A reader would care because such robustness is essential for practical deployment of efficient, meaning-focused communication systems in dynamic environments. The paper concludes by outlining promising research paths including generalization across multiple tasks and improving explainability.

Core claim

The paper establishes that robustness in deep JSCC for semantic communications is achieved either by training models to be inherently robust to channel variations or by enabling adaptation during deployment, and it structures the recent literature around these two classes to provide an overview and identify future directions such as multi-task generalization.

What carries the argument

The two-class categorization of robustness approaches into robust training methods and adaptive methods, with the latter subdivided into adaptive semantic feature selection, physical-layer adaptation, and semantic feature adaptation. This taxonomy organizes the surveyed literature to clarify how systems maintain performance under mismatched channel conditions.

If this is right

  • Robust training approaches can improve performance without requiring runtime changes to the system.
  • Adaptive approaches enable dynamic responses to varying channel conditions during operation.
  • Subdivisions such as semantic feature selection and physical-layer adaptation offer targeted strategies for handling mismatches.
  • Future systems could extend these ideas to multi-task scenarios and incorporate explainability mechanisms.

Where Pith is reading between the lines

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

  • The taxonomy could support hybrid designs that combine pre-training robustness with selective runtime adaptation.
  • Testing the categorized methods on measured wireless channels rather than simulated ones would reveal practical performance gaps.
  • Improved explainability might allow operators to diagnose why adaptation succeeds or fails in specific environments.

Load-bearing premise

That the selected literature accurately represents the full range of robustness methods and that the proposed categorization is both complete and useful for guiding future work.

What would settle it

A new robustness technique for deep JSCC that cannot be placed into either the robust training class or any of the three adaptive subclasses would require the taxonomy to be revised.

Figures

Figures reproduced from arXiv: 2604.04413 by Eunhye Hong, Taewoo Park, Yongjune Kim.

Figure 1
Figure 1. Figure 1: System model of a deep JSCC framework for semantic com [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
read the original abstract

Semantic communications (SCs) aim to transmit only the essential information required to perform given tasks, thereby improving communication efficiency. Deep learning-based joint source-channel coding (deep JSCC) has emerged as a promising approach for SC systems; however, its performance often degrades when the deployment channels differ from the training channel conditions, making robustness a critical requirement. This paper presents a structured overview of recent methodologies for enhancing the robustness of deep JSCC. Specifically, existing approaches are categorized into two classes: robust training approaches and adaptive approaches, with the latter further divided into adaptive semantic feature selection, physical-layer adaptation, and semantic feature adaptation. Finally, we discuss promising directions, including multi-task generalization and explainability in robust SC systems.

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 / 1 minor

Summary. The manuscript surveys methods for improving robustness in deep joint source-channel coding (deep JSCC) for semantic communications. It divides existing approaches into two classes—robust training approaches and adaptive approaches—with the adaptive class further split into adaptive semantic feature selection, physical-layer adaptation, and semantic feature adaptation. The paper concludes by outlining future directions such as multi-task generalization and explainability.

Significance. If the taxonomy is shown to be exhaustive and the reviewed literature representative, the survey would provide a useful organizing framework for the field, helping researchers compare training-based versus adaptive robustness strategies and highlighting open problems in multi-task and explainable robust semantic systems.

major comments (2)
  1. [Abstract and categorization section] Abstract and main categorization section: the manuscript states that existing approaches 'are categorized into two classes: robust training approaches and adaptive approaches' but supplies no search protocol, keyword list, database, date range, inclusion/exclusion rules, or explicit assignment criteria for placing a given method into one bin versus another. Without these, the claim that the taxonomy is complete and mutually exclusive cannot be verified and remains load-bearing for the paper's central contribution.
  2. [Adaptive approaches subsection] Adaptive approaches subsection: potential overlaps are not discussed (e.g., a training procedure that jointly optimizes for multiple channel conditions could be viewed as both a robust training method and a form of semantic feature adaptation). The absence of boundary definitions or worked examples of assignment leaves the three-way split inside the adaptive class open to ambiguity.
minor comments (1)
  1. A summary table listing representative papers under each category (with brief method descriptions and channel conditions tested) would improve readability and allow readers to quickly assess coverage.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our survey. We agree that greater transparency on categorization criteria and boundary definitions will strengthen the manuscript. We address each major comment below and will incorporate revisions accordingly.

read point-by-point responses

  1. Referee: [Abstract and categorization section] Abstract and main categorization section: the manuscript states that existing approaches 'are categorized into two classes: robust training approaches and adaptive approaches' but supplies no search protocol, keyword list, database, date range, inclusion/exclusion rules, or explicit assignment criteria for placing a given method into one bin versus another. Without these, the claim that the taxonomy is complete and mutually exclusive cannot be verified and remains load-bearing for the paper's central contribution.

    Authors: We acknowledge that the survey presents a structured overview of representative methods rather than a formal systematic review with PRISMA-style protocols. The two-class taxonomy (robust training vs. adaptive) is based on whether robustness is achieved primarily through offline training or through runtime adaptation mechanisms, as derived from our analysis of the literature. To address the concern, we will add a dedicated subsection (e.g., in Section II or the introduction) that explicitly describes the literature selection process, including search keywords (such as 'robust deep JSCC', 'semantic communications robustness', 'channel-adaptive JSCC'), databases consulted (arXiv, IEEE Xplore, ACM Digital Library), approximate date range (primarily 2018–2024), and high-level inclusion criteria focused on deep learning-based JSCC for semantic tasks. We will also state that the taxonomy organizes key approaches for comparison and does not claim exhaustive completeness or strict mutual exclusivity, thereby clarifying its scope. revision: yes

  2. Referee: [Adaptive approaches subsection] Adaptive approaches subsection: potential overlaps are not discussed (e.g., a training procedure that jointly optimizes for multiple channel conditions could be viewed as both a robust training method and a form of semantic feature adaptation). The absence of boundary definitions or worked examples of assignment leaves the three-way split inside the adaptive class open to ambiguity.

    Authors: We agree that some methods may exhibit characteristics of multiple categories, particularly at the boundary between robust training and adaptive approaches. We will revise the adaptive approaches subsection (and the overall categorization section) to include an explicit discussion of potential overlaps, along with boundary definitions and worked examples. For instance, we will define robust training as methods that optimize a single model offline for robustness (e.g., via multi-condition training or adversarial training) without runtime changes, while adaptive approaches involve dynamic adjustments at inference time. A multi-condition training method would remain under robust training unless it includes online feature or parameter adaptation; we will provide concrete paper examples to illustrate assignments to the three adaptive subcategories (semantic feature selection, physical-layer adaptation, semantic feature adaptation) and note any hybrid cases. revision: yes

Circularity Check

0 steps flagged

No circularity: pure literature survey with no derivations or self-referential results

full rationale

The manuscript is a survey paper that organizes existing external literature into categories (robust training vs. adaptive approaches with sub-divisions). No equations, predictions, fitted parameters, first-principles derivations, or quantitative claims are made. The taxonomy is presented as an organizational summary of reviewed works rather than a result derived from the paper's own inputs. No self-citation chains, ansatzes, or renamings of known results appear as load-bearing steps. The paper is self-contained as a review and does not reduce any claim to its own definitions or fits.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a survey paper; it does not introduce new parameters, axioms, or entities but summarizes existing literature on robust deep JSCC for semantic communications.

pith-pipeline@v0.9.0 · 5418 in / 950 out tokens · 42420 ms · 2026-05-10T20:20:05.937357+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Lightweight Low-SNR-Robust Semantic Communication System for Autonomous Driving

    eess.SY 2026-04 unverdicted novelty 5.0

    The proposed lightweight JSCC semantic communication system with structured pruning and quantization maintains image reconstruction quality and robustness under low SNR while being compatible with digital systems.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    A vision of 6G wireless sy stems: Applications, trends, technologies, and open research pro blems,

    W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless sy stems: Applications, trends, technologies, and open research pro blems,” IEEE Netw., vol. 34, no. 3, pp. 134–142, Oct. 2020

    work page 2020

  2. [2]

    6G R&D vision: Requirements and candidate technologies,

    E.-K. Hong, I. Lee, B. Shim, Y .-C. Ko, S.-H. Kim, S. Pack, K . Lee, S. Kim, J.-H. Kim, Y . Shin, Y . Kim, and H. Jung, “6G R&D vision: Requirements and candidate technologies,” J. Commun. Netw. , vol. 24, no. 2, pp. 232–245, Apr. 2022

    work page 2022

  3. [3]

    Beyond transmitting bits: Conte xt, seman- tics, and task-oriented communications,

    D. G¨ und¨ uz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Y ang, A. Y ener, K. K. Wong, and C.-B. Chae, “Beyond transmitting bits: Conte xt, seman- tics, and task-oriented communications,” IEEE J. Sel. Areas Commun. , vol. 41, no. 1, pp. 5–41, Jan. 2023

    work page 2023

  4. [4]

    Task- oriented communications for 6G: Vision, principles, and te chnologies,

    Y . Shi, Y . Zhou, D. Wen, Y . Wu, C. Jiang, and K. B. Letaief, “ Task- oriented communications for 6G: Vision, principles, and te chnologies,” IEEE Wireless. Commun. , vol. 30, no. 3, pp. 78–85, Jun. 2023

    work page 2023

  5. [5]

    Next generation advanced transceive r technolo- gies for 6G and beyond,

    C. Y ou, Y . Cai, Y . Liu, M. Di Renzo, T. M. Duman, A. Y ener, an d A. Lee Swindlehurst, “Next generation advanced transceive r technolo- gies for 6G and beyond,” IEEE J. Sel. Areas Commun. , vol. 43, no. 3, pp. 582–627, Mar. 2025

    work page 2025

  6. [6]

    Deep joi nt source- channel coding for wireless image transmission,

    E. Bourtsoulatze, D. B. Kurka, and D. G¨ und¨ uz, “Deep joi nt source- channel coding for wireless image transmission,” IEEE Trans. Cogn. Commun. Netw., vol. 5, no. 3, pp. 567–579, Sep. 2019

    work page 2019

  7. [7]

    Neural joint source-channel coding,

    K. Choi, K. Tatwawadi, A. Grover, T. Weissman, and S. Ermo n, “Neural joint source-channel coding,” in Proc. Int. Conf. Mach. Learn. (ICML) , Jun. 2019, pp. 1182–1192

    work page 2019

  8. [8]

    Deep joint source-channel c oding of images with feedback,

    D. B. Kurka and D. G¨ und¨ uz, “Deep joint source-channel c oding of images with feedback,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), May 2020, pp. 5235–5239

    work page 2020

  9. [9]

    Learning task-oriented co mmunication for edge inference: An information bottleneck approach,

    J. Shao, Y . Mao, and J. Zhang, “Learning task-oriented co mmunication for edge inference: An information bottleneck approach,” IEEE J. Sel. Areas Commun. , vol. 40, no. 1, pp. 197–211, Jan. 2022

    work page 2022

  10. [10]

    Robust i nformation bottleneck for task-oriented communication with digital m odulation,

    S. Xie, S. Ma, M. Ding, Y . Shi, M. Tang, and Y . Wu, “Robust i nformation bottleneck for task-oriented communication with digital m odulation,” IEEE J. Sel. Areas Commun. , vol. 41, no. 8, pp. 2577–2591, Aug. 2023

    work page 2023

  11. [11]

    Robust semantic communications with masked VQ-V AE enabled codebook,

    Q. Hu, G. Zhang, Z. Qin, Y . Cai, G. Y u, and G. Y . Li, “Robust semantic communications with masked VQ-V AE enabled codebook,” IEEE Trans. Wireless Commun., vol. 22, no. 12, pp. 8707–8722, Dec. 2023

    work page 2023

  12. [12]

    Distrib uted boosting classification over noisy communication channels,

    Y . Kim, J. Shin, Y . Cassuto, and L. R. V arshney, “Distrib uted boosting classification over noisy communication channels,” IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 141–154, Jan. 2023

    work page 2023

  13. [13]

    Joint tas k and data-oriented semantic communications: A deep separate so urce-channel coding scheme,

    J. Huang, D. Li, C. Huang, X. Qin, and W. Zhang, “Joint tas k and data-oriented semantic communications: A deep separate so urce-channel coding scheme,” IEEE Internet Things J. , vol. 11, no. 2, pp. 2255–2272, Jan. 2024

    work page 2024

  14. [14]

    Semantic communi cations for image recovery and classification via deep joint source a nd channel coding,

    Z. Lyu, G. Zhu, J. Xu, B. Ai, and S. Cui, “Semantic communi cations for image recovery and classification via deep joint source a nd channel coding,” IEEE Trans. Wireless Commun. , vol. 23, no. 8, pp. 8388–8404, Aug. 2024

    work page 2024

  15. [15]

    Deep learning en abled semantic communication systems,

    H. Xie, Z. Qin, G. Y . Li, and B. H. Juang, “Deep learning en abled semantic communication systems,” IEEE Trans. Signal Process., vol. 69, pp. 2663–2675, May 2021

    work page 2021

  16. [16]

    W ireless image transmission using deep source channel coding with at tention modules,

    J. Xu, B. Ai, W. Chen, A. Y ang, P . Sun, and M. Rodrigues, “W ireless image transmission using deep source channel coding with at tention modules,” IEEE Trans. Circuits Syst. Video Technol. , vol. 32, no. 4, pp. 2315–2328, Apr. 2022

    work page 2022

  17. [17]

    Deep learning for j oint source- channel coding of text,

    N. Farsad, M. Rao, and A. Goldsmith, “Deep learning for j oint source- channel coding of text,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), Apr. 2018, pp. 2326–2330

    work page 2018

  18. [18]

    Semantic communication systems for speech transmission,

    Z. Weng and Z. Qin, “Semantic communication systems for speech transmission,” IEEE J. Sel. Areas Commun. , vol. 39, no. 8, pp. 2434– 2444, Aug. 2021

    work page 2021

  19. [19]

    A unified multi- task semantic communication system for multimodal data,

    G. Zhang, Q. Hu, Z. Qin, Y . Cai, G. Y u, and X. Tao, “A unified multi- task semantic communication system for multimodal data,” IEEE Trans. Commun., vol. 72, no. 7, pp. 4101–4116, Jul. 2024

    work page 2024

  20. [20]

    Rate-adaptive coding mechanism for semantic communications with multi-modal data,

    Y . He, G. Y u, and Y . Cai, “Rate-adaptive coding mechanism for semantic communications with multi-modal data,” IEEE Trans. Commun., vol. 72, no. 3, pp. 1385–1400, Mar. 2024

    work page 2024

  21. [21]

    Regular ized deep joint source-channel coding for robust task-oriented sema ntic commu- nications,

    T. Park, E. Hong, Y .-S. Jeon, N. Lee, and Y . Kim, “Regular ized deep joint source-channel coding for robust task-oriented sema ntic commu- nications,” in Proc. IEEE Global Commun. Conf. (GLOBECOM) , Dec. 2025, pp. 4675–4680

    work page 2025

  22. [22]

    Robust deep joint source channel coding for task-o riented se- mantic communications,

    ——, “Robust deep joint source channel coding for task-o riented se- mantic communications,” arXiv preprint arXiv:2503.12907 , 2025

    work page arXiv 2025

  23. [23]

    Robust semantic communication via adversarial training,

    K. Wei, R. Xie, W. Xu, Z. Lu, and H. Xiao, “Robust semantic communication via adversarial training,” IEEE Trans. V eh. Technol. , vol. 74, no. 12, pp. 19 849–19 853, Dec. 2025

    work page 2025

  24. [24]

    Deep joint source-channel codin g for wireless image transmission with adaptive rate control,

    M. Y ang and H.-S. Kim, “Deep joint source-channel codin g for wireless image transmission with adaptive rate control,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP) , May 2022, pp. 5193–5197

    work page 2022

  25. [25]

    Predictive and adaptive deep coding for wireless image tra nsmission in semantic communication,

    W. Zhang, H. Zhang, H. Ma, H. Shao, N. Wang, and V . C. M. Leu ng, “Predictive and adaptive deep coding for wireless image tra nsmission in semantic communication,” IEEE Trans. Wireless Commun. , vol. 22, no. 8, pp. 5486–5501, Aug. 2023

    work page 2023

  26. [26]

    Deep joint source-channel coding for adaptive image transmissi on over MIMO channels,

    H. Wu, Y . Shao, C. Bian, K. Mikolajczyk, and D. G¨ und¨ uz, “Deep joint source-channel coding for adaptive image transmissi on over MIMO channels,” IEEE Trans. Wireless Commun. , vol. 23, no. 10, pp. 15 002– 15 017, 2024

    work page 2024

  27. [27]

    Rate adaptive mechanism for se mantic communication systems: A robustness verification approach ,

    H. Gao, G. Y u, and Y . Cai, “Rate adaptive mechanism for se mantic communication systems: A robustness verification approach ,” IEEE Netw., vol. 38, no. 4, pp. 216–223, Jul. 2024

    work page 2024

  28. [28]

    Joint source-cha nnel coding for channel-adaptive digital semantic communications,

    J. Park, Y . Oh, S. Kim, and Y .-S. Jeon, “Joint source-cha nnel coding for channel-adaptive digital semantic communications,” IEEE Trans. Cogn. Commun. Netw., vol. 11, no. 1, pp. 75–89, Feb. 2025

    work page 2025

  29. [29]

    SNR-adaptive deep join t source- channel coding for wireless image transmission,

    M. Ding, J. Li, M. Ma, and X. Fan, “SNR-adaptive deep join t source- channel coding for wireless image transmission,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP) , Jun. 2021, pp. 1555– 1559

    work page 2021

  30. [30]

    Deep learning-based adaptive joint source-chan nel coding using hypernetworks,

    S. Xie, H. He, H. Li, S. Song, J. Zhang, Y .-J. A. Zhang, and K. B. Letaief, “Deep learning-based adaptive joint source-chan nel coding using hypernetworks,” in Proc. IEEE Int. Mediterranean Conf. Commun. Netw. (MeditCom), Jul. 2024, pp. 191–196

    work page 2024

  31. [31]

    SNR-EQ-JSCC: Joint source-channe l coding with SNR-based embedding and query,

    H. Zhang and M. Tao, “SNR-EQ-JSCC: Joint source-channe l coding with SNR-based embedding and query,” IEEE Commun. Lett. , vol. 14, no. 3, pp. 881–885, 2025

    work page 2025

  32. [32]

    Transmit what you need: Task-ada ptive semantic communications for visual information,

    J. Park and S. W. Y oon, “Transmit what you need: Task-ada ptive semantic communications for visual information,” IEEE J. Sel. Areas Commun., vol. 43, no. 12, pp. 4182–4197, Oct. 2025

    work page 2025

  33. [33]

    Task-oriented explainable semantic commun ications,

    S. Ma, W. Qiao, Y . Wu, H. Li, G. Shi, D. Gao, Y . Shi, S. Li, an d N. Al-Dhahir, “Task-oriented explainable semantic commun ications,” IEEE Trans. Wireless Commun. , vol. 22, no. 12, pp. 9248–9262, Dec. 2023

    work page 2023

  34. [34]

    Attent ion-aware semantic communications for collaborative inference,

    J. Im, N. Kwon, T. Park, J. Woo, J. Lee, and Y . Kim, “Attent ion-aware semantic communications for collaborative inference,” IEEE Internet Things J. , vol. 11, no. 22, pp. 37 008–37 020, Nov. 2024

    work page 2024