Semantic HARQ for Intelligent Transportation Systems: Joint Source-Channel Coding-Powered Reliable Retransmissions

Xu Wang; Yaru Fu; Yongkang Li; Zheng Shi

REVIEW 3 major objections 5 minor 31 references

Semantic HARQ for Intelligent Transportation Systems: Joint Source-Channel Coding-Powered Reliable Retransmissions

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Integrating HARQ with JSCC makes semantic text transmission over fading channels noticeably more reliable, with the largest gains at low signal-to-noise ratios.

desk verdict A sensible extension of semantic HARQ to JSCC with a clear system design, but the BLEU-based training/evaluation loop and missing comparisons leave the claimed gains under-supported. read the letter →

arxiv 2504.14615 v1 pith:NDMDK5LR submitted 2025-04-20 cs.IT math.IT

Yongkang Li , Xu Wang , Zheng Shi , Yaru Fu This is my paper · ORCID

classification cs.ITmath.IT

keywords semanticcommunicationhybridautomaticrepeatrequestjointsource-channelcodingintelligenttransportationsystemsretransmissionschemesgenerativesignalreconstructionerrordetectiontexttransmissionoverfadingchannels

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Semantic HARQ retrofits a hybrid automatic repeat request loop onto a deep joint source-channel coding link for text, so that retransmissions are triggered by loss of meaning rather than by bit errors. The paper claims this raises end-to-end reliability on fading channels: all three proposed schemes beat a no-HARQ JSCC baseline, and the gains are largest at low signal-to-noise ratios. The central move is to give the receiver a learned error detector and a generative signal corrector that repairs channel-corrupted semantic vectors before deciding whether to ask for a retransmission. If the claim holds, semantic links in vehicular networks can run at lower transmit power or over longer ranges without sacrificing the meaning of exchanged text.

What carries the argument

The receiver-side reconstructor is the load-bearing mechanism. A discriminator network $D_A(\cdot)$ averages and classifies each received semantic vector as clean or corrupted; corrupted vectors pass to a conditional generative network $G_A(\cdot)$ that regenerates a cleaned version, while clean vectors pass through untouched. A transformer-based semantic error detector then maps the decoded sentence to an ACK/NACK confidence score, and the confidence scores from failed rounds double as weights in the combining rule $\sum_i a_i B_i$ with $a_i = \hat{p}_i / \sum_k \hat{p}_k$. For sem-HARQ-SC, an adaptive synonym-replacement mapping produces a rephrased version of the sentence for each retransmission, adding incremental semantic redundancy.

What would settle it

Run the same system over the reported Rayleigh fading model, collect decoded sentences, and have human raters or a downstream driving task label whether the meaning survived. If a substantial share of sentences with BLEU above 0.9 are judged to have lost meaning, or sentences below 0.9 still carry the intended meaning, the detector's training labels are wrong and the reliability gains are an artifact of the proxy rather than of true semantic fidelity.

Watch

Extended reading notes

Core claim

The paper's central claim is that HARQ can be moved from the bit level to the semantic level in JSCC-based text transmission. It reports that detecting semantic errors with a trained classifier and correcting corrupted received signals with a conditional generative network before retransmission lets all three proposed sem-HARQ schemes outperform a JSCC system without HARQ on both BLEU and sentence similarity. Among the schemes, keeping and combining previously failed signals (sem-HARQ-WC) beats discarding them (sem-HARQ-I), feature-level combining beats decision-level combining, confidence-weighted combining beats equal combining, and synonym-based incremental redundancy (sem-HARQ-SC) gives the best sentence similarity while trailing on BLEU.

Load-bearing premise

The whole loop is trained and evaluated on the assumption that a decoded sentence is semantically correct exactly when its BLEU score against the original is above 0.9; if that proxy mislabels meaning-bearing paraphrases or meaning-changing near-matches, the retransmission decisions and the reported gains inherit the error.

Editorial extensions

If this is right

At low SNR, all three proposed schemes raise BLEU and sentence similarity relative to a no-HARQ JSCC link; the largest reported gain is a 37.8% BLEU improvement at 0 dB for confidence-weighted feature-level combining.
Semantic retransmissions buy roughly 4 dB of effective SNR: the sem-HARQ system crosses the 0.9 BLEU/similarity threshold around 8 dB, while the no-HARQ baseline needs about 12 dB.
Reusing previously failed signals is the main lever: sem-HARQ-WC beats sem-HARQ-I in every comparison, and feature-level fusion beats decision-level fusion regardless of combination method.
Synonymous redundancy (sem-HARQ-SC) trades word-level fidelity for sentence-level meaning: it trails sem-HARQ-WC on BLEU but leads on sentence similarity.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A task-level definition of semantic correctness (for example, whether a vehicle can act on the transmitted instruction) would be a natural replacement for the BLEU>0.9 rule; the detector and combining weights could be retrained on that label, and the ranking of SC versus WC might flip because paraphrases preserve task meaning while lowering n-gram overlap.
The same discriminator-plus-generator corrector could be ported to image or video semantic links, with the local knowledge base constructed from a different corruption model and a modality-appropriate correctness label.
The ablation result that unfiltered correction hurts high-SNR performance suggests a confidence-gated bypass: a control policy could disable the generative corrector when the discriminator is confident the signal is clean, saving computation exactly where correction is unnecessary.
Because the gains concentrate at low SNR, sem-HARQ is best read as a coverage-extension tool for semantic links, not a peak-performance booster; deployed on a vehicular network it could let the same transmit power serve more distant or more obstructed vehicles.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Yongkang Li and colleagues have written a clear, workmanlike paper that extends semantic HARQ to JSCC-based text transmission. The core idea is plausible: use a learned detector to decide when to request retransmission, and combine signals across rounds at feature or decision level, with a synonym-based incremental redundancy option. The specific combination of a generative signal reconstructor (discriminator plus conditional generator) and a transformer-based error detector is new, and the three proposed schemes are sensible engineering. The ITS framing is mostly motivational; the model is a generic point-to-point text link over Rayleigh fading.

The paper does several things well. The system model is complete, the training procedure for the HARQ modules is described in detail, and the ablation study is informative: feeding all received signals through the corrector without the discriminator hurts performance, which validates the gating design. The simulation results are internally consistent and show the expected ranking among the schemes.

The main soft spot is the evaluation. The semantic error detector is trained on labels defined by BLEU>0.9, and BLEU is also the headline performance metric. So the retransmission policy is effectively fitted to the reported metric, and part of the measured BLEU gain comes from the labeling rule itself. Sentence similarity results are a step toward independent validation, but the detector is not trained on that metric, so the circularity is only partially broken. The paper acknowledges the label-noise issue but offers no human evaluation or task-level validation.

A second issue is the thin comparison set. The baseline is only no-HARQ. Previous semantic HARQ schemes ([22], [23], [25]-[28]) are cited but not compared, so the claim of superiority is not established. Two key parameters—the acceptance threshold lambda and maximum retransmissions M—are not reported, leaving the operating point and retransmission cost unspecified. Without those, the low-SNR gains could largely stem from extra transmission opportunities. The VC dimension bound in Section IV.D is a standard theorem quoted without adaptation to the specific model; it adds little.

This is an honest, readable paper that deserves a serious referee. I would recommend conditional acceptance only if the authors add comparisons to prior semantic HARQ, report lambda and M, and validate the BLEU threshold against human or task-level semantic judgments. For readers working on semantic communication reliability, this is worth reading, but the results should be treated with caution.

Referee Report

3 major / 5 minor

Summary. The paper proposes a semantic HARQ framework for JSCC-based text transmission in intelligent transportation systems. A receiver-side reconstructor, composed of a conditional generative network and a discriminator, attempts to repair corrupted semantic signals, while a transformer-based semantic error detector decides whether to send ACK or NACK. Three retransmission schemes are defined: Type I (discarding the erroneous signal and retransmitting the identical semantic signal), weighted combining at the feature or decision level (sem-HARQ-WC), and synonymous combining based on adaptive synonym replacement (sem-HARQ-SC). Experiments on the Europarl corpus over Rayleigh fading channels report BLEU and sentence similarity versus SNR, comparing the schemes against a no-HARQ JSCC baseline and among themselves.

Significance. If the reported gains are robust, the paper would make a useful contribution to an underexplored area: integrating HARQ with JSCC without a bit-level CRC. The strengths are a clear problem formulation, the explicit design of three distinct combining strategies, and an attempt to construct a local knowledge base for semantic error detection. The central limitation is the evaluation methodology: the semantic error detector is trained on labels defined by BLEU > 0.9, and BLEU is also the headline performance metric, which creates a circularity. The paper offers no human evaluation or downstream task validation, no comparison with prior semantic HARQ schemes, and no reporting of the retransmission budget and acceptance threshold. The sentence-similarity results provide some independent evidence, but the retransmission policy is not trained on that metric. Overall, the contribution is promising but preliminary; the empirical claim needs strengthening before the results can be accepted.

major comments (3)

[Section IV.D and Section V.C] The semantic error detector D_delta is trained on labels C=1 iff BLEU>0.9 (Section IV.D), and BLEU is used as the headline performance metric in Figs. 9 and 11 (Section V.C). The retransmission policy is therefore optimized to satisfy the very criterion used to measure success, so the reported BLEU gains may reflect fitting to the proxy rather than improved semantic correctness. The paper itself acknowledges in Section IV.A that 'labels in the training samples may not correspond to their true values,' but it never validates the BLEU threshold against human judgment or a downstream ITS task. The sentence-similarity metric is reported as well, but since the detector is not trained on it, it only partially addresses the circularity concern.
[Section II.A and Section V.C] The maximum retransmission count M and the acceptance threshold lambda are defined in Section II.A but their values are never reported in Section V. Without these values, the reader cannot distinguish gains from the proposed combining schemes from gains due to unlimited retransmission attempts. The no-HARQ baseline has no retransmission budget, so the comparisons in Figs. 9-12 do not account for the additional energy, latency, or spectrum consumed by retransmissions. Reporting performance at a fixed total resource budget, or at least reporting the average number of HARQ rounds and total transmission energy, is necessary to support the claim that sem-HARQ 'significantly improves reliability.'
[Section V.C] The only baseline is a no-HARQ JSCC system. Since HARQ by construction gives the receiver more transmission opportunities, outperforming no-HARQ is expected; the paper does not compare against prior semantic HARQ schemes (e.g., [22], [23], [25], [26]) or against a conventional bit-level HARQ wrapper around the same JSCC encoder. Additionally, Figs. 9-14 show no error bars or confidence intervals, and no code or data is released, so the statistical significance of the reported 30-38% improvements cannot be assessed.

minor comments (5)

[Section II.C.2] BERT is described as 'a pre-trained, decoder-only large language model,' but BERT is an encoder-only model; this is a factual error.
[Section III.C] The description of the synonym replacement range 'from L(m-1)/M to Lm/M' is ambiguous because these appear to be word indices rather than fractions, and the notation eSm is used before it is defined; the sentence 'applies a synonymous mapping to the T (·)' also contains an undefined placeholder.
[Section V.C and Figures 9-12] The figure legends use inconsistent scheme names (e.g., 'HARQ-SS-I', 'HARQ-SS-II-FC+Weight', 'HARQ-SS-IS') that do not match the sem-HARQ terminology used in the text; please unify the naming.
[Figures 13 and 14] The captions of Figs. 13 and 14 do not identify which line corresponds to which sample size, and no legend is visible in the reproduced figures; please add legends that map line styles to the different sample counts.
[Section I.A and Section IV.D] There are several language errors, including 'Zheng emphet al.' and 'an non-terrestrial network' in Section I.A, and 'hence, The constructed local knowledge base' with incorrect capitalization in Section IV.D; the manuscript should be carefully proofread.

Circularity Check

1 steps flagged · score 5.0 of 10

The semantic error detector is trained on BLEU>0.9 labels, the retransmission loop stops on that detector's output, and BLEU is then a headline evaluation metric; part of the reported reliability gain is a control-loop artifact of the metric.

self definitional [Section IV.D (Local Knowledge Base Generation and Training) combined with Section V.C (Verifications of Sem-HARQ Schemes)]
"we classify sentences with BLEU scores above 0.9 as belonging to the normal observation space, while those with scores below this threshold are categorized as part of the abnormal observation space. Specifically, when the BLEU score is below 0.9, the classification label C is set to 0; otherwise, it is set to 1."

The semantic error detector D_delta is trained with these BLEU-derived labels, and its ACK/NACK output decides whether to stop retransmitting. The evaluation then reports BLEU as a headline reliability metric. Hence the retransmission policy is a controller whose stopping rule is trained to recognize the BLEU>0.9 condition, so the reported BLEU improvement over No-HARQ is partly produced by the evaluation metric itself: the system keeps retransmitting until a BLEU-based classifier says the sentence is acceptable. This is a self-referential loop, not an independent test of semantic correctness.

full rationale

The paper is a systems paper with an empirical claim; there is no imported uniqueness theorem or self-citation chain that forces the architecture. The main circularity is in the evaluation loop: Section IV.D defines the 'normal observation space' and the classifier labels by the BLEU>0.9 threshold, and the same BLEU metric is used as a headline result in Section V.C. Since the semantic error detector drives retransmissions until it classifies a decoded sentence as normal, the HARQ policy is effectively a BLEU-threshold controller. This makes part of the BLEU gain over the No-HARQ baseline a consequence of the metric used to define both 'correct' and 'success.' The paper itself concedes in Section IV.A that the labels may not correspond to true semantic values. Sentence similarity is not the training target and therefore provides some independent signal, but it is correlated with BLEU and does not independently validate the feedback loop. Separately, the No-HARQ baseline does not consume retransmissions, so the comparison does not hold energy or latency fixed; that is a fairness concern rather than circularity. I therefore score partial circularity, not full equivalence: the underlying JSCC encoder/decoder, the generative corrector, and the combining schemes are trained separately and could genuinely contribute, so the central claim is not entirely forced.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on three free parameters (BLEU threshold, lambda, and M) and three domain assumptions. The BLEU threshold is the most consequential because it defines both the training labels and the evaluation metric. No new physical or conceptual entities are introduced.

free parameters (3)

BLEU threshold for semantic error labels = 0.9
Hand-chosen in Section IV.D to split normal vs abnormal observations; directly determines the labels used to train the semantic error detector.
Acceptance threshold lambda = Not specified
Used in sem-HARQ-I and others to decide ACK/NACK from detector confidence (Section III-A); value never given, affecting all retransmission decisions.
Maximum retransmissions M = Not specified
Defines the stopping condition for all schemes; its value impacts BLEU/latency trade-off and is not stated in simulations.

assumptions (3)

domain assumption Rayleigh fading channel with E{|h|^2}=1
Section V-A; all simulations rely on this channel model, which may not represent vehicular channels with Doppler and frequency-selective fading despite Eq. (2).
ad hoc to paper BLEU > 0.9 is a valid definition of semantic correctness
Section IV.D; the entire detector training and therefore the HARQ feedback uses this threshold, and no human or task-level validation is provided.
domain assumption Europarl corpus adequately represents ITS text traffic
Section V-A; a parliamentary corpus is used as the knowledge base, but vehicle-to-everything communication involves different text types such as safety messages.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Semantic HARQ for Intelligent Transportation Systems: Joint Source-Channel Coding-Powered Reliable Retransmissions." pith.science (2026). https://pith.science/paper/NDMDK5LR

@misc{pith2026250414615,
  author       = {Pith},
  title        = {Pith review of: Semantic HARQ for Intelligent Transportation Systems: Joint Source-Channel Coding-Powered Reliable Retransmissions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NDMDK5LR}},
  note         = {Machine review of arXiv:2504.14615}
}

read the original abstract

The surge of data traffic in Intelligent Transportation Systems (ITS) places a significant challenge on limited wireless resources. Semantic communication, which transmits essential semantics of the raw data, offers a promising solution by reducing redundancy and improving spectrum efficiency. However, high vehicle mobility, dynamic channel conditions, and dense vehicular networks severely impact transmission reliability in ITS. To address these limitations, we integrate Hybrid Automatic Repeat reQuest (HARQ) with Joint Source-Channel Coding (JSCC) to provide reliable semantic communications for ITS. To counteract the adverse effects of time-varying fading channels and noise, we propose a generative signal reconstructor module supported by a local knowledge base, which employs a discriminator for channel error detection and a conditional generative network for error correction. We propose three innovative semantic HARQ (sem-HARQ) schemes, Type I sem-HARQ (sem-HARQ-I), sem-HARQ with weighted combining (sem-HARQ-WC), and sem-HARQ with synonymous combining (sem-HARQ-SC) to enable reliable JSCC-based semantic communications. At the transmitter, both sem-HARQ-I and sem-HARQ-WC retransmit the same semantic signals, while sem-HARQ-SC introduces redundant semantics across different HARQ rounds through synonymous mapping. At the receiver, sem-HARQ-I performs semantic decoding based solely on the currently received signal. In contrast, sem-HARQ-WC enhances reliability by fusing the current received semantic signal with prior erroneous signals at the feature or decision level, thereby exploiting semantic information from failed HARQ rounds. Similarly, sem-HARQ-SC employs feature-level combining, leveraging incremental semantic redundancy to merge semantic features from retransmissions.

Figures

Figures reproduced from arXiv: 2504.14615 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 21 canonical work pages

[22]

Deep source-channel coding for sentence semantic transmission with HARQ,

P. Jiang, C.-K. Wen, S. Jin, and G. Y . Li, “Deep source-channel coding for sentence semantic transmission with HARQ,” IEEE Trans. Commun., vol. 70, no. 8, pp. 5225–5240, Aug. 2022

2022
[23]

Adaptive bit rate control in semantic communication with incremental knowledge-based HARQ,

Q. Zhou, R. Li, Z. Zhao, Y . Xiao, and H. Zhang, “Adaptive bit rate control in semantic communication with incremental knowledge-based HARQ,” IEEE Open J. Commun. Soc. , vol. 3, pp. 1076–1089, Aug. 2022

work page 2022
[25]

Adaptive HARQ design for semantic image transmission,

H. Liu, M.-M. Zhao, M. Lei, L. Li, Y . Cai, and M.-J. Zhao, “Adaptive HARQ design for semantic image transmission,” in Proc. IEEE Veh. Technol. Conf. (VTC-Fall), 2024, pp. 1–6

work page 2024
[28]

Semantic-aware jointed coding and routing design in large-scale satellite networks: A deep learning approach,

R. Gao, Y . Xu, H. Li, Q. Zhang, and Z. Yang, “Semantic-aware jointed coding and routing design in large-scale satellite networks: A deep learning approach,” IEEE/ACM Trans. Netw., vol. 32, no. 6, pp. 5415– 5429, Dec. 2024

work page 2024
[26]

Semantic communication for cooperative perception with HARQ,

Y . Sheng, H. Ye, L. Liang, and S. Jin, “Semantic communication for cooperative perception with HARQ,” inProc. IEEE Int. Workshop Mach. Learn. Signal Process. (MLSP) , 2024, pp. 1–6

work page 2024
[1]

A comprehensive survey on communication techniques for the realization of intelligent transportation systems in iot based smart cities,

Y . Rajkumar and S. Santhosh Kumar, “A comprehensive survey on communication techniques for the realization of intelligent transportation systems in iot based smart cities,” Peer-to-Peer Netw. Appl. , vol. 17, no. 3, pp. 1263–1308, Feb. 2024

work page 2024
[2]

A mathematical theory of communication,

C. E. Shannon, “A mathematical theory of communication,” Bell Syst. Tech. J., vol. 27, no. 3, pp. 379–423, Jul. 1948

1948
[3]

A Survey on Semantic Communications in Internet of Vehicles

S. Ye, Q. Wu, P. Fan, and Q. Fan, “A survey on semantic communications in internet of vehicles,” arXiv preprint arXiv:2503.03767 , Mar. 2025

work page Pith review arXiv 2025

Show all 31 references

[4]

Semantic communication for the internet of vehicles: A multiuser cooperative approach,

W. Xu, Y . Zhang, F. Wang, Z. Qin, C. Liu, and P. Zhang, “Semantic communication for the internet of vehicles: A multiuser cooperative approach,” IEEE Veh. Technol. Mag., vol. 18, no. 1, pp. 100–109, Jan. 2023

2023
[5]

Socially-inspired semantic communication codec updating for NTN-enabled intelligent transportation systems,

G. Zheng, Q. Ni, K. Navaie, and C. Zarakovitis, “Socially-inspired semantic communication codec updating for NTN-enabled intelligent transportation systems,” IEEE Trans. Intell. Transp. Syst., pp. 1–13, Mar. 2025

2025
[6]

Image segmentation semantic communication over internet of vehicles,

Q. Pan, H. Tong, J. Lv, T. Luo, Z. Zhang, C. Yin, and J. Li, “Image segmentation semantic communication over internet of vehicles,” in Proc. IEEE Wireless Commun. Netw. Conf. (WCNC) , 2023, pp. 1–6

2023
[7]

End-to-end fast training of com- munication links without a channel model via online meta-learning,

S. Park, O. Simeone, and J. Kang, “End-to-end fast training of com- munication links without a channel model via online meta-learning,” in Proc. IEEE 21st Int. Workshop Signal Process. Adv. Wireless Commun. (SPAWC), 2020, pp. 1–5

2020
[8]

Deep learning for joint source- channel coding of text,

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

2018
[9]

Deep learning enabled semantic communication systems,

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

2021
[10]

Toward semantic communications: Deep learning-based image semantic coding,

D. Huang, F. Gao, X. Tao, Q. Du, and J. Lu, “Toward semantic communications: Deep learning-based image semantic coding,” IEEE J. Sel. Areas Commun. , vol. 41, no. 1, pp. 55–71, Nov. 2023

2023
[11]

Wireless deep video semantic transmission,

S. Wang, J. Dai, Z. Liang, K. Niu, Z. Si, C. Dong, X. Qin, and P. Zhang, “Wireless deep video semantic transmission,” IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 214–229, Jan. 2023

2023
[12]

Deep semantic commu- nication for knowledge sharing in internet of vehicles,

Z. Wang, S. Leng, H. Zhang, and C. Yuen, “Deep semantic commu- nication for knowledge sharing in internet of vehicles,” IEEE Internet Things J., pp. 1–1, Jan. 2025

2025
[13]

Deep learning-empowered semantic communication systems with a shared knowledge base,

P. Yi, Y . Cao, X. Kang, and Y .-C. Liang, “Deep learning-empowered semantic communication systems with a shared knowledge base,” IEEE Trans. Wireless Commun., vol. 23, no. 6, pp. 6174–6187, Nov. 2023

2023
[14]

Domain knowledge driven semantic communication for image transmission over wireless channels,

A. Li, X. Liu, G. Wang, and P. Zhang, “Domain knowledge driven semantic communication for image transmission over wireless channels,” IEEE Wireless Commun. Lett. , vol. 12, no. 1, pp. 55–59, Jan. 2023

2023
[15]

Deep learning- enabled semantic communication systems with task-unaware transmitter and dynamic data,

H. Zhang, S. Shao, M. Tao, X. Bi, and K. B. Letaief, “Deep learning- enabled semantic communication systems with task-unaware transmitter and dynamic data,” IEEE J. Sel. Areas Commun. , vol. 41, no. 1, pp. 170–185, Jan. 2023

2023
[16]

Cognitive semantic communication systems driven by knowl- edge graph: Principle, implementation, and performance evaluation,

F. Zhou, Y . Li, M. Xu, L. Yuan, Q. Wu, R. Q. Hu, and N. Al- Dhahir, “Cognitive semantic communication systems driven by knowl- edge graph: Principle, implementation, and performance evaluation,” IEEE Trans. Commun. , vol. 72, no. 1, pp. 193–208, Jan. 2024

2024
[17]

A robust semantic text communication system,

X. Peng, Z. Qin, X. Tao, J. Lu, and L. Hanzo, “A robust semantic text communication system,” IEEE Trans. Wireless Commun., pp. 1–1, Sept. 2024. JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 14

2024
[18]

One-to-many semantic communication systems: Design, implementation, performance evaluation,

H. Hu, X. Zhu, F. Zhou, W. Wu, R. Q. Hu, and H. Zhu, “One-to-many semantic communication systems: Design, implementation, performance evaluation,” IEEE Commun. Lett. , vol. 26, no. 12, pp. 2959–2963, Dec. 2022

2022
[19]

A scalable semantic communi- cation system based on meta-learning,

H. Chen, S. Chen, C. He, and H. Li, “A scalable semantic communi- cation system based on meta-learning,” in Proc. Int. Wireless Commun. Mobile Comput. (IWCMC) , 2024, pp. 561–566

2024
[20]

Pilot optimization and channel estimation scheme for semantic com- munication: A framework for edge intelligence,

K. Kim, Y . K. Tun, M. Shirajum Munir, W. Saad, and C. S. Hong, “Pilot optimization and channel estimation scheme for semantic com- munication: A framework for edge intelligence,” in Proc. IEEE Net. Oper. Manag. Symp., 2024, pp. 1–7

2024
[21]

A lite distributed semantic communication system for internet of things,

H. Xie and Z. Qin, “A lite distributed semantic communication system for internet of things,” IEEE Sel. Areas Commun , vol. 39, no. 1, pp. 142–153, Jan. 2021

2021
[24]

Semantic communication with conceptual spaces,

D. Wheeler, E. E. Tripp, and B. Natarajan, “Semantic communication with conceptual spaces,” IEEE Communications Letters , vol. 27, no. 2, pp. 532–535, 2022

2022
[27]

Wireless semantic commu- nications for video conferencing,

P. Jiang, C.-K. Wen, S. Jin, and G. Y . Li, “Wireless semantic commu- nications for video conferencing,” IEEE J. Sel. Areas Commun , vol. 41, no. 1, pp. 230–244, Jan. 2023

2023
[29]

Semantic communication with conceptual spaces,

D. Wheeler, E. E. Tripp, and B. Natarajan, “Semantic communication with conceptual spaces,” IEEE Commun. Lett. , vol. 27, no. 2, pp. 532– 535, Feb. 2023

2023
[30]

Vapnik-chervonenkis dimension of neural nets,

P. L. Bartlett and W. Maass, “Vapnik-chervonenkis dimension of neural nets,” The handbook of brain theory and neural networks , pp. 1188– 1192, 2003

2003
[31]

Europarl: A parallel corpus for statistical machine transla- tion,

P. Koehn, “Europarl: A parallel corpus for statistical machine transla- tion,” in Proc. MT Summit , vol. 5, 2005, pp. 79–86

2005

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [22]

Deep source-channel coding for sentence semantic transmission with HARQ,

P. Jiang, C.-K. Wen, S. Jin, and G. Y . Li, “Deep source-channel coding for sentence semantic transmission with HARQ,” IEEE Trans. Commun., vol. 70, no. 8, pp. 5225–5240, Aug. 2022

2022

[2] [23]

Adaptive bit rate control in semantic communication with incremental knowledge-based HARQ,

Q. Zhou, R. Li, Z. Zhao, Y . Xiao, and H. Zhang, “Adaptive bit rate control in semantic communication with incremental knowledge-based HARQ,” IEEE Open J. Commun. Soc. , vol. 3, pp. 1076–1089, Aug. 2022

work page 2022

[3] [25]

Adaptive HARQ design for semantic image transmission,

H. Liu, M.-M. Zhao, M. Lei, L. Li, Y . Cai, and M.-J. Zhao, “Adaptive HARQ design for semantic image transmission,” in Proc. IEEE Veh. Technol. Conf. (VTC-Fall), 2024, pp. 1–6

work page 2024

[4] [28]

Semantic-aware jointed coding and routing design in large-scale satellite networks: A deep learning approach,

R. Gao, Y . Xu, H. Li, Q. Zhang, and Z. Yang, “Semantic-aware jointed coding and routing design in large-scale satellite networks: A deep learning approach,” IEEE/ACM Trans. Netw., vol. 32, no. 6, pp. 5415– 5429, Dec. 2024

work page 2024

[5] [26]

Semantic communication for cooperative perception with HARQ,

Y . Sheng, H. Ye, L. Liang, and S. Jin, “Semantic communication for cooperative perception with HARQ,” inProc. IEEE Int. Workshop Mach. Learn. Signal Process. (MLSP) , 2024, pp. 1–6

work page 2024

[6] [1]

A comprehensive survey on communication techniques for the realization of intelligent transportation systems in iot based smart cities,

Y . Rajkumar and S. Santhosh Kumar, “A comprehensive survey on communication techniques for the realization of intelligent transportation systems in iot based smart cities,” Peer-to-Peer Netw. Appl. , vol. 17, no. 3, pp. 1263–1308, Feb. 2024

work page 2024

[7] [2]

A mathematical theory of communication,

C. E. Shannon, “A mathematical theory of communication,” Bell Syst. Tech. J., vol. 27, no. 3, pp. 379–423, Jul. 1948

1948

[8] [3]

A Survey on Semantic Communications in Internet of Vehicles

S. Ye, Q. Wu, P. Fan, and Q. Fan, “A survey on semantic communications in internet of vehicles,” arXiv preprint arXiv:2503.03767 , Mar. 2025

work page Pith review arXiv 2025