On the Resilience of Underwater Semantic Wireless Communications

Filipe B. Teixeira; Jo\~ao Pedro Loureiro; Patr\'icia Delgado; Rui Campos; Tom\'as Feliciano Ribeiro

REVIEW 3 major objections 5 minor 1 cited by

On the Resilience of Underwater Semantic Wireless Communications

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read SAGE keeps image content recognizable when up to 20% of transmitted text is corrupted.

desk verdict A modest, honest resilience study whose headline threshold rests on an unvalidated random-error simulator; should be sent to review with conditions. read the letter →

arxiv 2506.23350 v1 pith:JUJV4LHQ submitted 2025-06-29 cs.NI

João Pedro Loureiro , Patrícia Delgado , Tomás Feliciano Ribeiro , Filipe B. Teixeira , Rui Campos This is my paper · ORCID

classification cs.NI

keywords underwateracousticcommunicationssemanticgenerativeAIimagecompressionerrorresilienceCLIPScoreBLIPStableDiffusion

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 communications try to send the meaning of an image rather than the image itself, and this paper asks how much textual corruption that idea survives in underwater acoustic channels. The authors test SAGE, a pipeline that turns images into short captions with BLIP, sends the caption as text, and reconstructs the image with Stable Diffusion. They inject random character substitutions, character deletions, and word deletions at controlled rates, and measure how well the output image still matches the original. The paper finds that meaningful reconstruction persists at about 15–20% character error ratio for substitution and deletion, and up to about 50% for word deletion, so semantic communication can absorb sizable error without collapsing.

What carries the argument

The load-bearing mechanism is the text-level error simulator placed between the BLIP image-to-text encoder and the Stable Diffusion text-to-image decoder. It injects one of three independent error types at a configurable ratio and feeds the corrupted caption to the generative decoder, with CLIPScore as the semantic similarity metric. CLIPScore, computed between the original and reconstructed images, is the instrument that reveals the resilience threshold; PSNR and SSIM serve as counterexamples that show why literal pixel comparison misses semantic preservation.

What would settle it

Replay recorded underwater acoustic modem bit-error patterns through the SAGE text decoder, or run SAGE over a real acoustic link, and compare CLIPScore against the paper's random-corruption curve; if semantic scores collapse at error rates the simulator calls tolerable, the resilience result is an artifact of the error model.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes an empirical resilience boundary for SAGE: semantic image reconstruction remains viable up to moderate error ratios, with CLIPScore falling only from roughly 60% to 50% at around 15% character substitutions or 20% character deletions, and a similar degradation appearing only near 50% word deletion. It also shows that pixel-level metrics (PSNR and SSIM) are essentially insensitive to this kind of degradation, so they cannot be used to judge semantic communication quality. Character-level corruption hurts more than word-level deletion, which the paper attributes to non-essential words carrying little meaning, while partial corruption of many words degrades the whole prompt.

Load-bearing premise

The simulated random substitutions, deletions, and word deletions represent the mismatches a real underwater acoustic channel produces.

Editorial extensions

If this is right

A SAGE-like link can skip heavy error correction for every character and still deliver useful images under acoustic noise, as long as the character error ratio stays below about 20%.
Underwater image transmission can be compressed to SMS-scale payloads without sacrificing semantic robustness.
Channel coding effort is better spent protecting characters than preserving every word, since whole-word deletion is tolerated far better than character corruption.
Evaluation of semantic systems should rely on semantic metrics like CLIPScore; PSNR and SSIM will wrongly indicate failure when images are semantically faithful but pixel-different.

Reading between the lines

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

If real underwater acoustic errors are bursty or correlated rather than random, these thresholds would likely move; a natural next step is replaying measured modem error traces through the simulator.
The CLIPScore numbers are relative to a particular control image, so the exact 50% floor is setup-dependent; the robust qualitative ordering (word deletion easier than character corruption) is the transferable result.
The resilience boundary is coupled to the captioning model's vocabulary; a future image-to-text model that uses rarer, more information-dense words may have a different tolerance curve.

Share X Bluesky LinkedIn Reddit HN

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

First, the useful thing: this is the first error-injection study on the SAGE image-to-text/text-to-image link, and it delivers a concrete, reproducible-in-spirit result: under random independent character substitution and deletion, CLIPScore starts degrading noticeably around 15-20% character error, while deleting whole words does less damage until much higher ratios. The PSNR/SSIM insensitivity is a nice demonstration that pixel metrics are the wrong tool here. The paper is honest about reusing SAGE; the self-citation is appropriate, and there is no fitted-parameter circularity.

The main soft spot is exactly the one the stress-test note flags. The simulator injects errors as independent random substitutions/deletions, but the abstract and conclusions treat it as representative of 'character errors observed in underwater acoustic channels' without any validation against measured traces, Bellhop-like models, or modem output. Acoustic errors are typically bursty and correlated; at the same CER, a burst that wipes out a phrase may hurt Stable Diffusion more than scattered single-character flips. So the 15-20% threshold is a property of the random-error model, not yet a property of underwater channels. The fix is not huge: either validate the error model or add a qualifier in the title/abstract that this is a first-order resilience screen under simulated text corruption.

The other weaknesses are minor. The dataset is 30 images; fine for a trend, but the paper should say so. Stable Diffusion is stochastic and they averaged 10 generations per condition, but no variance or confidence intervals are shown, so we cannot see whether the drop at 15-20% is bigger than run-to-run noise. The word-deletion condition uses a different error unit (words vs characters), which makes the comparison across error types a bit apples-to-oranges. 'Viable' is not defined quantitatively; a 10-point CLIPScore drop may or may not matter depending on application.

This is a modest but useful paper for people building text-based semantic links over acoustic modems. It deserves peer review, not a desk reject, with the error-model validation and variance reporting as conditions. I would not object to citing it as evidence that semantic links tolerate text errors in simulation, but I would not cite the specific threshold as a channel property.

Referee Report

3 major / 5 minor

Summary. This paper evaluates the resilience of SAGE, an image-to-text/text-to-image semantic communication framework, under simulated text-level transmission errors. The authors use BLIP to caption 30 underwater images, inject random character substitutions, character deletions, and word deletions via a custom simulator, feed the corrupted captions to Stable Diffusion for image reconstruction, and compare original and reconstructed images with PSNR, SSIM, and CLIPScore. The central claim is that SAGE maintains semantically meaningful reconstruction up to moderate error ratios, specifically 15–20% character error ratio, with word deletion appearing less harmful than character-level corruption (Section III-D).

Significance. If the result holds, this is a useful initial demonstration that a generative semantic communication pipeline can tolerate textual corruption, with potential bandwidth and robustness benefits for underwater acoustic links. The study has notable strengths: it builds on externally pretrained models (BLIP, Stable Diffusion, and CLIP), uses no fitted parameters, averages 10 Stable Diffusion runs per condition, and includes a control image as a semantic-similarity reference. However, the quantitative resilience threshold is tied to a synthetic independent-error model that is not validated against any physical acoustic channel, and the statistical evidence is limited by a 30-image dataset and the absence of variance or confidence-interval reporting. These issues currently limit the strength of the central claim.

major comments (3)

[Section III-A and III-D] The paper's central resilience threshold is derived entirely from a simulator that injects character substitutions, character deletions, and word deletions as independent random events. The abstract states that these are "character errors observed in underwater acoustic channels," and Section III-A relates CER to BER via the bound BER ∈ [CER/8, CER], but no comparison is made to measured acoustic channel traces, standard acoustic channel models (e.g., Bellhop), or bit-level modem outputs. Real underwater acoustic channels exhibit bursty, temporally correlated errors due to multipath, Doppler, and impulsive noise; under the same average CER, a burst corrupting several adjacent characters or an entire word can be substantially more destructive than scattered single-character errors, which a language model can often repair. At minimum, the authors should validate the simulator against a standard channel error model or real modem error traces, or restrict the claims to independent text errors and remove the "observed in underwater acoustic channels" wording. This is load-bearing because the 15–20% CER threshold in Section III-D and the comparison of error types in Section III-C are statements about resilience in acoustic environments.
[Section III methodology and Figs. 12–14] The authors state that each Stable Diffusion image was generated 10 times "to account for randomness," but the reported figures show only average CLIPScore, PSNR, and SSIM curves, with no error bars, confidence intervals, or significance tests. Text-to-image diffusion models are highly stochastic, and with only 30 captions per condition, the reported 10-point CLIPScore drop (e.g., from 60% to 50% around 15–20% CER) may be comparable to run-to-run variance. The authors should report per-condition standard deviations or confidence intervals and test whether the observed degradation at 15–20% CER is statistically significant. Without this, the central "viable up to 15–20% CER" claim is not quantitatively supported.
[Section III-A and III-C] The simulator defines the error ratio for types 1 and 2 as the percentage of affected characters, but for type 3 as the percentage of affected words; Section III-C then refers to "CER" for all three types and concludes that character-level corruption is more harmful than word deletion. Comparing "15–20% CER" with "50% CER" across different units (characters versus words) is not an equivalent comparison of information loss, because a word contains multiple characters and word deletion also removes contextual structure. The authors should use a common corruption measure (e.g., expected characters removed or message-level edit distance) or explicitly compare semantic degradation per unit of textual information affected.

minor comments (5)

[Section II] There is a typo in Section II: "sign-inficantly" should be "significantly."
[Section III-B] The statement that PSNR "may range between 20 and 80 dB" is not generally correct; PSNR values depend on the data range and bit depth and are not confined to that interval. Please rephrase as an indicative range for the images used.
[References] Reference [23] is an informal Medium blog post about CLIP similarity. For CLIPScore, the authors should cite a more standard source, such as Hessel et al., "CLIPScore: A Reference-free Evaluation Metric for Image Captioning," 2021.
[Section III] The claim that repeated BLIP runs "consistently yielded nearly identical captions" is asserted without quantitative evidence; please report the observed caption similarity or specify the decoding settings that made the output deterministic.
[Section III] The evaluation uses only 30 underwater images and a single BLIP caption per image, with no per-image breakdown. Reporting per-image CLIPScore distributions would clarify how representative the average behavior is and how sensitive the conclusions are to individual images.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SAGE resilience is measured against external pretrained models with an open-loop error sweep, and the only self-citation introduces the framework rather than supplying the result.

full rationale

The paper's derivation chain is an empirical evaluation, not a derivation from assumptions that already contain the conclusion. SAGE is taken from the authors' prior work [12], but the resilience claim is tested independently: BLIP generates captions, the custom simulator injects controlled text errors, Stable Diffusion reconstructs images, and similarity is scored with PSNR, SSIM, and CLIPScore against the original images and a control image. No parameter is fitted to the data whose behavior is then 'predicted'; the error ratio is an input variable swept across values, and CLIPScore is an external pretrained metric. The BER bounds in Section III-A are a definitional mapping from CER to BER and are presented only as context, not as a predicted result. The single self-citation [12] defines SAGE itself and does not carry the burden of the resilience conclusion. The main weakness, that the simulator uses random independent character-level errors rather than validated bursty acoustic channel traces, is a threat to external validity and not a circularity, because the simulator's assumptions do not logically force the measured CLIPScore outcomes. The paper is self-contained against external benchmarks for what it claims to measure, so no circular step is present.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The paper introduces no fitted constants or new entities. Its central claim rests on two domain assumptions: that injected text errors approximate underwater acoustic impairments, and that a moderate CLIPScore decline is an acceptable proxy for meaningful reconstruction.

assumptions (2)

domain assumption Text-level random character substitutions, deletions, and word deletions are a sufficient proxy for underwater acoustic transmission errors.
The central resilience claim relies on this; the simulator injects independent errors, but Section III-A provides no validation against real acoustic channel measurements.
domain assumption A moderate decline in CLIPScore, e.g., from 60% to 50%, constitutes retaining semantically meaningful image content.
The paper never defines a threshold for meaningful reconstruction; resilience is inferred from CLIPScore remaining around 50% at 15-20% CER in Section III-D.

how reviews work

0 comments

Cite this review

Pith. "Pith review of On the Resilience of Underwater Semantic Wireless Communications." pith.science (2026). https://pith.science/paper/JUJV4LHQ

@misc{pith2026250623350,
  author       = {Pith},
  title        = {Pith review of: On the Resilience of Underwater Semantic Wireless Communications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JUJV4LHQ}},
  note         = {Machine review of arXiv:2506.23350}
}

read the original abstract

Underwater wireless communications face significant challenges due to propagation constraints, limiting the effectiveness of traditional radio and optical technologies. Long-range acoustic communications support distances up to a few kilometers, but suffer from low bandwidth, high error ratios, and multipath interference. Semantic communications, which focus on transmitting extracted semantic features rather than raw data, present a promising solution by significantly reducing the volume of data transmitted over the wireless link. This paper evaluates the resilience of SAGE, a semantic-oriented communications framework that combines semantic processing with Generative Artificial Intelligence (GenAI) to compress and transmit image data as textual descriptions over acoustic links. To assess robustness, we use a custom-tailored simulator that introduces character errors observed in underwater acoustic channels. Evaluation results show that SAGE can successfully reconstruct meaningful image content even under varying error conditions, highlighting its potential for robust and efficient underwater wireless communication in harsh environments.

Figures

Figures reproduced from arXiv: 2506.23350 by the authors.

**Figure 2.** Comparison between a semantic communications sys [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Semantic communication pipeline employed for a single [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: Flowchart of the text-level channel error simulator used [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]

**Figure 5.** Figure 5: Comparison of similarity metrics for the original image and: JPEG Compressed versions; a control image; and AI [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

**Figure 9.** Figure 9: Average SSIM vs Error Percentage (Error Type 1). [PITH_FULL_IMAGE:figures/full_fig_p005_9.png]

**Figure 7.** Figure 7: Average PSNR vs Error Percentage (Error Type 2). [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]

**Figure 11.** Figure 11: Average SSIM vs Error Percentage (Error Type 3). [PITH_FULL_IMAGE:figures/full_fig_p005_11.png]

**Figure 12.** Figure 12: Average CLIP Similarity Score vs Error Percentage [PITH_FULL_IMAGE:figures/full_fig_p006_12.png]

**Figure 13.** Figure 13: Average CLIP Similarity Score vs Error Percentage [PITH_FULL_IMAGE:figures/full_fig_p006_13.png]

**Figure 14.** Figure 14: Average CLIP Score vs Error Percentage (Error Type [PITH_FULL_IMAGE:figures/full_fig_p006_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

Semantic Communication for the Internet of Underwater Things: Architectures, Applications, Challenges, and Future Directions
eess.SP 2026-01 reject novelty 2.0 of 10

A survey of semantic communication for underwater IoT that compiles architectures, applications, and future directions, but contains internally inconsistent performance claims and many non-archival citations.

Reference graph

Works this paper leans on

23 extracted references · 18 canonical work pages · cited by 1 Pith paper

[1]

Survey on high reliability wireless communication for underwater sensor networks,

S. Li, W. Qu, C. Liu, T. Qiu, and Z. Zhao, “Survey on high reliability wireless communication for underwater sensor networks,” in Journal of Network and Computer Applications , 2019

work page 2019
[2]

Eval- uation of IEEE 802.11 Underwater Networks Operating at 700 MHz, 2.4 GHz and 5 GHz,

F. Teixeira, P. Freitas, L. Pessoa, R. Campos, and M. Ricardo, “Eval- uation of IEEE 802.11 Underwater Networks Operating at 700 MHz, 2.4 GHz and 5 GHz,” in Proceedings of the International Conference on Underwater Networks & Systems, Rome, Italy, November 12 - 14, 2014, 2014, pp. 11:1–11:5

work page 2014
[3]

A novel simulation platform for underwater data muling communications using autonomous underwater vehicles,

F. B. Teixeira, B. M. Ferreira, N. Moreira, N. Abreu, M. Villa, J. P. Loureiro, N. A. Cruz, J. C. Alves, M. Ricardo, and R. Campos, “A novel simulation platform for underwater data muling communications using autonomous underwater vehicles,” Computers, vol. 10, no. 10, p. 119, 2021

work page 2021
[4]

D. B. Alves, Blue Economy . Fundac ¸ ˜ao Francisco Manuel dos Santos, January 2022, ISBN: 9789899004955

work page 2022
[5]

DURIUS: A Multimodal Underwater Communications Approach for Higher Performance and Lower Energy Consumption,

J. P. Loureiro, F. B. Teixeira, and R. Campos, “DURIUS: A Multimodal Underwater Communications Approach for Higher Performance and Lower Energy Consumption,” in 2023 IEEE 9th WF-IoT , 2023

work page 2023
[6]

Adaptive and Reli- able Underwater Wireless Video Streaming using Data Muling,

J. P. Loureiro, F. B. Teixeira, and R. Campos, “Adaptive and Reli- able Underwater Wireless Video Streaming using Data Muling,” in IEEE/MTS OCEANS 2021, Porto and San Diego , 2021

work page 2021
[7]

Underwater acoustic modems,

S. Sendra, J. Lloret, J. M. Jimenez, and L. Parra, “Underwater acoustic modems,” IEEE Sensors Journal , vol. 16, no. 11, pp. 4063–4071, 2016

work page 2016
[8]

Semantic communications: Principles and challenges,

Z. Qin, X. Tao, J. Lu, and G. Y . Li, “Semantic communications: Principles and challenges,” CoRR, vol. abs/2201.01389, 2022. [Online]. Available: https://arxiv.org/abs/2201.01389

arXiv 2022

Show all 23 references

[9]

Design and experimental demonstration of underwater wireless optical communica- tion system based on semantic communication paradigm,

J. Xu, Z. Huang, Y . Gao, W. Zhai, H. Qiu, and Y . Ji, “Design and experimental demonstration of underwater wireless optical communica- tion system based on semantic communication paradigm,” Opt. Express, vol. 32, no. 2, pp. 2188–2201, Jan 2024

2024
[10]

Semantic communication in underwater communication: Advantage, problem and solution — a survey,

J. Zhang, W. Sun, Y . Zhao, and H. Du, “Semantic communication in underwater communication: Advantage, problem and solution — a survey,” in 2023 8th ICSP , 2023, pp. 2120–2123

2023
[11]

An underwater acoustic semantic communication approach to underwater image transmission,

Y . Zhang, H. Li, B. Li, L. Li, Z. Weibo, H. Wang, and P. Ren, “An underwater acoustic semantic communication approach to underwater image transmission,” Intelligent Marine Technology and Systems, vol. 3, 02 2025

2025
[12]

A semantic- oriented approach for underwater wireless communications using gener- ative ai,

J. P. Loureiro, A. Mateus, F. B. Teixeira, and R. Campos, “A semantic- oriented approach for underwater wireless communications using gener- ative ai,” in2024 15th IFIP Wireless and Mobile Networking Conference (WMNC), 2024, pp. 70–74

2024
[13]

Learning transferable visual models from natural language supervision,

A. R. et al., “Learning transferable visual models from natural language supervision,” CoRR, vol. abs/2103.00020, 2021

2021 arXiv
[14]

Vilbert: Pretraining task- agnostic visiolinguistic representations for vision-and-language tasks,

J. Lu, D. Batra, D. Parikh, and S. Lee, “Vilbert: Pretraining task- agnostic visiolinguistic representations for vision-and-language tasks,” CoRR, vol. abs/1908.02265, 2019

1908 arXiv
[15]

BLIP: bootstrapping language-image pre-training for unified vision-language understanding and generation,

J. Li, D. Li, C. Xiong, and S. C. H. Hoi, “BLIP: bootstrapping language-image pre-training for unified vision-language understanding and generation,” CoRR, 2022

2022
[16]

Blip-2: bootstrapping language- image pre-training with frozen image encoders and large language models,

J. Li, D. Li, S. Savarese, and S. Hoi, “Blip-2: bootstrapping language- image pre-training with frozen image encoders and large language models,” 2023

2023
[17]

Zero-shot text-to-image generation,

A. Ramesh, M. Pavlov, G. Goh, S. Gray, C. V oss, A. Radford, M. Chen, and I. Sutskever, “Zero-shot text-to-image generation,” CoRR, vol. abs/2102.12092, 2021

2021 arXiv
[18]

Photorealistic text-to-image diffusion models with deep language understanding,

C. S. et al., “Photorealistic text-to-image diffusion models with deep language understanding,” 2022. [Online]. Available: https://arxiv.org/abs/2205.11487

2022 arXiv
[19]

High- resolution image synthesis with latent diffusion models,

R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” CoRR, 2021

2021
[20]

U-shape transformer for underwater image enhancement,

L. Peng, C. Zhu, and L. Bian, “U-shape transformer for underwater image enhancement,” IEEE Transactions on Image Processing , vol. 32, pp. 3066–3079, 2023

2023
[21]

Chapter 4 - digital picture formats and representations,

D. R. Bull and F. Zhang, “Chapter 4 - digital picture formats and representations,” in Intelligent Image and Video Compression (Second Edition). Oxford: Academic Press, 2021

2021
[22]

Image quality assessment: from error visibility to structural similarity,

Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Trans- actions on Image Processing , vol. 13, no. 4, pp. 600–612, 2004

2004
[23]

Unlocking OpenAI CLIP. Part 2: image similarity,

“Unlocking OpenAI CLIP. Part 2: image similarity,” [Accessed: 10/04/2025]. [Online]. Available: https://medium.com/@jeremy- k/unlocking-openai-clip-part-2-image-similarity-bf0224ab5bb0

2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Survey on high reliability wireless communication for underwater sensor networks,

S. Li, W. Qu, C. Liu, T. Qiu, and Z. Zhao, “Survey on high reliability wireless communication for underwater sensor networks,” in Journal of Network and Computer Applications , 2019

work page 2019

[2] [2]

Eval- uation of IEEE 802.11 Underwater Networks Operating at 700 MHz, 2.4 GHz and 5 GHz,

F. Teixeira, P. Freitas, L. Pessoa, R. Campos, and M. Ricardo, “Eval- uation of IEEE 802.11 Underwater Networks Operating at 700 MHz, 2.4 GHz and 5 GHz,” in Proceedings of the International Conference on Underwater Networks & Systems, Rome, Italy, November 12 - 14, 2014, 2014, pp. 11:1–11:5

work page 2014

[3] [3]

A novel simulation platform for underwater data muling communications using autonomous underwater vehicles,

F. B. Teixeira, B. M. Ferreira, N. Moreira, N. Abreu, M. Villa, J. P. Loureiro, N. A. Cruz, J. C. Alves, M. Ricardo, and R. Campos, “A novel simulation platform for underwater data muling communications using autonomous underwater vehicles,” Computers, vol. 10, no. 10, p. 119, 2021

work page 2021

[4] [4]

D. B. Alves, Blue Economy . Fundac ¸ ˜ao Francisco Manuel dos Santos, January 2022, ISBN: 9789899004955

work page 2022

[5] [5]

DURIUS: A Multimodal Underwater Communications Approach for Higher Performance and Lower Energy Consumption,

J. P. Loureiro, F. B. Teixeira, and R. Campos, “DURIUS: A Multimodal Underwater Communications Approach for Higher Performance and Lower Energy Consumption,” in 2023 IEEE 9th WF-IoT , 2023

work page 2023

[6] [6]

Adaptive and Reli- able Underwater Wireless Video Streaming using Data Muling,

J. P. Loureiro, F. B. Teixeira, and R. Campos, “Adaptive and Reli- able Underwater Wireless Video Streaming using Data Muling,” in IEEE/MTS OCEANS 2021, Porto and San Diego , 2021

work page 2021

[7] [7]

Underwater acoustic modems,

S. Sendra, J. Lloret, J. M. Jimenez, and L. Parra, “Underwater acoustic modems,” IEEE Sensors Journal , vol. 16, no. 11, pp. 4063–4071, 2016

work page 2016

[8] [8]

Semantic communications: Principles and challenges,

Z. Qin, X. Tao, J. Lu, and G. Y . Li, “Semantic communications: Principles and challenges,” CoRR, vol. abs/2201.01389, 2022. [Online]. Available: https://arxiv.org/abs/2201.01389

arXiv 2022