Importance-Aware Semantic Communication in MIMO-OFDM Systems Using Vision Transformer

Jihun Park; Joohyuk Park; Yongjeong Oh; Yo-Seb Jeon

REVIEW 2 major objections 3 minor 34 references

Importance-Aware Semantic Communication in MIMO-OFDM Systems Using Vision Transformer

T0 review · 2 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Attention scores from a pretrained Vision Transformer can guide joint quantization, subcarrier mapping, and power allocation in MIMO-OFDM semantic communication, and this beats conventional baselines in ideal and finite-blocklength settings

desk verdict Worth a serious referee, but I can only assess the abstract—the full text in my copy is garbled, so the claimed gains need verification. read the letter →

arxiv 2508.07696 v1 pith:GNBQ4RH5 submitted 2025-08-11 eess.SP cs.ITmath.IT

Joohyuk Park , Yongjeong Oh , Jihun Park , Yo-Seb Jeon This is my paper · ORCID

classification eess.SPcs.ITmath.IT

keywords semanticcommunicationMIMO-OFDMvisiontransformerattentionimportance-awareresourceallocationjointquantizationandpowerfiniteblocklengthtransmissionblockcoordinatedescentmulti-viewimageclassification

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

This paper argues that in a MIMO-OFDM semantic communication link, the transmitter can decide how to spend scarce wireless resources by asking a pretrained Vision Transformer which parts of the image matter. The proposed IA-QSMPA framework uses the model's attention scores as a semantic-importance map, then jointly chooses quantization levels, subcarrier assignment, and power allocation so that important features travel over high-quality subchannels. The joint problem is nonconvex, and the paper solves it with block coordinate descent, adding a segment-wise linear approximation of the channel-dispersion penalty to handle finite blocklength transmissions where errors are unavoidable. On a multi-view classification task, this attention-guided resource allocation beats conventional baselines in both ideal and finite-blocklength regimes, improving task accuracy while using the link more efficiently. A sympathetic reader would take the paper's claim to be that semantic importance, extracted for free from a pretrained ViT, can be turned directly into physical-layer resource decisions.

What carries the argument

The load-bearing object is the ViT attention map: the pretrained model's attention scores are used as a semantic-importance prior that ranks feature tokens. This ranking drives a joint resource-optimization problem (quantization level, subcarrier mapping, power allocation) solved by block coordinate descent. In the finite blocklength extension, a segment-wise linear approximation of the channel dispersion penalty keeps the objective amenable to the same block-wise updates. What carries the argument is the transfer from the attention map to a per-token priority that is then matched to subchannel quality.

What would settle it

Take the IA-QSMPA transmitter and compare it against a variant that recomputes attention on the received, channel-degraded features (or re-estimates importance per subchannel) at progressively lower SNR. If the fixed clean-input attention map loses its advantage once channel distortion is large enough to reorder which features are task-critical, the paper's central premise fails. A simpler check: replace the attention map with uniform importance and measure the accuracy/latency gap; if the gap vanishes on high-distortion channels, the importance signal is not doing the claimed work.

Watch

Extended reading notes

Core claim

The central discovery is that attention scores from a pretrained Vision Transformer provide a usable, task-relevant importance map for physical-layer resource allocation in semantic communication. IA-QSMPA couples three decisions that are usually made separately: how many bits each feature gets (quantization), which subcarrier carries it (mapping), and how much power each subcarrier is allocated. It routes high-attention features to subchannels with high effective SNR and allocates more bits and power to them, subject to total power and latency constraints. The optimization is nonconvex, and the paper handles it with block coordinate descent; for the finite blocklength regime, the channel di

Load-bearing premise

The load-bearing premise is that attention scores computed by a pretrained Vision Transformer on clean input remain a reliable measure of which features matter after quantization, subcarrier mapping, and power allocation reshape the transmitted signal; if channel distortion changes which features matter, the importance signal could be miscalibrated.

Editorial extensions

If this is right

If IA-QSMPA is right, MIMO-OFDM semantic links can allocate resources without retraining the transmitter for every channel: a frozen pretrained ViT supplies the importance map.
Resource allocation for semantic communication can be posed as one joint quantization-mapping-power optimization and solved efficiently, rather than tuned heuristically.
Finite blocklength operation is within reach of the same framework using piecewise-linear dispersion approximation, making the approach relevant for low-latency, short-packet wireless.
Task accuracy and communication efficiency improve together, meaning fewer bits over worse channels can suffice when the transmitted features respect semantic importance.
The demonstrated gains on multi-view classification rest on a mechanism that is not tied to that specific task architecture, so the same attention-priority idea can be carried to other transformer-based tasks.

Reading between the lines

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

The paper leaves implicit that the attention map is computed on clean input; I would expect the importance ranking to need re-calibration at very low SNR, where channel distortion may reorder which features are task-critical after reception.
A natural extension is to reuse the same attention-priority machinery for multi-user scheduling, where subcarriers and power are contested across different users' importance maps rather than a single transmitter's map.
A testable extension is ablating the three coupled components: fixing mapping and power and varying only quantization, or fixing quantization and varying mapping, to see which of the three decisions contributes most to the reported gains.
One could compare pretrained attention against a lightweight learned importance scorer trained with the same objective, which would show whether the frozen ViT's attention remains competitive when importance is optimized for the actual channel.

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

Colleague,

Quick take: this is a plausible paper, but I can't judge the details because the full text I have is unreadable—the PDF extraction is garbled beyond recovery. So my comments are based on the abstract plus the visible structure of the manuscript.

What's actually new: the IA-QSMPA framework, which uses attention scores from a pretrained ViT to guide joint quantization, subcarrier mapping, and power allocation in MIMO-OFDM. That's a sensible new application of ViT interpretability to physical-layer resource allocation. The extension to finite blocklength, with a segment-wise linear approximation of the channel dispersion penalty, is a legitimate technical contribution for practical short-packet systems.

What the paper does well: the problem is well motivated—most semantic-communication work stops at source coding; here they take it down to the physical layer. The block-coordinate-descent approach is appropriate for the nonconvex joint optimization, and the MVP-N multi-view task is a reasonable testbed.

Soft spots: the main concern, which you also flagged, is whether clean-input attention scores remain a reliable importance signal after quantization and channel distortion. The abstract doesn't address this. It's a real question, and the paper should show experiments (e.g., comparing with random importance, or with a fine-tuned ViT) to prove the mechanism works. Without full text, I can't tell if they do. The abstract also gives no numerical gains, which is typical but makes the 'significant outperformance' hard to assess.

Also, the full-text corruption is a practical problem for peer review. I'd want a clean version to verify the equations and simulations.

Bottom line: if the simulation results are as reported, this is a solid contribution to semantic communication. Even with my uncertainty, it's not obviously wrong, and it targets a real gap. I'd send it out for review, and I'd insist the authors provide code or at least detailed simulations. I wouldn't desk reject it. A reading group on semantic communication would find it worth discussing, though I wouldn't cite it in my own work right now.

Referee Report

2 major / 3 minor

Summary. The paper proposes an importance-aware quantization, subcarrier mapping, and power allocation (IA-QSMPA) framework for semantic communication in MIMO-OFDM systems. A pretrained Vision Transformer is used to extract attention-based importance scores that guide the joint allocation of quantization levels, subcarriers, and power; a block coordinate descent algorithm solves the nonconvex problem, and a finite-blocklength extension uses a segment-wise linear approximation of the channel dispersion penalty. The abstract reports that simulations on MVP-N multi-view classification show significant improvement over conventional methods. In the version provided to me, the full text is largely unreadable because of character corruption; consequently the derivations, algorithm details, and experimental results cannot be checked.

Significance. The motivation—using semantic importance to allocate physical-layer resources—is timely and, if substantiated, could be a useful contribution to semantic communication. The paper would be strengthened by making the importance mechanism falsifiable, for example through ablations that vary the attention signal and compare against channel-aware importance. No reproducible code, machine-checked proofs, or legible numerical results are available in the submitted manuscript, so the significance remains conditional.

major comments (2)

[Full text (system-model and optimization equations)] The supplied full text is corrupted (mojibake/character substitution), so the equations, algorithm pseudocode, and simulation tables cannot be read. The abstract's central claims—the nonconvex problem formulation, the block coordinate descent algorithm, the segment-wise linear approximation of the channel dispersion penalty, and the reported significant gains—are therefore unverifiable in this version. Please provide a clean source file and ensure all equations and result tables are legible. Without this, no technical verification is possible.
[Abstract] The importance mechanism is not specified sufficiently to assess its validity. The framework uses attention scores from a pretrained ViT, presumably computed on clean input, to set quantization levels, subcarrier mapping, and power allocation. The abstract gives no evidence or argument that these scores remain a reliable ordering of feature importance after quantization, subcarrier assignment, and channel noise, especially at low SNR or short blocklength. If residual errors reorder which features are task-relevant, the optimizer is using miscalibrated weights. Please add an ablation comparing fixed clean-attention importance with channel-aware or iteratively re-estimated importance, and specify whether the ViT is fine-tuned on the MVP-N task. This is a correctness-risk concern about the central mechanism.

minor comments (3)

[Abstract] The phrase 'significantly outperforms' should be supported by quantitative metrics, such as accuracy versus SNR or blocklength, with confidence intervals or error bars.
[Abstract and full text] The extraction of 'attention-based importance' should be defined: which ViT layer/heads are used, how the attention scores are aggregated, and how they map to quantization and power allocation decisions.
[Abstract] The MVP-N dataset should be cited and briefly described, including the number of views, train/test split, and any preprocessing. This is needed to judge the generalization of the claimed gains.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ViT attention scores are a fixed exogenous importance input; resource allocation and task-accuracy results are derived independently.

full rationale

The paper's derivation chain starts from attention-based importance scores extracted from a pretrained ViT, as stated in the abstract: "The proposed framework exploits attention-based importance extracted from a pretrained ViT to jointly optimize quantization levels, subcarrier mapping, and power allocation." These scores enter as fixed weights into a joint optimization over resource variables. The claimed outcome is superior task performance on the MVP-N classification benchmark, not the attention scores themselves. No equation in the visible text equates the optimization output to the attention input, and no parameter is fitted to the evaluation metric and then renamed as a prediction. The finite-blocklength channel-dispersion approximation is an optimization surrogate, not a fitted prediction. The ViT is used as an exogenous feature/importance extractor; its attention weights are not derived from the resource-allocation solution. The possibility that clean-image attention could be miscalibrated under quantization and channel distortion is a correctness risk about the method's assumptions, not a circularity in the derivation. No self-citation chain, uniqueness theorem, or definitional identity is invoked to force the central claim. Therefore the derivation is self-contained with respect to circularity, and the score is 0.

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

The central claim rests on the reliability of ViT attention as a semantic importance measure, the convergence of block coordinate descent, and the accuracy of the channel dispersion approximation. None of these are validated in the abstract alone.

assumptions (3)

ad hoc to paper Attention scores from a pretrained ViT reflect semantic importance for the downstream task.
The IA-QSMPA framework uses these scores to drive quantization, subcarrier mapping, and power allocation; the abstract states this but provides no independent validation of the proxy's reliability under channel distortion.
domain assumption Block coordinate descent converges to a good solution of the nonconvex joint optimization problem.
The abstract says a block coordinate descent algorithm is employed, but convergence guarantees, local optimality, and complexity are not discussed in the abstract.
ad hoc to paper The segment-wise linear approximation of the channel dispersion penalty is accurate enough for joint optimization.
In the finite blocklength extension, the paper introduces this approximation to make the problem tractable; the abstract does not quantify the approximation error.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Importance-Aware Semantic Communication in MIMO-OFDM Systems Using Vision Transformer." pith.science (2026). https://pith.science/paper/GNBQ4RH5

@misc{pith2026250807696,
  author       = {Pith},
  title        = {Pith review of: Importance-Aware Semantic Communication in MIMO-OFDM Systems Using Vision Transformer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GNBQ4RH5}},
  note         = {Machine review of arXiv:2508.07696}
}

read the original abstract

This paper presents a novel importance-aware quantization, subcarrier mapping, and power allocation (IA-QSMPA) framework for semantic communication in multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems, empowered by a pretrained Vision Transformer (ViT). The proposed framework exploits attention-based importance extracted from a pretrained ViT to jointly optimize quantization levels, subcarrier mapping, and power allocation. Specifically, IA-QSMPA maps semantically important features to high-quality subchannels and allocates resources in accordance with their contribution to task performance and communication latency. To efficiently solve the resulting nonconvex optimization problem, a block coordinate descent algorithm is employed. The framework is further extended to operate under finite blocklength transmission, where communication errors may occur. In this setting, a segment-wise linear approximation of the channel dispersion penalty is introduced to enable efficient joint optimization under practical constraints. Simulation results on a multi-view image classification task using the MVP-N dataset demonstrate that IA-QSMPA significantly outperforms conventional methods in both ideal and finite blocklength transmission scenarios, achieving superior task performance and communication efficiency.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages

[1]

Luo, H.-H

X. Luo, H.-H. Chen, and Q. Guo, `` Semantic communications: Overview, open issues, and future research directions ,'' IEEE Wireless Commun., vol. 29, no. 1, pp. 210--219, Feb. 2022

work page 2022
[2]

Gündüz, Z

D. Gündüz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C.-B. Chae, `` Beyond transmitting bits: Context, semantics, and task-oriented communications ,'' IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 5--41, Jan. 2023

work page 2023
[3]

Zhang, H

C. Zhang, H. Zou, S. Lasaulce, W. Saad, M. Kountouris, and M. Bennis, `` Goal-oriented communications for the IoT and application to data compression ,'' IEEE Internet Things Mag., vol. 5, no. 4, pp. 58--63, Dec. 2022

work page 2022
[4]

Erdemir, T.-Y

E. Erdemir, T.-Y. Tung, P. L. Dragotti, and D. Gündüz, `` Generative joint source-channel coding for semantic image transmission ,'' IEEE J. Sel. Areas Commun., vol. 41, no. 8, pp. 2645--2657, Aug. 2023

work page 2023
[5]

D. B. Kurka and D. Gündüz, `` Bandwidth-agile image transmission with deep joint source-channel coding ,'' IEEE Trans. Wireless Commun., vol. 20, no. 12, pp. 8081--8095, Dec. 2021

work page 2021
[6]

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

work page 2021
[7]

L. Yan, Z. Qin, R. Zhang, Y. Li, and G. Y. Li, `` Resource allocation for text semantic communications ,'' IEEE Wireless Commun. Lett., vol. 11, no. 7, pp. 1394--1398, Jul. 2022

work page 2022
[8]

Weng and Z

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

Show all 34 references

[9]

K. Choi, K. Tatwawadi, A. Grover, T. Weissman, and S. Ermon, `` Neural joint source-channel coding ,'' in Proc. Int. Conf. Mach. Learn. (ICML), Long Beach, CA, USA, Jun. 2019, pp. 1182--1192

2019
[10]

Q. Fu, H. Xie, Z. Qin, G. Slabaugh, and X. Tao, `` Vector quantized semantic communication system ,'' IEEE Commun. Lett., vol. 12, no. 6, pp. 982--986, Jun. 2023

2023
[11]

Huang, K

J. Huang, K. Yuan, C. Huang, and K. Huang, `` D ^2 -JSCC: Digital deep joint source-channel coding for semantic communications ,'' IEEE J. Sel. Areas Commun., vol. 43, no. 4, pp. 1246--1261, Apr. 2025

2025
[12]

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

2025
[13]

Y. Oh, J. Park, J. Choi, J. Park, and Y.-S. Jeon, `` Blind training for channel-adaptive digital semantic communications ,'' 2025, arXiv:2501.02273v2

2025 arXiv
[14]

Y. Bo, Y. Duan, S. Shao, and M. Tao, `` Joint coding-modulation for digital semantic communications via variational autoencoder ,'' IEEE Trans. Commun., vol. 72, no. 9, pp. 5626--5640, Sep. 2024

2024
[15]

L. Guo, W. Chen, Y. Sun, and B. Ai, `` Digital-SC: Digital semantic communication with adaptive network split and learned non-linear quantization ,'' IEEE Trans. Cogn. Commun. Netw., early access, Dec. 2024, doi: 10.1109/TCCN.2024.3510586

2024
[16]

Zhang, P

G. Zhang, P. Yang, Y. Cai, Q. Hu, and G. Yu, `` From analog to digital: Multi-order digital joint coding-modulation for semantic communication ,'' IEEE Trans. Commun., early access, Dec. 5, 2024, doi: 10.1109/TCOMM.2024.3511949

2024
[17]

H. Wu, Y. Shao, C. Bian, K. Mikolajczyk, and D. Gündüz, `` Deep joint source-channel coding for adaptive image transmission over MIMO channels ,'' IEEE Trans. Wireless Commun., vol. 23, no. 10, pp. 15\,002--15\,017, Oct. 2024

2024
[18]

M. Yang, C. Bian, and H.-S. Kim, `` OFDM-guided deep joint source channel coding for wireless multipath fading channels ,'' IEEE Trans. Cogn. Commun. Netw., vol. 8, no. 2, pp. 584--599, Jun. 2022

2022
[19]

J. Park, H. Kim, J. Shin, Y. Oh, and Y.-S. Jeon, `` End-to-end training and adaptive transmission for OFDM-based semantic communication ,'' 2025, to be appeared in ICT Express

2025
[20]

H. Xie, Z. Qin, and G. Y. Li, `` Task-oriented multi-user semantic communications for VQA ,'' IEEE Wireless Commun. Lett., vol. 11, no. 3, pp. 553--557, Mar. 2022

2022
[21]

J. Shao, Y. Mao, and J. Zhang, `` Task-oriented communication for multidevice cooperative edge inference ,'' IEEE Trans. Wireless Commun., vol. 22, no. 1, pp. 73--87, Jan. 2023

2023
[22]

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, Mar. 2023

2023
[23]

L. Teng, W. An, C. Dong, and X. Xu, `` sDMCM—semantic digital modulation constellation mapping scheme for semantic communication ,'' IEEE Internet Things J., early access, Feb. 2025, doi: 10.1109/JIOT.2025.3545667

2025
[24]

H. Gao, G. Yu, Y. He, and Y. Liu, `` Semantic feature scheduling and rate control in multi-modal distributed network ,'' IEEE Trans. Wireless Commun., vol. 23, no. 12, pp. 19\,199--19\,214, Dec. 2024

2024
[25]

K. Zhou, G. Zhang, Y. Cai, Q. Hu, G. Yu, and A. L. Swindlehurst, `` Feature allocation for semantic communication with space-time importance awareness ,'' IEEE Trans. Wireless Commun., early access, May 2025, doi: 10.1109/TWC.2025.3569320

2025
[26]

J. Im, N. Kwon, T. Park, J. Woo, J. Lee, and Y. Kim, `` Attention-aware semantic communications for collaborative inference ,'' IEEE Internet Things J., vol. 11, no. 22, pp. 37\,008--37\,020, Nov. 2024

2024
[27]

J. Park, Y. Oh, Y. Kim, and Y.-S. Jeon, `` Vision transformer-based semantic communications with importance-aware quantization ,'' 2024, arXiv:2412.06038

2024 arXiv
[28]

Peng and R

L. Peng and R. Vidal, `` Block coordinate descent on smooth manifolds: Convergence theory and twenty-one examples ,'' 2023, arXiv:2305.14744

2023 arXiv
[29]

R. Wang, T. S. Kim, J.-S. Kim, and H.-J. Lee, `` Towards real-world multi-view object classification: dataset, benchmark, and analysis ,'' IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 7, pp. 5653--5664, Jul. 2024

2024
[30]

Goldsmith, Wireless communications

A. Goldsmith, Wireless communications . 1em plus 0.5em minus 0.4em Cambridge, U.K.: Cambridge Univ. Press, 2005

2005
[31]

X. Ling, B. Wu, P.-H. Ho, F. Luo, and L. Pan, `` Fast water-filling for agile power allocation in multi-channel wireless communications ,'' IEEE Commun. Lett., vol. 16, no. 8, pp. 1212--1215, Aug. 2012

2012
[32]

F. Gao, T. Cui, and A. Nallanathan, `` Optimal training design for channel estimation in decode-and-forward relay networks with individual and total power constraints ,'' IEEE Trans. Signal Process., vol. 56, no. 12, pp. 5937--5949, Dec. 2008

2008
[33]

Z. Ma, M. Xiao, Y. Xiao, Z. Pang, H. V. Poor, and B. Vucetic, `` High-reliability and low-latency wireless communication for internet of things: Challenges, fundamentals, and enabling technologies ,'' IEEE Internet Things J., vol. 6, no. 5, pp. 7946--7970, Oct. 2019

2019
[34]

Touvron, M

H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. Jégou, `` Training data-efficient image transformers & distillation through attention ,'' in Proc. Int. Conf. Mach. Learn. (ICML), Jul. 2021, pp. 10\,347--10\,357

2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Luo, H.-H

X. Luo, H.-H. Chen, and Q. Guo, `` Semantic communications: Overview, open issues, and future research directions ,'' IEEE Wireless Commun., vol. 29, no. 1, pp. 210--219, Feb. 2022

work page 2022

[2] [2]

Gündüz, Z

D. Gündüz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C.-B. Chae, `` Beyond transmitting bits: Context, semantics, and task-oriented communications ,'' IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 5--41, Jan. 2023

work page 2023

[3] [3]

Zhang, H

C. Zhang, H. Zou, S. Lasaulce, W. Saad, M. Kountouris, and M. Bennis, `` Goal-oriented communications for the IoT and application to data compression ,'' IEEE Internet Things Mag., vol. 5, no. 4, pp. 58--63, Dec. 2022

work page 2022

[4] [4]

Erdemir, T.-Y

E. Erdemir, T.-Y. Tung, P. L. Dragotti, and D. Gündüz, `` Generative joint source-channel coding for semantic image transmission ,'' IEEE J. Sel. Areas Commun., vol. 41, no. 8, pp. 2645--2657, Aug. 2023

work page 2023

[5] [5]

D. B. Kurka and D. Gündüz, `` Bandwidth-agile image transmission with deep joint source-channel coding ,'' IEEE Trans. Wireless Commun., vol. 20, no. 12, pp. 8081--8095, Dec. 2021

work page 2021

[6] [6]

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

work page 2021

[7] [7]

L. Yan, Z. Qin, R. Zhang, Y. Li, and G. Y. Li, `` Resource allocation for text semantic communications ,'' IEEE Wireless Commun. Lett., vol. 11, no. 7, pp. 1394--1398, Jul. 2022

work page 2022

[8] [8]

Weng and Z

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