Pith. sign in

REVIEW 4 major objections 4 minor 57 references

LiTCom: A Lightweight Transmitter and Inference-Capable Receiver Framework for 6G Uplink

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read LiTCom shows that a transmitter can low-pass filter images and skip heavy channel coding, trusting a generative AI receiver to infer lost detail, and beat conventional and learned baselines at low SNR.

desk verdict LiTCom's core idea is worth a look, but the headline SNR gains over Deep-JSCC are inflated by an unequal bandwidth comparison; still deserves serious peer review. read the letter →

arxiv 2607.13114 v1 pith:JIOUHULI submitted 2026-07-14 eess.SP

Chunmei Xu , Siqi Zhang , Zhi Ding , Yi Ma , Rahim Tafazolli This is my paper · ORCID
classification eess.SP
keywords 6GuplinksemanticcommunicationgenerativeAIreceiverlightweightsourcecodingerror-distributingchannelqualityofexperienceimportance-awarepowerallocationlow-SNRtransmission
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

LiTCom asks what a 6G uplink looks like when the edge device is intentionally dumb and the base station is very smart. The paper argues that for image transmission, the transmitter need only keep the low-frequency content—object structure and layout—using a mean-filter downsampler, and can transmit with weak or no channel coding, because a large pre-trained generative model at the receiver can hallucinate the missing high-frequency detail and tolerate scattered bit errors. The design is justified by two properties: representing sufficiency (the transmitted representation must preserve enough semantic essence) and error robustness (the receiver can absorb independent, unstructured residual errors far better than structured bursts). This yields up to 8 dB SNR gain over a 5G NR-like JPEG+LDPC link and 2.5 dB over the Deep-JSCC baseline, while cutting transmitter computations by more than 95%. A reader should care because it reframes the uplink bottleneck as a compute asymmetry problem, not a coding-rate problem.

What carries the argument

The load-bearing mechanism is the pairing of a semantic-preserving source code with an error-distributing channel code. The source code is a block-wise low-pass (mean) filter that discards high-frequency detail and compresses the image by a factor r, costing only one multiplication and block-size additions per block. The channel code is deliberately weak or absent, chosen to make residual errors approximately independent and unstructured at low SNR rather than minimising BER. On top of this, an importance-aware water-filling power allocation assigns more power to more significant bit planes (weights gamma_k = 2^{2(k-1)}), minimizing an importance-weighted MSE surrogate of QoE. The receiver-s

What would settle it

Run a controlled comparison at fixed normalized MSE between (a) uncoded transmission producing i.i.d. bit errors and (b) an LDPC or convolutional code at the same residual MSE but with structured post-decoding burst errors; if the generative reconstructor scores no worse on NIQE/CLIP under structured errors, Assumption 2 and the motivation for error-distributing codes are refuted. A second, simpler check: measure the actual residual error distribution of the uncoded scheme under bursty fading or imperfect CSI and test whether it still resembles the independent-noise regime the GenAI model was

Watch

Extended reading notes

Core claim

The central discovery is that channel coding should not be judged by bit-error rate alone in a semantic-communication link. At low SNR, strong codes (LDPC, Turbo) decode poorly and leave structured, correlated residual errors that a generative decoder cannot fix, whereas uncoded or weakly coded transmission produces scattered, nearly independent errors that the generative prior can denoise. The paper formalizes this as an 'error-distributing code' designed to shape the residual error distribution toward the perturbation class that maximizes reconstructed QoE. On the source side, a simple block low-pass filter preserves the low-frequency semantic essentials, and the receiver's pre-trained dif

Load-bearing premise

The framework stands or falls on Assumption 2: at the same mean-squared error, independent random errors degrade reconstructed semantic quality less than structured or correlated errors; if that ordering reverses, the entire 'transmit uncoded and let the GenAI receiver fix it' design loses its justification.

Editorial extensions

If this is right

  • If LiTCom is right, 6G uplink transmitters can be drastically simplified: a low-pass filter plus uncoded or weak-FEC transmission suffices for image services, cutting transmitter-side computation by over 95%.
  • Strong FEC codes may be actively harmful at low SNR for semantic tasks, because post-decoding error bursts are structured and harder for generative models to repair; code selection should be SNR- and decoder-aware.
  • Perceived uplink coverage—the region of (compression rate, SNR) pairs meeting QoE thresholds—can be extended by several dB and by compression-rate margin compared to bit-fidelity-based coverage.
  • Importance-aware power allocation across bit planes yields closed-form water-filling solutions and consistently improves QoE over equal-power allocation in the paper's simulations.
  • The framework is portable to any modality with a mature generative restoration model (audio, video) and to any generative decoder, with operating points shifting but the architecture principle unchanged.

Reading between the lines

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

  • The paper's Assumption 2 suggests a reframing of channel-code design: instead of maximizing post-decoding BER gains, codes should be designed to maximize the entropy or independence of residual errors, potentially leading to new 'generative-friendly' code families—an avenue the paper only gestures at.
  • Because the receiver's generative model is trained on natural image degradations, the approach may transfer to audio or video but will likely fail on modalities lacking strong generative priors, such as radar point clouds or haptic signals; this implies that semantic-communication gains are bounded by the quality of the prior, not by channel capacity.
  • If the 8 dB SNR gain holds under fading with imperfect CSI, the effective coverage area of a base station could grow substantially, since SNR gains translate into range gains; this is a deployment-level consequence the paper quantifies only via effective-SNR arguments.
  • A testable extension is to adapt the receiver generative model to channel-error-contaminated representations (fine-tuning on structured error patterns), which the paper lists as future work but which could close the gap when burst errors are unavoidable.
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.

Referee Report

4 major / 4 minor

Summary. The paper proposes LiTCom, an uplink image-transmission framework that deliberately offloads almost all processing to the receiver: the transmitter applies block-wise low-pass filtering at compression rate r, partitions the compressed pixels into bit-importance sub-streams, optionally applies weak FEC or no channel coding, and modulates with M-QAM; the receiver runs the SUPIR generative restoration model to infer a high-quality reconstruction. A NIQE/CLIP-based QoE metric is introduced, perceived uplink coverage is defined, and importance-aware water-filling power allocation is derived and simulated. The central claims are SNR gains up to 8 dB over a JPEG+LDPC 5G NR-like baseline and 2.5 dB over Deep-JSCC, with a transmitter-side computation reduction of over 95%.

Significance. The architectural concept is timely and potentially important for 6G uplink scenarios with resource-constrained transmitters and infrastructure-side AI accelerators: move source and channel coding complexity to the receiver-side generative decoder. The paper is explicit about its design principles, gives closed-form algorithms, and validates on a concrete system model, which is a strength. However, the headline numerical claims are not yet established because the Deep-JSCC comparison is made at equal nominal 'r' but unequal channel resources, and the Deep-JSCC baseline is trained on a different distribution. If these comparison issues are corrected, the framework could be a useful design point; the current manuscript therefore requires substantial revision rather than acceptance.

major comments (4)
  1. [Sec. II.B.1, Sec. III, Sec. VI.C, Sec. VI.D.2] The SNR-gain comparison with Deep-JSCC is not apples-to-apples. In Sec. II.B.1, r is defined as the spatial subsampling ratio (S=rI). With 8-bit pixels and 4-QAM, the LiTCom transmitter in Sec. III sends 8r/log2(M)=4r channel symbols per source pixel, whereas Deep-JSCC's r is the channel bandwidth ratio k/n, i.e., r channel uses per source pixel. Thus at r=9% LiTCom uses 0.36 channel uses/pixel versus 0.09 for Deep-JSCC—a factor of 4. The paper's own resource-occupancy estimates in Sec. VI.D.2 (1500 ms for LiTCom vs 375 ms for Deep-JSCC) confirm this factor. Consequently the reported 2–2.5 dB SNR gains in Figs. 7 and 9 may be a bandwidth advantage rather than evidence for the generative-inference design. The comparison should be redone at equal spectral efficiency, e.g., equal channel uses per source pixel, or the coverage plots should explicitly trade SNR against bandwidth.
  2. [Sec. VI.A, Sec. VI.B, Fig. 11] The Deep-JSCC baseline is trained on CIFAR10 at Eb/No=0 and evaluated on 640×512 images at low SNR. This is stated in Sec. VI.B, but its consequence is not addressed: Deep-JSCC is out-of-distribution in both resolution and SNR regime, which can only exaggerate the gap. The test image set is never identified, and no error bars are shown on any QoE curve. Please compare with a Deep-JSCC model that is retrained or at least evaluated on the same test resolution/SNR distribution, and report the image set, number of runs, and variability (especially given the stochasticity of the SUPIR decoder).
  3. [Eqs. (17), (34), (37)] The power-allocation formulas contain sign/algebra errors as printed. In Eq. (17), α_c>0 and β_c>0 give a BER that grows exponentially with SNR, which is nonsensical; the usual model would be exp(-β_c snr_k). In Eq. (34), the second logarithm appears to have √f_snr in the numerator, whereas substituting (33) into (32) places it in the denominator. In Eq. (37), the argument -α_c β_c/λ* is negative for λ*>0 and α_c,β_c>0, so the logarithm is undefined. These issues propagate into Algorithms 1 and 2. Please correct the equations and verify the corresponding simulation code, or the analytical development cannot be trusted.
  4. [Sec. II.A.2, Fig. 6] Assumption 2—that at the same MSE, independent/unstructured residual errors degrade semantic QoE less than structured/correlated errors—is the cornerstone of the weak-FEC/uncoded design. The current support is one illustrative error-distribution figure and a qualitative statement, and the conclusion itself acknowledges that burst errors, structured decoding failures, etc. can break the scheme. Please provide a quantitative experiment that varies error structure (e.g., correlation length or block-error size) at a fixed MSE and measures NIQE/CLIP, to give the reader confidence that the design principle is not just a single-image anecdote.
minor comments (4)
  1. [Sec. VI.B] Figures 10 and 11 are referenced before Fig. 4 in the text; the figure numbering/order should be fixed.
  2. [Table I] To support the '>95% transmitter-side computation' claim, the Deep-JSCC row should report FLOPs or multiply-adds in the same units as the other rows, not just parameter count.
  3. [Sec. VI.A] The definition of r for the JPEG baseline should be clarified: is it bits-per-pixel after JPEG compression, or a normalized quality factor? Without this, the coverage plots in Fig. 9 are hard to interpret.
  4. [Eq. (18)] The 'at most one bit per pixel' assumption should be stated as an approximation with a validity range; at low SNR where BER is not small, multi-bit errors are not negligible and the IMSE surrogate may be inaccurate.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline SNR gains are generated by external models (SUPIR, NIQE, CLIP) and textbook/standard BER equations, not by self-referential fitting.

full rationale

LiTCom's claimed SNR gains are produced by end-to-end simulations in Sec. VI in which the receiver is the externally pre-trained SUPIR [35] and the QoE scores are computed with the externally defined NIQE [51] and CLIP [52] models; those scores are not generated by the paper's own equations. The source-coding step is an elementary LPF operation (Sec. II.B.1, Eq. 8) and the channel-uncoded BER is the textbook Q-function (Eq. 16), so the headline QoE curves are not a restatement of an assumed input. Assumptions 1 and 2 in Sec. II.A are explicitly assumptions supported by Figs. 5-6, not derived predictions, and the paper itself concedes in Sec. VI.A and the Conclusion that the regime is conditional ('LiTCom is therefore robust, conditioning on the residual errors remaining within this training degradation regime of the GenAI model'). The reuse of [27] for the weak-FEC BER approximation (Eq. 17) and IMSE surrogate (Eq. 18) is a dependency on the authors' prior fitted model; it does not constitute a 'prediction' of this paper, and the main GenAI-Uncoded comparisons do not rely on it. The reviewer's concern that the compression rate r denotes different quantities for LiTCom and Deep-JSCC (spatial subsampling vs. channel bandwidth ratio) is a benchmark-fairness issue, not a circular reduction: the compared numbers may be inequivalent in spectral efficiency, but they are not equivalent by construction. No load-bearing step collapses into its own input.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests on domain assumptions about how a generative prior behaves under channel noise, plus a few chosen or fitted parameters (QoE thresholds, BER-model coefficients, bit-importance weights). No new physical entities are introduced.

free parameters (3)
  • QoE thresholds Dth_NIQE and Dth_CLIP = [5, 0.1]
    Chosen in Sec. VI.A to define 'satisfactory' quality; all coverage gains and required-SNR numbers depend on this arbitrary threshold.
  • Weak-FEC BER model coefficients alpha_c, beta_c = not given (from Ref. [27])
    Eq. (17) approximates BER as alpha_c * exp(beta_c * snr); the paper states these are obtained via data fitting, so the channel-coded power allocation is built on fitted constants.
  • Bit-importance weights gamma_k = 2^{2(k-1)}
    Assigned in Sec. III as bit-position significance; drives the water-filling allocation but is not derived or validated against human perception.
assumptions (6)
  • domain assumption Assumption 1: semantic QoE degradation is positively related to MSE of the received representation
    Sec. II.A. Used to relax an intractable error-robustness set to an MSE-threshold condition and to justify IMSE as the optimization objective.
  • domain assumption Assumption 2: at equal MSE, independent/unstructured errors cause less QoE degradation than structured/correlated errors
    Sec. II.A. This is the reason uncoded and weak-FEC schemes are preferred over strong codes at low SNR.
  • domain assumption SUPIR's generative prior covers the degradation from LPF compression plus low-SNR channel errors
    Sec. VI.A states robustness is conditioned on residual errors remaining within the training degradation regime; burst errors, high mobility, structured decoding failures are explicitly excluded.
  • domain assumption At most one bit error per pixel in the source representation
    Used in Eq. (18) to approximate IMSE as sum gamma_k * ber_k; at low SNR with BER ~0.1 or higher, multiple bit errors per pixel are likely, making the approximation questionable.
  • domain assumption AWGN/parallel sub-channel model approximates fading and CSI errors as effective SNR reduction
    Sec. VI.A justifies evaluation under AWGN; this avoids simulating time/frequency-selective fading and imperfect CSI.
  • domain assumption NIQE and CLIP jointly capture human semantic QoE
    The entire QoE definition and coverage concept depend on treating NIQE<=5 and CLIP distance<=0.1 as satisfactory; no user study is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LiTCom: A Lightweight Transmitter and Inference-Capable Receiver Framework for 6G Uplink." pith.science (2026). https://pith.science/paper/JIOUHULI

@misc{pith2026260713114,
  author       = {Pith},
  title        = {Pith review of: LiTCom: A Lightweight Transmitter and Inference-Capable Receiver Framework for 6G Uplink},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JIOUHULI}},
  note         = {Machine review of arXiv:2607.13114}
}
read the original abstract

This paper introduces LiTCom, a lightweight transmitter and inference-capable receiver framework, designed to enable robust 6G uplink communication under low signal-to-noise (SNR) conditions. It embraces the resource asymmetry between edge devices and the network infrastructure. LiTCom simplifies transmitter design by applying basic low-pass filtering for source coding and minimal channel coding, significantly reducing the processing complexity. The receiver employs large-scale generative artificial intelligence (GenAI) models to infer high semantic-fidelity content from highly distorted and degraded signals beyond traditional decoding capabilities. Furthermore, efficient power allocation strategies are developed by exploiting data importance to improve system performance, which is measured by the introduced quality of experience (QoE) metric. Simulation results validate the effectiveness of the proposed LiTCom framework and the lightweight coding design. Compared with the 5G NR-like baseline (using JPEG source coding and LDPC channel coding) and the Deep-JSCC baseline, LiTCom achieves SNR gains up to 8 dB and 2.5 dB, respectively, while reducing over 95% transmitter-side computations.

Figures

Figures reproduced from arXiv: 2607.13114 by the authors.

Figure 1
Figure 1. Illustrative diagram of sufficient and insufficient representations, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. System model of the proposed LiTCom framework for robust 6G uplink transmission. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The comparison between the perceived coverage and conventional [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Representing sufficiency of LiTCom under perfect transmission. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 6
Figure 6. Figure 6: Error distribution across the source representations at a compression [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The QoE Performance comparison: (a) DNIQE at a compression rate of r = 2%, (b) DNIQE at a compression rate of r = 4%, (c) DNIQE at a compression rate of r = 9%, (d) DCLIP at a compression rate of r = 2%, (e) DCLIP at a compression rate of r = 4%, (c) DCLIP at a compres…
Figure 8
Figure 8. Figure 8: Performance comparison in terms of normalized MSE under the [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Perceived coverage of LiTCom in comparison with traditional wireless [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: The original source, compressed representation, and the reconstructions under the proposed LiTCom framework with perfect transmission. The first [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: The reconstructions under LiTCom, 5G NR-like systems, and Deep-JSCC where [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Visualization comparison between the proposed GenAI-Uncoded-WF and the Deep-JSCC baseline at low SNRs at a compression rate of [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 2 linked inside Pith

  1. [1]

    Toward 6G networks: Use cases and technologies,

    M. Giordani, M. Polese, M. Mezzavilla, S. Rangan, and M. Zorzi, “Toward 6G networks: Use cases and technologies,”IEEE Commun. Mag., vol. 58, no. 3, pp. 55–61, 2020

    2020

  2. [2]

    On the road to 6G: Visions, requirements, key technologies, and testbeds,

    C.-X. Wang, X. You, X. Gao, X. Zhu, Z. Li, C. Zhang, H. Wang, Y . Huang, Y . Chen, H. Haaset al., “On the road to 6G: Visions, requirements, key technologies, and testbeds,”IEEE Commun. Surv. Tutor., vol. 25, no. 2, pp. 905–974, 2023

    2023

  3. [3]

    Semantic communications for future internet: Fundamentals, applications, and challenges,

    W. Yang, H. Du, Z. Q. Liew, W. Y . B. Lim, Z. Xiong, D. Niyato, X. Chi, X. Shen, and C. Miao, “Semantic communications for future internet: Fundamentals, applications, and challenges,”IEEE Commun. Sur. Tutor., vol. 25, no. 1, pp. 213–250, 2022

    2022

  4. [4]

    5G; NR; nr and ng-ran overall description (release 19),

    3rd Generation Partnership Project (3GPP), “5G; NR; nr and ng-ran overall description (release 19),” 3GPP, Technical Specification (TS) 38.300, 2025

    2025

  5. [5]

    AI-on-RAN: Enabling monetizable differ- entiated connectivity for AI,

    A.-R. Alliance, “AI-on-RAN: Enabling monetizable differ- entiated connectivity for AI,” AI-RAN Alliance, White Paper, 2026. [Online]. Available: https://ai-ran.org/documents/ AI-RAN-WG3-AI-on-RAN-Whitepaper.pdf

    2026

  6. [6]

    The JPEG still picture compression standard,

    G. K. Wallace, “The JPEG still picture compression standard,”IEEE Trans. Consumer Electronics, vol. 38, no. 1, pp. xviii–xxxiv, 1992

    1992

  7. [7]

    Overview of the h. 264/avc video coding standard,

    T. Wiegand, G. J. Sullivan, G. Bjontegaard, and A. Luthra, “Overview of the h. 264/avc video coding standard,”IEEE Trans. Circuits Syst. Video Technol., vol. 13, no. 7, pp. 560–576, 2003

    2003

  8. [8]

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

    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, seman- tics, and task-oriented communications,”IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 5–41, 2022

    2022

Show all 57 references
  1. [9]

    Error detecting and error correcting codes,

    R. W. Hamming, “Error detecting and error correcting codes,”The Bell System Tech. J., vol. 29, no. 2, pp. 147–160, 1950

    1950

  2. [10]

    Convolutional codes and their performance in communica- tion systems,

    A. Viterbi, “Convolutional codes and their performance in communica- tion systems,”IEEE Trans. Commun. Tech., vol. 19, no. 5, pp. 751–772, 1971

    1971

  3. [11]

    Near shannon limit error- correcting coding and decoding: Turbo-codes. 1,

    C. Berrou, A. Glavieux, and P. Thitimajshima, “Near shannon limit error- correcting coding and decoding: Turbo-codes. 1,” inIEEE Int. Conf. Commun., vol. 2. IEEE, 1993, pp. 1064–1070

    1993

  4. [12]

    Design of capacity-approaching irregular low-density parity-check codes,

    T. J. Richardson, M. A. Shokrollahi, and R. L. Urbanke, “Design of capacity-approaching irregular low-density parity-check codes,”IEEE Trans. Inf. theory, vol. 47, no. 2, pp. 619–637, 2001

    2001

  5. [13]

    Semantics-empowered communication for networked intelligent systems,

    M. Kountouris and N. Pappas, “Semantics-empowered communication for networked intelligent systems,”IEEE Commun. Mag., vol. 59, no. 6, pp. 96–102, 2021

    2021

  6. [14]

    A theory of semantic communication,

    Y . Shao, Q. Cao, and D. Gündüz, “A theory of semantic communication,” IEEE Trans. Mobile Computing, vol. 23, no. 12, pp. 12 211–12 228, 2024

    2024

  7. [15]

    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, 2021

    2021

  8. [16]

    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, 2021

    2021

  9. [17]

    Deep joint source- channel coding for wireless image transmission,

    E. Bourtsoulatze, D. B. Kurka, and D. Gündüz, “Deep joint source- channel coding for wireless image transmission,”IEEE Trans. Cognitive Commun. Networking, vol. 5, no. 3, pp. 567–579, 2019

    2019

  10. [18]

    Deep joint source-channel coding for semantic communications,

    J. Xu, T.-Y . Tung, B. Ai, W. Chen, Y . Sun, and D. Gündüz, “Deep joint source-channel coding for semantic communications,”IEEE Commun. Mag., vol. 61, no. 11, pp. 42–48, 2023

    2023

  11. [19]

    Generative joint source-channel coding for semantic image transmission,

    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., 2023

    2023

  12. [20]

    Resitok: A resilient tokenization- enabled framework for ultra-low-rate and robust image transmission,

    Z. Liu, Y . Ma, and R. Tafazolli, “Resitok: A resilient tokenization- enabled framework for ultra-low-rate and robust image transmission,” arXiv preprint arXiv:2505.01870, 2025

    2025 arXiv

  13. [21]

    OFDM-based digital semantic communication with importance awareness,

    C. Liu, C. Guo, Y . Yang, W. Ni, and T. Q. Quek, “OFDM-based digital semantic communication with importance awareness,”IEEE Trans. Commun., vol. 72, no. 10, pp. 6301–6315, 2024

    2024

  14. [22]

    From analog to digital semantic communications: Architectures, challenges, and future directions,

    J. Zou, W. Xie, J. Xiao, Y . Liang, J. M. Moualeu, and L. Yang, “From analog to digital semantic communications: Architectures, challenges, and future directions,”IEEE Wireless Commun., vol. 32, no. 5, pp. 56– 62, 2025

    2025

  15. [23]

    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. S. Areas Commun., vol. 41, no. 1, pp. 230–244, 2022

    2022

  16. [24]

    A survey on semantic communication networks: Architecture, security, and privacy,

    S. Guo, Y . Wang, N. Zhang, Z. Su, T. H. Luan, Z. Tian, and X. Shen, “A survey on semantic communication networks: Architecture, security, and privacy,”IEEE Commun. Surv. Tutor., vol. 27, no. 5, pp. 2860–2894, 2024

    2024

  17. [25]

    Gen- erative semantic communications with foundation models: Perception- error analysis and semantic-aware power allocation,

    C. Xu, M. B. Mashhadi, Y . Ma, R. Tafazolli, and J. Wang, “Gen- erative semantic communications with foundation models: Perception- error analysis and semantic-aware power allocation,”IEEE J. S. Areas Commun., 2025

    2025

  18. [26]

    Latency-aware generative semantic communications with pre-trained diffusion models,

    L. Qiao, M. B. Mashhadi, Z. Gao, C. H. Foh, P. Xiao, and M. Bennis, “Latency-aware generative semantic communications with pre-trained diffusion models,”IEEE Wireless Commun. Lett., 2024

    2024

  19. [27]

    Data-importance-aware power allocation for adaptive real-time communication in computer vision applications,

    C. Xu, Y . Ma, R. Tafazolli, and J. Wang, “Data-importance-aware power allocation for adaptive real-time communication in computer vision applications,”IEEE J. S. Areas Commun., vol. 43, no. 12, pp. 4015– 4026, 2025

    2025

  20. [28]

    AI-native 6G empowering intelligent RAN with accelerated compute,

    X. Lin, L. Kundu, S. Cammerer, Y . Huang, C. Dick, C. Santhosam, R. Gadiyar, R. Wiesmayr, and C. Studer, “AI-native 6G empowering intelligent RAN with accelerated compute,”IEEE Wireless Commun., vol. 32, no. 6, pp. 11–14, 2025

    2025

  21. [29]

    Big AI models for 6G wireless networks: Opportunities, challenges, and research directions,

    Z. Chen, Z. Zhang, and Z. Yang, “Big AI models for 6G wireless networks: Opportunities, challenges, and research directions,”IEEE 13 Fig. 10. The original source, compressed representation, and the reconstructions under the proposed LiTCom framework with perfect transmission. ...

    2024

  22. [30]

    Large generative AI models for telecom: The next big thing?

    L. Bariah, Q. Zhao, H. Zou, Y . Tian, F. Bader, and M. Debbah, “Large generative AI models for telecom: The next big thing?”IEEE Commun. Mag., vol. 62, no. 11, pp. 84–90, 2024

    2024

  23. [31]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inProc. Int. conf. machine learning (ICML), Virtual, Jul. 2021, pp. 8748–8763

    2021

  24. [32]

    Language mod- els are few-shot learners,

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askellet al., “Language mod- els are few-shot learners,”Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

    1901

  25. [33]

    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,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., New Orleans, Louisiana, Jun. 2022, pp. 10 684–10 695

    2022

  26. [34]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” inProc. IEEE/CVF Int. Conf. Computer Vision (ICCV), Paris, France, Oct. 2023, pp. 3836–3847

    2023

  27. [35]

    Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild,

    F. Yu, J. Gu, Z. Li, J. Hu, X. Kong, X. Wang, J. He, Y . Qiao, and C. Dong, “Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild,” 2024

    2024

  28. [36]

    Diffbir: Toward blind image restoration with generative diffusion prior,

    X. Lin, J. He, Z. Chen, Z. Lyu, B. Dai, F. Yu, Y . Qiao, W. Ouyang, and C. Dong, “Diffbir: Toward blind image restoration with generative diffusion prior,” inProc. Eur. Conf. Comput. Vis.Springer, 2024, pp. 430–448

    2024

  29. [37]

    Semantic image inpainting with deep generative models,

    R. A. Yeh, C. Chen, T. Yian Lim, A. G. Schwing, M. Hasegawa-Johnson, and M. N. Do, “Semantic image inpainting with deep generative models,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2017, pp. 5485–5493

    2017

  30. [38]

    Generative im- age inpainting with contextual attention,

    J. Yu, Z. Lin, J. Yang, X. Shen, X. Lu, and T. S. Huang, “Generative im- age inpainting with contextual attention,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2018, pp. 5505–5514

    2018

  31. [39]

    Tech- niques for measuring quality of experience,

    F. Kuipers, R. Kooij, D. De Vleeschauwer, and K. Brunnström, “Tech- niques for measuring quality of experience,” inWired/Wireless Internet Communications. Berlin, Heidelberg: Springer Berlin Heidelberg, 2010, pp. 216–227

    2010

  32. [40]

    Towards a QoE model to evaluate holographic augmented reality devices,

    L. Zhang, H. Dong, and A. El Saddik, “Towards a QoE model to evaluate holographic augmented reality devices,”IEEE MultiMedia, vol. 26, no. 2, pp. 21–32, 2018

    2018

  33. [41]

    QoE-driven data communication framework for consumer electronics in tele-healthcare system,

    T. Zhang, X. Zhou, J. Liu, B. Cheng, X. Xu, L. Qi, Q. Tian, and Z. Wan, “QoE-driven data communication framework for consumer electronics in tele-healthcare system,”IEEE Trans. Consumer Electronics, vol. 69, no. 4, pp. 719–733, 2023

    2023

  34. [42]

    QoE in video transmission: A user experience-driven strategy,

    T. Zhao, Q. Liu, and C. W. Chen, “QoE in video transmission: A user experience-driven strategy,”IEEE Commun. Surv. Tutor., vol. 19, no. 1, pp. 285–302, 2016

    2016

  35. [43]

    QoE-oriented mobile virtual reality game in distributed edge networks,

    Y . Zhang, L. Pu, T. Lin, and J. Yan, “QoE-oriented mobile virtual reality game in distributed edge networks,”IEEE Trans. Multimedia, vol. 25, pp. 9132–9146, 2023

    2023

  36. [44]

    Survey on QoE\QoS correlation models for multimedia services,

    M. Alreshoodi and J. Woods, “Survey on QoE\QoS correlation models for multimedia services,”Int. J. Distrib. Parallel Syst., vol. 4, no. 3, p. 53, 2013

    2013

  37. [45]

    Ber comparison between convolu- tional, turbo, ldpc, and polar codes,

    B. Tahir, S. Schwarz, and M. Rupp, “Ber comparison between convolu- tional, turbo, ldpc, and polar codes,” inInt. Conf. Telecommunications (ICT), 2017, pp. 1–7

    2017

  38. [46]

    To code, or not to code: Lossy source-channel communication revisited,

    M. Gastpar, B. Rimoldi, and M. Vetterli, “To code, or not to code: Lossy source-channel communication revisited,”IEEE Trans. Inf. Theory, vol. 49, no. 5, pp. 1147–1158, 2003

    2003

  39. [47]

    Blind audio bandwidth extension: A diffusion-based zero-shot approach,

    E. Moliner, F. Elvander, and V . Välimäki, “Blind audio bandwidth extension: A diffusion-based zero-shot approach,”IEEE/ACM Trans. Audio, Speech, Language Process., vol. 32, pp. 5092–5105, 2024

    2024

  40. [48]

    Diffvsr: Revealing an effective recipe for taming robust video super-resolution against complex degradations,

    X. Li, Y . Liu, S. Cao, Z. Chen, S. Zhuang, X. Chen, Y . He, Y . Wang, and Y . Qiao, “Diffvsr: Revealing an effective recipe for taming robust video super-resolution against complex degradations,” inProc. IEEE Int. Conf. Computer Vision (ICCV), Hawaii, US, Oct. 2025, pp. 15 319–15 328

    2025

  41. [49]

    Structured IB: Improving information bottleneck with structured feature learning,

    H. Yang, Y . Wu, D. Wen, Y . Zhou, and Y . Shi, “Structured IB: Improving information bottleneck with structured feature learning,” inProc. AAAI Conf. Artif. Intell, vol. 39, no. 20, 2025, pp. 21 922–21 928

    2025

  42. [50]

    Goldsmith,Wireless Communications

    A. Goldsmith,Wireless Communications. Cambridge University Press, 2005

    2005

  43. [51]

    Making a “completely blind

    A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely blind” image quality analyzer,”IEEE Signal Process. Lett., vol. 20, no. 3, pp. 209–212, 2012

    2012

  44. [52]

    Semantic-aware power allocation for generative semantic communications with foundation models,

    C. Xu, M. B. Mashhadi, Y . Ma, and R. Tafazolli, “Semantic-aware power allocation for generative semantic communications with foundation models,” 2024. [Online]. Available: https://arxiv.org/abs/2407.03050

    2024 arXiv

  45. [53]

    NR; physical layer procedures for data (release 17),

    3rd Generation Partnership Project (3GPP), “NR; physical layer procedures for data (release 17),” 3GPP, Technical Specification (TS) 38.214, 2025. [Online]. Available: https://www.3gpp.org/ftp/Specs/ archive/38_series/38.214

    2025

  46. [54]

    Boyd and L

    S. Boyd and L. Vandenberghe,Convex Optimization. Cambridge University Press, 2004

    2004

  47. [55]

    M. K. Simon and M.-S. Alouini,Digital communication over fading channels. John Wiley & Sons, 2004

    2004

  48. [56]

    NR; multiplexing and channel coding,

    3GPP, “NR; multiplexing and channel coding,” 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 38.212, 2026, version 19.3.0, Release 19. [Online]. Available: https://www.3gpp.org/ ftp/Specs/archive/38_series/38.212/ Chunmei Xu(Member, IEEE) received the B...

    2026

  49. [1990]

    grand challenges in IoT

    From 1990 to 2000, he was a faculty member of Auburn University and later, University of Iowa. Prof. Ding joined the College of Engineering at UC Davis in 2000. His major research interests and expertise cover the areas of wireless networking, communications, signal processing...

    1990

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.