Pith. sign in

REVIEW 4 major objections 7 minor 31 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

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

T0 review · glm-5.2

Video stays recognizable at -8 dB SNR and 90% packet loss

2026-07-09 14:53 UTC pith:YAM7QSOE

load-bearing objection MamVSC: Mamba-based semantic video communication with CSI-adaptive modules; headline result is strong but rests on perfect-CSI assumption the 4 major comments →

arxiv 2607.07293 v1 pith:YAM7QSOE submitted 2026-07-08 cs.ET cs.MM

-8 dB SNR + 90% Packet Loss: MamVSC -- CSI-Guided Semantic Mamba for Extreme-Robust Video Semantic Communication

Lei Teng , Senran Fan , Chen Dong , Haotai Liang , Xiaodong Xu , Ping Zhang This is my paper
classification cs.ET cs.MM
keywords semanticchannellosspacketcommunicationdistortiondynamicallysystem
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes a video semantic communication system called MamVSC that simultaneously addresses two distinct forms of wireless distortion: semantic deviation (continuous noise-induced corruption of semantic vectors) and semantic erasure (catastrophic loss of entire semantic segments due to packet loss). The central mechanism is a channel-state-information (CSI)-guided Mamba architecture that dynamically adjusts three things based on reported channel conditions: (1) the granularity of semantic feature extraction via a CSI-Guided Attentive State-Space Equation, (2) the distance between semantic vectors and their clustering centers in a semantic channel codec, and (3) the recovery strategy for lost packets. The load-bearing claim is that this system maintains usable video quality—MS-SSIM above 0.6 and PSNR above 21 dB—at an SNR of -8 dB and a packet loss rate of 90% in an AWGN channel, conditions under which both traditional codecs (which exhibit a cliff effect) and prior semantic communication systems (designed for only one distortion type) fail entirely.

Core claim

The paper claims that by jointly modeling semantic deviation and semantic erasure through CSI-guided adaptive mechanisms—rather than treating them in isolation—a Mamba-based semantic communication system can preserve recognizable video structure under channel conditions so extreme that the signal is buried 8 dB below the noise floor and 90% of packets never arrive. The key architectural innovation is the CSI-Guided Attentive State-Space Equation (CGASE), which injects channel quality metrics (SNR and packet loss rate) directly into the state-space model's output projection matrix, thereby steering the Mamba scanning mechanism to prioritize semantically critical information that survives both

What carries the argument

The central mechanism is the CSI-Guided Attentive State-Space Equation (CGASE), which modifies the Mamba state-space model's output equation by adding a CSI-derived vector (mapping SNR and packet loss rate through a linear layer) to the output projection matrix. This is combined with a semantic clustering module that groups semantically similar tokens before state-space modeling, and a dynamic semantic channel codec that adjusts the distance between semantic vectors and their clustering centers based on CSI—pulling vectors closer to centers under poor conditions to concentrate robustness, and spreading them apart under good conditions to preserve detail.

Load-bearing premise

The system assumes that accurate channel state information—specifically SNR and packet loss rate—is simultaneously available at both transmitter and receiver. Under the extreme conditions the system claims to handle (-8 dB SNR), the feedback channel carrying this information would itself suffer severe noise, yet the paper does not address how reliable CSI is obtained, how feedback delay is handled, or what happens when the CSI estimates are wrong.

What would settle it

If the system is tested with delayed, noisy, or mismatched CSI at the transmitter and receiver—rather than the idealized simultaneous CSI assumed in the evaluation—and the adaptive modules (CGASE, semantic channel codec, packet loss recovery) degrade performance below that of a non-adaptive baseline, the core contribution of CSI-guided adaptation collapses, leaving only an incremental architecture swap from Swin Transformer to Mamba.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the system performs as claimed, emergency-response and military video links could remain functional in environments where current systems produce only static, bridging the gap between graceful degradation and total failure.
  • The principle of CSI-guided semantic granularity control could extend beyond video to other modalities—audio, sensor telemetry, or text—where the semantic importance of individual features varies with channel quality.
  • Joint handling of deviation and erasure through a single adaptive framework suggests that future 6G semantic communication standards might treat channel noise and packet loss as a unified distortion space rather than separate protocol-layer concerns.
  • The dynamic clustering-center approach implies that semantic communication systems can trade detail for robustness in a continuous, channel-aware fashion, potentially eliminating the cliff effect that has constrained digital communication since Shannon.

Where Pith is reading between the lines

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

  • The system's reliance on accurate CSI at both transmitter and receiver raises an unaddressed question: at -8 dB SNR, the feedback channel carrying CSI would itself be severely degraded, and the paper does not discuss how CSI reliability is maintained under the very conditions the system claims to handle.
  • If the adaptive modules operate on incorrect CSI (e.g., stale or noisy feedback), the semantic clustering, granularity adjustment, and packet loss recovery could amplify distortion rather than mitigate it—a failure mode the paper's evaluation does not test.
  • The claim of MS-SSIM > 0.6 at 90% packet loss implies the system is reconstructing substantial content from only 10% of semantic segments; this suggests the semantic encoder is achieving extreme compression ratios that may lose fine-grained detail invisible to MS-SSIM but critical for tasks like object recognition or tracking.
  • Extending the evaluation to fading channels (not just AWGN) and to scenarios with CSI delay or mismatch would directly test whether the adaptive framework's gains are robust to realistic feedback imperfections.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes MamVSC, a Mamba-based video semantic communication system that jointly addresses semantic deviation (channel noise) and semantic erasure (packet loss) through three CSI-guided modules: a CSI-Guided Attentive State-Space Equation (CGASE) in the Semantic Mamba Block, a Semantic Channel Codec with dynamic semantic clustering centers, and an Adaptive Packet Loss Recovery Module. The headline result is MS-SSIM > 0.6 and PSNR > 21 dB at SNR = -8 dB and 90% packet loss rate in AWGN. The system is compared against H.264/H.265+LDPC and two prior semantic communication systems (MSTVSC, MDVSC) across multiple resolutions and distortion conditions.

Significance. The simultaneous treatment of semantic deviation and semantic erasure is a meaningful contribution to video semantic communication, as most prior work addresses only one. The Mamba backbone with CSI-guided semantic grouping is a reasonable architectural choice, and the experimental comparisons against both traditional codecs and prior semantic systems (Figs. 10-15) are fairly comprehensive within the tested conditions. The dynamic semantic clustering center mechanism (Eq. 26) is an interesting approach to channel-adaptive robustness. However, the significance of the headline result is tempered by the assumption of perfect CSI under the very conditions where CSI acquisition is most difficult.

major comments (4)
  1. §III, Eqs. (25), (26), (27): All three proposed adaptive modules depend on accurate CSI (SNR and packet loss rate) at both transmitter and receiver. The headline result (SNR = -8 dB, 90% packet loss) is obtained under the implicit assumption of perfect CSI. At -8 dB SNR, the receiver must estimate SNR from severely corrupted signals, and the feedback channel to the transmitter would experience comparable or worse noise. The paper does not discuss CSI estimation method, feedback channel model, feedback delay, or CSI error sensitivity anywhere in the manuscript. If CSI is unreliable under these conditions, the packet loss recovery module (Eq. 27) would apply the wrong blend of high-SNR and low-SNR paths, and the semantic channel codec (Eq. 26) would set incorrect distances to clustering centers—potentially degrading performance below a non-adaptive baseline. The paper should include a CSI
  2. error sensitivity analysis: train and/or evaluate with noisy or delayed CSI to demonstrate that the adaptive modules provide robustness benefits even when CSI is imperfect. Without this, the headline result represents an upper bound whose practical achievability is unclear.
  3. §IV.B, Figs. 12-15: All experiments are conducted exclusively in AWGN channels. The title and abstract claim 'extreme-robust' video semantic communication, which implies practical robustness. AWGN does not capture fading, multipath, Doppler effects, or interference—channel conditions that are standard in wireless video transmission evaluation. The performance under more realistic channel models (e.g., Rayleigh fading, 3GPP TR 38.901) should be evaluated, or the claims should be explicitly scoped to AWGN only.
  4. §IV, Figs. 10-15: No error bars, confidence intervals, or statistical significance tests are reported for any experimental result. Given that the performance differences between MamVSC and MSTVSC are sometimes described as 'slight' (e.g., §IV.B.2, Fig. 10), it is unclear whether these differences are statistically significant. Multiple random seeds or repeated trials should be reported to establish the reliability of the comparisons.
minor comments (7)
  1. The paper heavily self-cites prior work from the same group: MSTVSC [17], sDMCM [22], MDVSC [16], and the semantic channel codec concept from [18]. While the central claim is experimentally validated rather than circularly derived, the novelty boundary between MamVSC and MSTVSC [17] should be made clearer—specifically, which modules are entirely new versus adapted from [17].
  2. No computational complexity or latency analysis is provided. For real-time video communication, the inference latency of the Mamba-based codec and the adaptive modules is a practical concern, especially at 4K resolution (3840 × 2160).
  3. §III.A, Eq. (25): The notation S and L in the term (C+P+S+L) is introduced without clear definition. S presumably relates to CSI-derived SNR and L to packet loss rate, but this should be stated explicitly.
  4. §III.C, Fig. 8: The histogram shows received semantic information distributions at different SNRs, but the specific sDMCM [22] modulation parameters and constellation size are not stated in this section, making the figure difficult to interpret without consulting [22].
  5. §II, Fig. 2: The frame structure diagram references SDS, OH, AH, FCS, FEC but the figure is not fully legible in the provided text. The relationship between semantic data segment size and packet size is not specified.
  6. §IV.A.3, Eq. (28): CBR is defined as v/l, but the specific values of v and l for the tested configurations are not tabulated. The CBR = 0.0078 used in Figs. 11 and 14-15 should be contextualized with the actual bandwidth savings relative to uncompressed transmission.
  7. The abstract states 'a adaptive packet loss recovery module'—should be 'an adaptive.'

Circularity Check

0 steps flagged

No significant circularity; central claim is experimentally measured, not derived from self-cited prior work

full rationale

The paper's central claim—MS-SSIM > 0.6 and PSNR > 21 dB at -8 dB SNR and 90% packet loss—is an empirical result measured against external benchmarks (H.264, H.265) and the authors' own prior systems (MSTVSC [17], MDVSC [16]). The self-citations are extensive but not circular: the paper does not claim to derive its headline performance from these prior works. Instead, it builds on prior architectural components (the MSTVSC framework, sDMCM modulation) and proposes new modules (CGASE, dynamic semantic channel codec, adaptive packet loss recovery) with independently specified equations (Eq. 25, 26, 27). The semantic channel codec concept from [18] is cited as inspiration, but the specific dynamic clustering center scaling mechanism (Eq. 26) is newly formulated here. The Attentive State-Space Equation is attributed to MambaIRv2 [24] (external, different authors), and the CSI-guided extension (CGASE, Eq. 25) is a novel modification. No equation reduces to its own input by construction, no prediction is a renamed fit, and no uniqueness theorem is invoked. The self-citations establish architectural lineage and baseline comparisons, which is normal incremental research practice. The perfect-CSI assumption is a validity concern (correctness risk), not a circularity issue—the experimental results are not tautologically forced by assuming what they claim to measure. Score 2 reflects the heavy self-citation pattern where baselines are dominated by the authors' own prior systems, but the central claim retains independent empirical content validated against external standards (H.264/H.265 + LDPC).

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 2 invented entities

The system assumes that accurate channel state information (SNR and packet loss rate) is available at both transmitter and receiver simultaneously (§III, Eq. 25, Eq. 26, Eq. 27). At -8 dB SNR, obtaining reliable CSI feedback is itself a significant challenge—the feedback channel would also suffer severe noise. The paper does not discuss CSI feedback reliability, delay, or error. If CSI is unreliable under the very conditions the system claims to handle, the adaptive modules (CGASE, semantic channel codec, packet loss recovery) would operate on incorrect channel estimates, potentially degrading rather than improving performance.

free parameters (7)
  • W (semantic channel codec weight) = learned via training
    Controls distance between semantic vectors and clustering centers (Eq. 26); generated by a weight module from CSI fusion, trained end-to-end
  • WP (packet loss recovery weight) = learned via training
    Blends high-SNR and low-SNR recovery paths (Eq. 27); generated by CSI-controlled weight module with sigmoid, trained end-to-end
  • CSI linear layer parameters (S, L in CGASE) = learned via training
    Maps SNR and packet loss rate to d-dimensional vector added to output matrix C (Eq. 25); trained end-to-end
  • GOP size n = 4
    Set by hand; number of frames per group of pictures
  • Quantization bits = 3
    Set by hand; all semantic models use 3-bit quantization
  • Number of semantic clusters = not specified
    Used in semantic clustering module but exact count not stated
  • Network depth (N1, N2, N3) = not specified
    Number of Semantic Mamba Blocks per stage not given
axioms (4)
  • domain assumption CSI (SNR and packet loss rate) is available at both transmitter and receiver
    Used throughout (Eqs. 25-27) but never justified; at -8 dB SNR, CSI feedback reliability is itself questionable
  • domain assumption AWGN channel model is adequate for evaluating extreme-robustness claims
    All experiments use AWGN; no fading, multipath, or realistic channel models tested
  • domain assumption Semantic information can be meaningfully clustered into discrete groups
    Foundation of the semantic clustering module and SGN-unfold process (§III-A-3)
  • standard math State-space models (Mamba) provide adequate global modeling for video semantic extraction
    Builds on established SSM theory; supported by [9], [10], [24]
invented entities (2)
  • Dynamic Semantic clustering centers (S_b) no independent evidence
    purpose: Serve as reference points for adjusting semantic vector distances based on channel conditions
    Computed per-GOP from semantic features; no external validation that this clustering corresponds to meaningful semantic categories
  • Semantic erasure matrices (M_c, M_i) independent evidence
    purpose: Mark positions of lost semantic information for packet loss recovery
    Generated from application-layer header segment tags; falsifiable by checking packet loss patterns against header verification

pith-pipeline@v1.1.0-glm · 16759 in / 4350 out tokens · 146047 ms · 2026-07-09T14:53:06.666114+00:00 · methodology

0 comments
read the original abstract

Semantic communication, leveraging joint source-channel coding, is designed to mitigate semantic distortion introduced by the channel. However, most current studies focus solely on semantic deviation distortion caused by physical wireless channels, while overlooking semantic erasure distortion due to packet loss. A CSI-Guided Mamba-based video semantic wireless digital communication system (MamVSC) employing semantic grouping is proposed to simultaneously address both semantic deviation and erasure distortions. In this system, a semantic Mamba module, guided by channel state information (CSI) feedback, is utilized to dynamically adjust the granularity of extracted semantic information, adapting to channel conditions. Furthermore, a Semantic Channel Codec based on dynamic Semantic clustering centers is introduced, where the distance between semantic vectors within the same semantic class and their corresponding Semantic clustering center is dynamically adjusted according to channel conditions, enhancing robustness against channel noise. Additionally, a adaptive packet loss recovery module, dynamically adaptive to the CSI, is proposed. The system achieves an MS-SSIM greater than 0.6 and a PSNR exceeding 21 dB at an SNR of -8 dB and a packet loss rate of 90% in AWGN channel.

Figures

Figures reproduced from arXiv: 2607.07293 by Chen Dong, Haotai Liang, Lei Teng, Ping Zhang, Senran Fan, Xiaodong Xu.

Figure 1
Figure 1. Figure 1: The architecture diagram of the semantic digital communication system is presented, illustrating the two sources of semantic distortion in wireless communication: the wireless channel and packet loss. At the transmitter, source data is processed by a semantic encoder to extract semantic information. This information undergoes semantic-level interleaving and segmentation, with an application-layer header (A… view at source ↗
Figure 3
Figure 3. Figure 3: The overall architecture of our MamVSC for wireless video transmission. The orange box represents the newly proposed framework in this paper, comprising the Semantic Mamba Block, Semantic Channel Codec, and Adaptive Packet Loss Recovery Module, while the white box denotes the basic modules proposed in our previous article [17]. CSI is incorporated into the codec to enhance robustness against semantic disto… view at source ↗
Figure 4
Figure 4. Figure 4: The architecture of the spatial semantic code. The network architecture of the Spatial Semantic Encoder￾Decoder is depicted in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The architecture of the Semantic Mamba Module (SMM). 2) CSI-Guided Attentive State-Space Equation: The ob￾jective is to modify the output matrix C ∈ R L×d , where L = HW represents the flattened image sequence length and d denotes the number of hidden states in Mamba, to achieve adaptive semantic information extraction based on channel conditions. To this end, a CSI-Guided Attentive State-Space Equation (C… view at source ↗
Figure 6
Figure 6. Figure 6: The architecture of the Semantic Channel Codec [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Diagram of the Semantic Channel Codec Functionality. In the figure, the red arrow represents convergence toward the semantic clustering center, while the blue arrow represents divergence from the semantic clustering center. (characteristic features) and Semantic clustering centers (com￾mon information). The proposed semantic channel codec is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Histogram of Received Semantic Information Distribution under Different SNRs. under varying SNR conditions. It can be observed that at high SNR, the semantic information distribution approximates a Gaussian distribution, primarily concentrated around values 3 and 4. As SNR decreases, the distribution shifts toward the extremes, with the proportions of values 0 and 7 increasing rapidly. This is because lowe… view at source ↗
Figure 9
Figure 9. Figure 9: The architecture of the Adaptive Packet Loss Recovery Module. The proposed Adaptive Individual Packet Loss Recovery Module is depicted in [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The relationship between PSNR and MS-SSIM performance of different encoding-decoding schemes and CBR under error-free transmission is depicted. + 1/2 LDPC, except at the 3840 × 2160 resolution. Compared to other video semantic communication systems, MamVSC significantly outperforms MDVSC and slightly outperforms MSTVSC. Regarding MS-SSIM, MamVSC demonstrates clear superiority over the channel-robust combi… view at source ↗
Figure 11
Figure 11. Figure 11: An example of visual comparison is provided, with the CBR set to 0.0078. For the first two rows, the first column displays the original frame, the second column shows a cropped patch from the original frame, and the third to sixth columns present the reconstructed frames using different schemes as shown in [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The three-dimensional surfaces plot illustrating the variation of semantic performance metrics MS-SSIM with SNR and semantic erasure probability in AWGN channel is presented. Additionally, performance curves of MS-SSIM versus SNR are shown for the case of zero semantic erasure probability, together with MS-SSIM versus semantic erasure probability under a fixed SNR of 20 dB [PITH_FULL_IMAGE:figures/full_f… view at source ↗
Figure 13
Figure 13. Figure 13: The three-dimensional surfaces plot illustrating the variation of semantic performance metrics PSNR with SNR and semantic erasure probability in AWGN channel is presented. Additionally, performance curves of PSNR versus SNR are shown for the case of zero semantic erasure probability, together with PSNR versus semantic erasure probability under a fixed SNR of 20 dB [PITH_FULL_IMAGE:figures/full_fig_p009_13.png] view at source ↗
Figure 15
Figure 15. Figure 15: Visualization results under extreme semantic distortion conditions (SNR = -8 dB, packet loss rate = 90%). The first row shows the original frame. The second to fourth rows present the reconstructed video frames obtained by MamVSC, MSTVSC, and MDVSC, respectively. The subtitles indicate SNR/the packet loss/PSNR/MS-SSIM values. in this extremely challenging setting, the MamVSC scheme is still able to preser… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · 3 internal anchors

  1. [1]

    Intellicise wireless networks from semantic commu- nications: A survey, research issues, and challenges,

    P. Zhanget al., “Intellicise wireless networks from semantic commu- nications: A survey, research issues, and challenges,”IEEE Commun. Surveys Tuts., vol. 27, no. 3, pp. 2051–2084, Jun. 2025

    work page 2051

  2. [2]

    IMT-2030 framework and study groups,

    International Telecommunication Union, “IMT-2030 framework and study groups,” ITU-R Study Groups, 2025. [Online]. Available: https://www.itu.int/en/ITU-R/study-groups/rsg5/rwp5d/imt- 2030/Pages/default.aspx

    work page 2030

  3. [3]

    A mathematical theory of communication,

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

    work page 1948

  4. [4]

    Recent contributions to the mathematical theory of com- munication,

    W. Weaver, “Recent contributions to the mathematical theory of com- munication,”ETC: Rev. Gen. Semantics, vol. 10, no. 4, pp. 261–281, 1953

    work page 1953

  5. [5]

    Semantic communications: Overview, open issues, and future research directions,

    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

  6. [6]

    A paradigm shift toward semantic communications,

    K. Niuet al., “A paradigm shift toward semantic communications,”IEEE Commun. Mag., vol. 60, no. 11, pp. 113–119, Nov. 2022

    work page 2022

  7. [7]

    Toward wisdom-evolutionary and primitive-concise 6G: A new paradigm of semantic communication networks,

    P. Zhanget al., “Toward wisdom-evolutionary and primitive-concise 6G: A new paradigm of semantic communication networks,”Engineering, vol. 8, pp. 60–73, 2022

    work page 2022

  8. [8]

    Semantic- importance-aware communication over MIMO fading channels,

    H. Liang, C. Dong, W. An, Z. Bao, X. Xu, and R. Meng, “Semantic- importance-aware communication over MIMO fading channels,”IEEE Internet Things J., vol. 12, no. 18, pp. 38540–38555, Sept. 15, 2025

    work page 2025

  9. [9]

    Vision Mamba: Efficient visual representation learning with bidirectional state space model,

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, and X. Wang, “Vision Mamba: Efficient visual representation learning with bidirectional state space model,” inProc. 41st Int. Conf. Mach. Learn., 2024, pp. 62429– 62442

    work page 2024

  10. [10]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,” arXiv preprint arXiv:2312.00752, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  11. [11]

    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

    work page 2021

  12. [12]

    Deep learning for joint source- channel coding of text,

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

    work page 2018

  13. [13]

    A perceptually mo- tivated approach for low-complexity speech semantic communication,

    X. Chen, J. Wang, L. Xu, J. Huang, and Z. Fei, “A perceptually mo- tivated approach for low-complexity speech semantic communication,” IEEE Internet Things J., vol. 11, no. 12, pp. 22054–22065, Jun. 2024

    work page 2024

  14. [14]

    DeepJSCC-f: Deep joint source-channel coding of images with feedback,

    D. B. Kurka and D. G ¨und¨uz, “DeepJSCC-f: Deep joint source-channel coding of images with feedback,”IEEE J. Sel. Areas Inf. Theory, vol. 1, no. 1, pp. 178–193, May 2020

    work page 2020

  15. [15]

    Nonlinear transform source-channel coding for semantic communications,

    J. Daiet al., “Nonlinear transform source-channel coding for semantic communications,”IEEE J. Sel. Areas Commun., vol. 40, no. 8, pp. 2300– 2316, Aug. 2022

    work page 2022

  16. [16]

    MDVSC—Efficient wireless model division video se- mantic communication,

    Z. Baoet al., “MDVSC—Efficient wireless model division video se- mantic communication,”IEEE Internet Things J., vol. 12, no. 2, pp. 1109–1124, Jan. 2025

    work page 2025

  17. [17]

    Conquering high packet-loss erasure: MoE Swin transformer-based video semantic communication,

    L. Teng, S. Fan, C. Dong, H. Liang, Z. Bao, and X. Xu,et al., “Conquering high packet-loss erasure: MoE Swin transformer-based video semantic communication,”IEEE Trans. Cogn. Commun. Netw., vol. 12, pp. 9279–9296, 2026

    work page 2026

  18. [18]

    On the discussion of the semantic communication with the transmission of MNIST dataset,

    K. Liu, C. Dong, X. Xu, and G. Liu, “On the discussion of the semantic communication with the transmission of MNIST dataset,” inProc. 8th Int. Conf. Comput. Commun. Syst. (ICCCS), Guangzhou, China, 2023, pp. 875–879

    work page 2023

  19. [19]

    A Mathematical Theory of Semantic Communication

    K. Niu and P. Zhang, “A mathematical theory of semantic communica- tion,” arXiv:2401.13387, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  20. [20]

    W AFI-VSC: Wireless adaptive frame interpolation video semantic communication,

    X. Qi, N. Ma, Z. Bao, Y . Liu, C. Dong, and X. Xu, “W AFI-VSC: Wireless adaptive frame interpolation video semantic communication,” inProc. 16th Int. Conf. Wireless Commun. Signal Process. (WCSP), Hefei, China, 2024, pp. 1503–1508. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11

    work page 2024

  21. [21]

    Toward adaptive semantic communications: Efficient data transmission via online learned nonlinear transform source-channel coding,

    J. Daiet al., “Toward adaptive semantic communications: Efficient data transmission via online learned nonlinear transform source-channel coding,”IEEE J. Sel. Areas Commun., vol. 41, no. 8, pp. 2609–2627, Aug. 2023

    work page 2023

  22. [22]

    sDMCM—Semantic digital modulation constellation mapping scheme for semantic communication,

    L. Teng, W. An, C. Dong, and X. Xu, “sDMCM—Semantic digital modulation constellation mapping scheme for semantic communication,” IEEE Internet Things J., vol. 12, no. 12, pp. 20885-20901, 15 June, 2025

    work page 2025

  23. [23]

    Video enhance- ment with task-oriented flow,

    T. Xue, B. Chen, J. Wu, D. Wei, and W. T. Freeman, “Video enhance- ment with task-oriented flow,”Int. J. Comput. Vis., vol. 127, no. 8, pp. 1106–1125, Aug. 2019

    work page 2019

  24. [24]

    MambaIRv2: Attentive state space restoration,

    H. Guo, Y . Guo, Y . Zha, Y . Zhang, W. Li, T. Dai, S.-T. Xia, and Y . Li, “MambaIRv2: Attentive state space restoration,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Nashville, TN, USA, Jun. 2025, pp. 28124–28133

    work page 2025

  25. [25]

    Common test conditions and software reference configurations,

    F. Bossenet al., “Common test conditions and software reference configurations,” Int. Organ. Standard., Geneva, Switzerland, document JCTVC-L1100, 2013

    work page 2013

  26. [26]

    UVG dataset: 50/120fps 4K sequences for video codec analysis and development,

    A. Mercat, M. Viitanen, and J. Vanne, “UVG dataset: 50/120fps 4K sequences for video codec analysis and development,” inProc. 11th ACM Multimedia Syst. Conf., Istanbul, Turkey, Jun. 2020, pp. 297–302

    work page 2020

  27. [27]

    Overview of the H.264/A VC video coding standard,

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

    work page 2003

  28. [28]

    Overview of the high efficiency video coding (HEVC) standard,

    G. J. Sullivan, J.-R. Ohm, W. Han, and T. Wiegand, “Overview of the high efficiency video coding (HEVC) standard,”IEEE Trans. Circuits Syst. Video Technol., vol. 22, no. 12, pp. 1649–1668, Dec. 2012

    work page 2012

  29. [29]

    Design of low-density parity check codes for 5G new radio,

    T. J. Richardson and S. Kudekar, “Design of low-density parity check codes for 5G new radio,”IEEE Commun. Mag., vol. 56, no. 3, pp. 28–34, Mar. 2018

    work page 2018

  30. [30]

    Multiscale structural similarity for image quality assessment,

    Z. Wang, E. P. Simoncelli, and A. C. Bovik, “Multiscale structural similarity for image quality assessment,” inProc. 37th Asilomar Conf. Signals, Syst., Comput., Pacific Grove, CA, USA, Nov. 2003, pp. 1398– 1402

    work page 2003

  31. [31]

    End-to-end Optimized Image Compression

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “End-to-end opti- mized image compression,”arXiv preprint, 2016. [Online]. Available: https://arxiv.org/abs/1611.01704

    work page internal anchor Pith review Pith/arXiv arXiv 2016