{"id":"f79b8286-94db-45b2-b58d-d618f17371ea","arxiv_id":"2411.12776","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A cross-layer framework integrates Deep JSCC semantic coding with encryption, CRC, LDPC, and retransmission, adapting error protection to semantic importance for panoramic video.","lead":"This paper proposes a cross-layer encrypted semantic communication framework that lets semantic video transmission work with standard network layers by adding encryption, CRC checks, and retransmission. It is relevant because it could reduce bandwidth for panoramic video in 6G while keeping important content protected under weak channels.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (29) defines majority voting with argmin, selecting the minority bit across retransmissions; if the simulator follows this equation, the adaptive retransmission mechanism cannot produce the reported gains.","rationale":"I examined the strongest claim: CLESC achieves an 85% bandwidth reduction over H.264 at equal WS-PSNR and maintains quality at low SNR. The claim is simulation-based and depends on the adaptive cross-layer mechanism, specifically on how retransmissions are combined. The reader's weakest assumption was that the entropy map faithfully measures semantic importance. That is a legitimate modeling concern and would benefit from an equal-error-protection ablation. However, I found a more immediate, internal inconsistency: Eq. (29) uses argmin where majority voting requires argmax. This is not a disagreement with community consensus; it is a contradiction within the paper's own algorithm. If followed literally, it would make retransmissions harmful rather than helpful, so the reported retransmission and low-SNR results cannot be attributed to the described mechanism. The concern is concrete and testable: implement the algorithm as written and compare against the argmax correction. The reader's conditional verdict remains appropriate; the paper should not be accepted until Eq. (29) is fixed and the experiments are either reproduced or clarified. I set agreement to 'partial' because my concern targets the same adaptive mechanism but differs from the reader's entropy-importance concern.","tokens_in":22011,"tokens_out":4060,"duration_ms":46866,"concrete_test":"Implement Algorithm 2 exactly as written, with Eq. (29) using argmin over b of Σ_u 1[c_{v,u}=b], and rerun the Section VI-B experiments at CBR=0.04, SNR=5 dB, and the retransmission sweep. If decoded WS-PSNR does not collapse relative to a single transmission, then the published equation does not describe the implemented decoder. Separately, run the same experiments with argmax; if the reported gains reappear, Eq. (29) must be corrected in the manuscript before the adaptive retransmission results are accepted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that CLESC/APVST reduces bandwidth by 85% and preserves quality at low SNR—rests on the adaptive cross-layer mechanism in Algorithm 1 and 2. That mechanism uses majority voting (MV) to combine retransmissions, defined in Section IV as Eq. (29): ˆcv = argmin_b Σ_{u=1}^U f_I(c_{v,u}=b). Because f_I equals 1 when the v-th codeword bit in retransmission u equals b, the sum counts how many retransmissions contain bit value b; minimizing over b picks the value with the fewest votes. Correct majority combining requires argmax, not argmin. As written, the rule systematically inverts high-priority bits, and since priority level 5 allows up to 10 retransmissions, the effect would be strongest for the most protected packets. Figs. 5c, 6c, and 7c therefore cannot be reproduced from the published algorithm unless Eq. (29) is a typographical error and the implementation actually used argmax. The paper provides no code or erratum, so this internal inconsistency is a load-bearing obstacle to the bandwidth and low-SNR claims. The reader's concern about entropy-based importance is valid but secondary: even a correct priority mapping cannot be validated until the stated combining rule is corrected.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes CLESC, a cross-layer encrypted semantic communication framework for panoramic video transmission, and APVST, a Deep JSCC-based neural codec with a weighted attention module and a latitude adaptive module. The framework integrates semantic feature extraction, Deep JSCC encoding, encryption, CRC, LDPC channel coding, and retransmissions, with an adaptive cross-layer mechanism that maps per-group entropy (semantic importance) to CRC length, LDPC rate, and retransmission count. The authors claim that, compared with traditional cross-layer transmission using H.264/H.265, CLESC reduces bandwidth consumption by 85% while maintaining WS-PSNR, and that it avoids the conventional cliff effect at low SNR. The claims are supported by simulations over panoramic video datasets against H.264, H.265, DVST, and an APVST variant without the WA module.","tokens_in":22313,"tokens_out":6535,"duration_ms":71375,"significance":"If the stated results hold, the paper addresses a real and timely problem: making semantic communication compatible with a traditional protocol stack (encryption, CRC, LDPC, retransmission) while retaining compression and robustness gains. The framework is coherent, the problem is well motivated, and the evaluation against external baselines is a strength. However, the evidence is entirely simulation-based, no code or data is released, and the central quantitative claim rests on at least one internal inconsistency in the combining rule and on an asymmetric baseline comparison. The proposed entropy-based importance mechanism is also not validated by any ablation that isolates it from the codec gains. These issues must be resolved before the 85% bandwidth claim and the low-SNR robustness claim can be considered supported.","major_comments":[{"comment":"The majority-voting combining rule is written as argmin over b of the number of retransmissions whose bit equals b. This selects the minority bit, not the majority bit. Since priority level 5 in Table II allows up to 10 retransmissions, applying Eq. (29) as written would increasingly invert the most protected bits, so the low-SNR gains and the curves in Figs. 5(c), 6(c), and 7(c) cannot be reproduced from the published algorithm. Please correct the equation to use argmax if that is what was implemented, or state explicitly which combining rule was actually simulated. The subscript in c_{v,k} should also index the retransmission u, not a subcarrier index k.","section":"Section IV, Eq. (29)"},{"comment":"The baseline comparison is asymmetric: H.264 and H.265 are simulated with a fixed LDPC rate of 1/2 and a maximum retransmission count of 4, while APVST uses the adaptive LDPC rates 2/3, 1/2, 1/3 and retransmission limits of 2 through 10 given in Table II. The claimed 85% bandwidth reduction therefore compares unequal protection and resource budgets. Please report matched-overhead comparisons, for example by equalizing average total channel symbols or average retransmission counts, or by presenting rate-distortion curves where each scheme operates at its own optimal settings.","section":"Section VI-A3 and VI-B1"},{"comment":"The paper's headline claim that APVST reduces bandwidth by 85% compared with H.264 is not tied to a specific operating point. In Fig. 5(a) the horizontal gap between the APVST and H.264 curves varies with CBR, so the percentage reduction depends on the chosen WS-PSNR level or CBR. Please state the exact comparison point, report the corresponding WS-PSNR values, and indicate how sensitive the 85% figure is to that choice.","section":"Section VI-B1"},{"comment":"There is no ablation that isolates the adaptive cross-layer mechanism, which is the paper's central contribution. The comparisons in Figs. 5-7 contrast APVST against DVST, H.264, and H.265, but they do not compare APVST with entropy-based priority against APVST with equal CRC, LDPC, and retransmission settings at matched overhead. Without such an ablation, the observed gains could be entirely due to the semantic codec rather than to the CLESC priority mechanism. Please add a control experiment, for example all packets assigned to the middle importance level, or a random priority assignment.","section":"Section VI, overall"},{"comment":"The adaptive mechanism treats the entropy map e_t as a faithful measure of semantic importance, and this map is the same entropy signal that the training loss in Eq. (27) minimizes. The paper provides no evidence that entropy ranking by groups corresponds to the contribution of those groups to final WS-PSNR or WS-SSIM. This is a modeling assumption rather than a derived result. Please validate it, for example by comparing the proposed entropy-based protection allocation with an allocation based on actual per-group distortion sensitivity, and by reporting the resulting WS-PSNR and WS-SSIM at low SNR.","section":"Sections III-A1, III-A2, and Algorithm 1"}],"minor_comments":[{"comment":"The quantization set Q = {0, 2, 4, 6, 8, 10, 16, 20, 26, 32, 20, 48, 56, 64, 80, 96} contains the value 20 twice; this is presumably a typo, and the set should be monotone.","section":"Section VI-A1"},{"comment":"The heading uses \"APSVT\" where the acronym should be \"APVST\".","section":"Section V-B heading"},{"comment":"The initialization of the packet data for the 0-th retransmission uses the variable u from the preceding importance-indication loop; it should use the 0-th retransmission index for the packet data as well.","section":"Algorithm 2, line 14"},{"comment":"The curves show no error bars or confidence intervals, even though the simulation involves random channel fading and stochastic retransmissions. Reporting mean plus/minus standard deviation would make the comparisons, especially the 85% claim, more persuasive.","section":"Figures 5-7 and 9"},{"comment":"The bandwidth-reduction percentages appear inconsistent across figures: Section VI-B1 reports 85% and 33% for H.264 and H.265, respectively, while Section VI-B2 reports 44% and 85%. Please clarify that these correspond to different metrics (WS-PSNR vs. WS-SSIM) and identify the exact operating points.","section":"Section VI-B1 and VI-B2"},{"comment":"The text says the reconstruction loss used for training is LPIPS, whereas Sections VI-B1 and VI-B2 state that WMSE and negative WS-SSIM are used. Please clarify whether separate models are trained and evaluated for each metric.","section":"Section VI-B3"}],"recommendation":"major_revision","confidential_remarks":"The framework is promising and the topic is timely, but the central claims currently rest on an internal inconsistency in Eq. (29), an asymmetric baseline comparison, and the absence of an ablation isolating the adaptive cross-layer mechanism. The 85% bandwidth claim in the abstract is likely to attract attention and should be verified after correcting the combining rule and re-running matched-overhead comparisons. The authors should also be asked to either release code or provide enough implementation detail to reproduce the simulation. The novelty claim of being \"the first to achieve compatibility\" should be checked against existing cross-layer semantic HARQ works such as SemHARQ [22] and the RS-coded HARQ approach in [23]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper's real contribution is the CLESC framework itself: mapping semantic importance onto CRC length, LDPC rate, and retransmission count inside a conventional protocol stack, with encryption added. That is a sensible integration step for 6G immersive services, and the APVST codec, while incremental over the authors' WCNC 2024 work and DVST, is competent. The paper deserves credit for taking the compatibility problem seriously rather than hand-waving at the physical layer.\n\nBut there is a load-bearing problem. Equation (29) defines majority voting for retransmission combining with an argmin over the indicator sum. For a binary bit with U retransmissions, that selects the value with fewer votes, i.e., the minority bit. Correct combining requires argmax. As printed, the algorithm would systematically invert the most heavily protected bits, so the reported WS-PSNR/WS-SSIM/LPIPS gains in Figs. 5-7 cannot be reproduced. Either the simulator used argmax, in which case the equation is a typographical error, or it followed the paper, in which case the results are unexplained. The paper provides no code or erratum, so this is a blocking issue, not a cosmetic one.\n\nThe reader's other concerns are valid and secondary. There is no ablation isolating the entropy-based priority mechanism against equal error protection, so the adaptive mechanism's contribution is unverified. The baselines are asymmetric: H.264/H.265 are fixed at rate-1/2 LDPC and at most four retransmissions, while the semantic scheme adapts across a wider table. No error bars are reported, and the entropy-as-importance assumption is plausible but not tested against alternatives. These are all fixable with code release, an equal-protection baseline, and variance reporting.\n\nThe paper is coherent aside from Eq. (29), and the authors engage seriously with the prior art. This should go to peer review, but only with a strong request to correct the voting rule, release code and trained models, and add the missing ablation. Without those, the headline 85% bandwidth claim is not something I would rely on.","headline":"A plausible cross-layer semantic communication framework that is currently undermined by a load-bearing majority-voting typo and missing reproducibility, so the 85% bandwidth claim is not yet credible.","tokens_in":679,"tokens_out":1694,"would_cite":false,"duration_ms":38020,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Semantic video transmission can cut bandwidth by 85 percent while surviving low-SNR channels, claims new cross-layer framework.","keywords":["semantic communication","cross-layer design","deep joint source-channel coding","panoramic video","entropy model","adaptive transmission","WS-PSNR","encrypted communication"],"falsifier":"Train or simulate a version of APVST that replaces the entropy-based importance labels with a fixed equal-protection policy (same CRC length, LDPC rate, and retransmission count for all groups), holding total bandwidth constant, and compare WS-PSNR over the same channel SNR range; if the equal-protection version matches or beats the entropy-based version, the central adaptive mechanism is not delivering the claimed benefit.","tokens_in":21836,"feed_emoji":"🎥","tokens_out":3987,"duration_ms":39365,"temperature":0.7,"pith_summary":"This paper claims that semantic communication, which compresses video at the level of meaning rather than pixels, can be grafted onto the ordinary layered protocol stack without losing its advantage. The authors build a cross-layer encrypted semantic communication framework in which a deep JSCC codec for panoramic video is followed by encryption, CRC, LDPC coding, and retransmission, with all of those classic error-control steps adapted per packet according to a learned entropy map. Their central evidence is that, at equal WS-PSNR, the scheme uses about 85 percent less bandwidth than an H.264-based cross-layer system, and that it degrades gracefully at low SNR where conventional transmission shows a cliff effect. The paper's significance is that it makes a case for semantic communication being deployable inside existing mobile networks instead of replacing them.","feed_headline":"Semantic video cuts bandwidth by 85 percent in cross-layer test","feed_subtitle":"Panoramic transmission keeps quality at low SNR by protecting high-entropy semantic packets.","key_machinery":"The load-bearing mechanism is the entropy map $e_t$ of the semantic feature map $y_t$, which is grouped into spatial blocks and converted into a set of importance indications $l_{i,t}$ that control three mapping functions in the transmitter: the choice of CRC generator polynomial, the LDPC generator matrix, and the maximum retransmission count. These mappings let the physical and data-link layers allocate error-control resources according to estimated semantic importance, while the latitude-adaptive module and a weighted spatial attention module shape the entropy map itself by restricting the information dimension of each feature point according to latitude.","core_discovery":"The paper's core claim is that semantic communication and traditional cross-layer transmission are compatible, and that the combination can be made adaptive to channel conditions. In the proposed CLESC framework, the application layer extracts semantic features and encodes them at variable length with a Deep JSCC network, then labels each packet with an importance indication derived from the entropy map of the feature map. The data link and physical layers read that label and choose CRC strength, LDPC rate, and retransmission count accordingly, so that high-priority semantic content receives stronger protection under poor channels. The accompanying APVST codec adds a latitude-adaptive module and a weight-attention module to exploit equirectangular projection redundancy. The authors report that the resulting system exceeds H.264- and H.265-based cross-layer schemes in WS-PSNR, WS-SSIM, and LPIPS at equal bandwidth, and specifically that it achieves the same WS-PSNR with 85 percent less bandwidth than H.264 while avoiding the cliff effect at low SNR.","pith_inferences":["Beyond the paper: the entropy-based priority rule assumes that high-entropy features dominate perceptual quality; a natural test is to compare the scheme against a variant that protects the low-entropy but attended regions the user is actually looking at, which the current model does not model.","Beyond the paper: the reported gains are against H.264/H.265 with fixed LDPC and retransmission settings; the comparison would be fairer if the traditional schemes were also given the same latitude-adaptive rate allocation.","Beyond the paper: the compatibility claim suggests an incremental deployment path — semantic codecs could be introduced as an application-layer upgrade inside an otherwise unchanged mobile network."],"forward_implications":["If the 85 percent bandwidth reduction holds at equal WS-PSNR, panoramic video services could be delivered over existing cellular infrastructure at a fraction of the current resource cost.","The framework extends the principle of semantic-aware hybrid ARQ to the whole protocol stack, implying that entropy-based importance labels could be used by schedulers and resource allocators in 6G networks.","Because the design inserts encryption before channel coding, it shows that semantic communication need not sacrifice security or error detection to gain compression.","The same CLESC structure is claimed to generalize to other modal data, since the cross-layer adaptation only requires an entropy map or equivalent importance map."],"supporting_citations":[{"why":"Provides the DVST deep video semantic transmission baseline that CLESC builds on and compares against.","marker":"[10]"},{"why":"The authors' prior panoramic video semantic transmission codec that APVST extends with the latitude-adaptive and weight-attention modules.","marker":"[13]"},{"why":"Nonlinear transform source-channel coding used for the I-frame coding within a group of pictures.","marker":"[9]"},{"why":"Defines WS-PSNR, the primary metric on which the 85 percent bandwidth reduction claim is based.","marker":"[17]"},{"why":"Defines WS-SSIM, the perceptual quality metric used in the evaluation.","marker":"[18]"},{"why":"H.264 standard is the main traditional baseline against which the bandwidth reduction is measured.","marker":"[52]"},{"why":"SemHARQ provides the prior semantic-aware retransmission idea that the adaptive cross-layer mechanism generalizes.","marker":"[22]"},{"why":"Swin transformer v2 is the backbone of the Deep JSCC encoder and decoder in APVST.","marker":"[49]"}],"fun_headline_variants":["Semantic cross-layer video slashes bandwidth 85% at low SNR","Adaptive semantic protection trims video bandwidth by 85%","Cross-layer semantic coding beats H.264 with 85% less bandwidth","Semantic video protection shrinks bandwidth 85% under poor channels","Important semantic packets get stronger codes; bandwidth drops 85%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The framework's gains depend on the learned entropy map of the semantic feature map being a faithful measure of how much each spatial group contributes to the final WS-PSNR and WS-SSIM at the receiver; if entropy does not track perceptual or task importance, the priority-based resource allocation will not deliver the claimed efficiency.","fun_headline_variants_meta":{"raw":{"variants":["Semantic cross-layer video slashes bandwidth 85% at low SNR","Adaptive semantic protection trims video bandwidth by 85%","Cross-layer semantic coding beats H.264 with 85% less bandwidth","Semantic video protection shrinks bandwidth 85% under poor channels","Important semantic packets get stronger codes; bandwidth drops 85%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000478,"raw_usage":{"total_tokens":2370,"prompt_tokens":951,"completion_tokens":1419,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":1328}},"tokens_in":567,"tokens_out":1419,"duration_ms":11092,"temperature":1.0,"reasoning_tokens":1328,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T17:43:10.666839+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train or simulate a version of APVST that replaces the entropy-based importance labels with a fixed equal-protection policy (same CRC length, LDPC rate, and retransmission count for all groups), holding total bandwidth constant, and compare WS-PSNR over the same channel SNR range; if the equal-protection version matches or beats the entropy-based version, the central adaptive mechanism is not delivering the claimed benefit.","supporting_citations":[{"cited_title":"Semantic communication-enabled wireless adaptive panoramic video transmission,","cited_arxiv_id":null,"evidence_quote":"The authors' prior panoramic video semantic transmission codec that APVST extends with the latitude-adaptive and weight-attention modules."},{"cited_title":"Weighted-to-spherically-uniform quality evaluation for omnidirectional video,","cited_arxiv_id":null,"evidence_quote":"Defines WS-PSNR, the primary metric on which the 85 percent bandwidth reduction claim is based."},{"cited_title":"Weighted-to- spherically-uniform SSIM objective quality evaluation for panoramic video,","cited_arxiv_id":null,"evidence_quote":"Defines WS-SSIM, the perceptual quality metric used in the evaluation."},{"cited_title":"Overview of the H. 264/A VC video coding standard,","cited_arxiv_id":null,"evidence_quote":"H.264 standard is the main traditional baseline against which the bandwidth reduction is measured."},{"cited_title":"Swin transformer v2: scaling up capacity and resolution,","cited_arxiv_id":null,"evidence_quote":"Swin transformer v2 is the backbone of the Deep JSCC encoder and decoder in APVST."}],"review_version":1}