{"id":"54362aa9-01f1-4419-9560-1afe5cff9a60","arxiv_id":"2506.22903","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey that organizes limited CSI feedback for RIS-assisted wireless communications around channel reconstruction and RIS configuration, highlighting structured sparsity and other channel features.","lead":"This paper reviews how limited feedback of channel state information works in wireless systems that use reconfigurable intelligent surfaces (RISs). It groups existing research into two use cases, channel reconstruction and RIS configuration, and outlines key channel features and open problems.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Structured-sparsity and DL-based guidelines are the load-bearing core of the survey; Secs. 3.4, 3.5, 5.2, and 5.5 concede they degrade outside sparse far-field channels and lack measured-data validation, so the central claim is a conditional design rule, not a universal one.","rationale":"The paper is a survey, so the central claim is a synthesis rather than a theorem. The claim would be secure if the three named features (position-dependent fluctuation, high-dimensional sub-channel matrix, structured sparsity) were robust across the use cases and if the surveyed schemes demonstrated overhead reduction under those features. The high-dimensional sub-channel matrix is uncontroversial. The structured-sparsity pillar, however, is explicitly conditional: Sec. 3.4 assumes sparse scatterers around BS and RIS; Sec. 3.5 says feedback overhead in [21]/[22] grows with path count; Sec. 5.2 says near-field breaks far-field codebooks; Sec. 5.5 says DL-based feedback has only been tested in simulation. The manuscript itself contains these limitations, so it is not internally inconsistent. The concern is external validity: the abstract presents the features as 'unique' and as 'guidelines' without guarding the conditions, and the paper gives no quantitative robustness sweep (e.g., varying scattering richness, user/RIS mobility, bandwidth, near-field distance). This does not require rejection; a conditional acceptance with an explicit scope statement would match the evidence. The reader's weakest assumption already captured this, so the verdict is unchanged.","tokens_in":12475,"tokens_out":5463,"duration_ms":67090,"concrete_test":"Use QuaDRiGa or an equivalent ray tracer to generate an RIS-assisted FDD channel at 3.5 GHz in a rich-scattering MUSR/MUMR topology (e.g., 20 clusters, 32 RIS elements, 8 users), then implement the structured-sparsity feedback of [21]/[22] and the adaptive codebook of [27] on that channel. Measure the fraction of non-zero beamspace columns shared across users and the feedback-overhead/NMSE trade-off, and compare against the sparse mmWave channel used in the original papers. If the shared-support fraction drops below roughly 80%, or if the feedback overhead needed to reach a target reconstruction NMSE exceeds brute-force CSI feedback, then the central design guideline fails outside the sparse-scatterer regime, directly testing the enabling assumption of Sec. 3.4.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The survey's central claim—that three RIS-induced channel features should drive limited-feedback design—stands only where those features actually persist. The strongest support, structured-sparsity feedback (Sec. 3.4), explicitly assumes 'where both the BS and RIS are bounded by sparse scatterers' and a shared, slowly varying BS-RIS channel; Sec. 3.5 then concedes that the feedback overhead of [21]/[22] grows with the number of paths. Sec. 5.2 concedes that near-field channels break far-field DFT/beamspace assumptions and that existing schemes suffer 'severe performance degradation.' Sec. 5.5 concedes that every AI-based feedback scheme has been evaluated only on simulated data, with no measured CSI samples. The paper also asserts in Sec. 2.1.2 that multi-reflection paths are negligible without giving a quantitative regime; if that fails, the cascaded channel dimension and feedback structure change. None of these concessions is fatal by itself, but together they mean that the 'distilled' features are conditional guidelines, not universal properties of RIS-assisted channels. The manuscript provides no robustness sweep over scattering richness, mobility, bandwidth, or near-field distance, so the reader cannot tell how much of the claimed overhead reduction survives outside the specific scenarios where each cited method was validated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a survey of limited feedback design in RIS-assisted wireless systems. It proposes that RIS-specific channel features—position-dependent channel fluctuation, ultra-high-dimensional sub-channel matrices, and structured sparsity—should serve as guidelines for feedback design, and it classifies feedback into two main use cases: channel reconstruction (Sec. II-A) and RIS configuration (Sec. II-B). The survey reviews codebook-based, DL-based, channel-customization, and structured-sparsity methods for channel reconstruction (Sec. III), then describes three feedback protocols for RIS configuration: user-to-BS, BS-to-RIS, and user-to-RIS (Sec. IV), and closes with future directions (Sec. V). The contribution is organizational and tutorial rather than a new technical result.","tokens_in":12728,"tokens_out":6310,"duration_ms":65548,"significance":"The survey is timely and clearly organized; the SUSR/SUMR/MUSR/MUMR breakdown and the two-category taxonomy provide a useful framework for researchers entering the area. The paper is honest about the conditional validity of the design rules: Sec. 3.5 states that sparsity-based gains diminish as the number of paths grows, Sec. 5.2 warns that near-field channels break far-field assumptions, and Sec. 5.5 concedes that all AI-based feedback schemes have been evaluated only on simulated data. These caveats partially answer the concern that the 'distilled features' are being over-generalized. The main value of the paper lies in the synthesis of recent results; there are no machine-checked proofs, reproducible code, or new experimental data, so the assessment rests on the accuracy of the literature descriptions, which appear faithful to the cited works.","major_comments":[],"minor_comments":[{"comment":"The abstract lists three channel features, but the body also treats time correlation (Sec. 3.2, Sec. 4.1) and angle-dependent phase shifts (Sec. 2.2.3) as design drivers; the abstract should either add these or state explicitly that the three listed features are a representative subset.","section":"Abstract / Sec. III"},{"comment":"Because the structured-sparsity design rule applies only under sparse-scatterer and far-field conditions, and the near-field discussion in Sec. 5.2 concedes severe performance degradation, the paper should collect the applicability conditions of each design guideline in one place, such as a short paragraph at the end of Sec. III, to prevent readers from treating the features as universal.","section":"Sec. 3.4 / Sec. 5.2"},{"comment":"A systematic comparison table summarizing assumptions, feedback overhead, computational complexity, and applicable scenario for the methods in Refs. [21], [22], [27], [33], [34], and [29] would materially improve the survey; the prose comparison in Sec. 3.5 is useful but does not allow readers to weigh the methods side by side.","section":"Sec. 3.5"},{"comment":"The term 'position-dependent channel fluctuation' is not formally defined; the body discusses how RIS placement and mobility affect channel dynamics (Secs. 2.1.1 and 2.1.2), but the abstract's phrase should be tied to a specific definition or equation.","section":"Abstract / Sec. 2.1"},{"comment":"The claim that multi-reflection channels between RISs are negligible is stated without a quantitative path-loss comparison; adding a short expression for the multiplicative path-loss penalty or citing a study that quantifies the regime would make the assumption easier to evaluate.","section":"Sec. 2.1.2"},{"comment":"The caption of Fig. 1 lists scenario abbreviations but does not explain the row/column structure or the meaning of the arrows; please expand the caption to make the figure self-contained.","section":"Fig. 1"},{"comment":"There is a typo: 'near-filed' appears twice and should be 'near-field'.","section":"Sec. 5.2"},{"comment":"Ref. [43] is an arXiv preprint; if a peer-reviewed version has appeared, the published version should be cited instead.","section":"References"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is within the scope of a communications magazine. A notable share of the key cited schemes are from the authors' own group (e.g., Refs. [26], [27], [33], [34], [38]); this is not improper, but an explicit statement of literature-selection criteria would help the reader judge whether the taxonomy is biased toward the group's own line of work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a competent, readable survey of limited feedback in RIS systems, not a research contribution. It earns its place as an organizational reference, but the advertised 'distilled channel features' are conditional design heuristics, and the paper leans heavily on the authors' own prior papers.\n\nWhat's new: a clean two-way split—channel reconstruction vs RIS configuration—and a sensible progression of user-RIS topologies (SUSR/SUMR/MUSR/MUMR). The comparison in Sec. 3.5 is honest: it admits adaptive codebooks degrade in mobility, DL needs retraining, and structured-sparsity overhead grows with path count. The angle-dependent phase-shifter discussion and capacitance-based codebook point (Sec. 2.2.3) is a genuinely useful observation often missing from surveys. The descriptions of cited works look accurate.\n\nSoft spots: the central claim in the abstract overreaches. The three 'unique features' only hold under specific conditions: sparse scatterers around BS and RIS, dominant single-reflection cascades, and a slowly varying shared BS-RIS channel. The paper itself concedes these in Secs. 3.4, 3.5, 5.2, and 5.5, and Sec. 5.5 admits every AI-based scheme has been tested only on simulated data. Sec. 2.1.2 asserts multi-reflection paths are negligible without a quantitative regime; that should be supported or softened. Near-field is waved away as 'severe degradation' without a robustness sweep. The self-citation concentration ([14], [26], [27], [33], [34], [39], [43]) is visible, but the cited works are the actual primary sources, so it is not by itself disqualifying. The manuscript is also not final: placeholder dates ('Received: XX XX, 202X') and typos like 'near-filed' suggest another editing pass.\n\nBottom line: as a review article, it is useful for groups entering RIS feedback, and a careful referee could turn it into a solid reference. The abstract and conclusions should be tempered to present guidelines as regime-dependent, and the future-directions section should include a robustness discussion. It deserves peer review, not desk rejection, but with requests for revision. I wouldn't cite it over the primary sources in my own work, but I'd point students to it.","headline":"A competent survey with a useful taxonomy, but its 'distilled features' are regime-dependent heuristics and it leans heavily on the authors' own prior work.","tokens_in":13148,"tokens_out":2551,"would_cite":false,"duration_ms":25217,"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":"Limited feedback in RIS networks should be designed around three channel features the RIS itself creates, not legacy CSI.","keywords":["limited feedback","reconfigurable intelligent surface","channel state information","structured sparsity","beamspace channel","codebook design","deep learning","FDD systems"],"falsifier":"Measure the non-zero column support of the beamspace cascaded channel across many users in a rich-scattering environment with multiple interacting RISs: if the shared column indices and offset/ratio structure disappear, the structured-sparsity feedback schemes lose their overhead advantage. Equally direct is comparing CSI recovery accuracy under a capacitance-based codebook versus a phase-shift codebook while sweeping the incidence angle at the RIS; the angle-dependent model predicts a growing gap that phase-only feedback cannot close.","tokens_in":12311,"feed_emoji":"📡","tokens_out":4646,"duration_ms":49021,"temperature":0.7,"pith_summary":"This paper argues that limited CSI feedback in RIS-assisted systems fails if it is treated as an extension of conventional massive-MIMO feedback, because the RIS introduces channel properties that legacy schemes ignore: channel fluctuations tied to the RIS position, an ultra-high-dimensional sub-channel matrix at the BS–RIS link, and structured sparsity across users. The authors organize the field into two use cases: feeding back enough information to reconstruct the BS–RIS–user channel at the base station, and feeding back control instructions to configure the RIS itself. From this, they extract design principles: allocate feedback bits according to the sub-channel that dominates, exploit the shared BS–RIS channel so one user can feed back common structure for all, and build codebooks from capacitance values rather than phase shifts because the RIS phase response depends on incidence angle. A sympathetic reader would take the paper as a case that RIS feedback is a distinct problem with its own governing features, and that future schemes should be judged by how well they exploit those features.","feed_headline":"Three RIS channel features should drive limited-feedback design","feed_subtitle":"A review maps position-dependent fluctuation, ultra-high dimension, and structured sparsity into concrete feedback schemes.","key_machinery":"The carrying objects are the four user–RIS topologies (single/multiple user times single/multiple RIS), the cascaded BS–RIS–user channel matrix, and its beamspace projection under DFT matrices. The central identities are single-structured sparsity (different users share the same non-zero column indices of the hybrid-domain cascaded channel because the BS–RIS channel is shared) and triple-structured sparsity (the non-zero columns are identical up to a location offset and cascaded path-gain ratio). A second mechanism is two-timescale feedback, in which the high-dimensional, slowly varying BS–RIS channel is fed back once per large timescale while the low-dimensional, fast-varying RIS–user channel is fed back each small timescale. A third is the angle-dependent phase-shifter model, which makes capacitance the natural codebook dimension for RIS configuration rather than phase shift.","core_discovery":"The paper's central claim is that the design of limited feedback in RIS-assisted FDD systems should be guided by RIS-specific channel features rather than by end-to-end channel models borrowed from conventional systems. It classifies feedback into channel reconstruction and RIS configuration, and distills three features: RIS position-dependent channel fluctuation (which sub-channel is static versus dynamic depends on where the RIS is deployed), the ultra-high-dimensional sub-channel matrix (the BS–RIS and RIS–user links taken together create far more parameters than a standard MIMO channel), and structured sparsity (when the cascaded channel is projected into the beamspace domain, different users share the same non-zero column indices, and the non-zero columns differ only by location offset and path-gain ratio). The paper further claims that RIS configuration feedback must account for active–passive beamforming interplay, the large number of configuration parameters, and the angle-dependent phase-shift hardware, which together push codebook design toward capacitance-based codewords. These features are then mapped onto concrete schemes: adaptive cascaded codebooks, autoencoder and attention-based compression, channel customization that reshapes rich scattering into a few strong paths, shared-structure sparsity feedback, two-timescale deep-learning feedback, and learning-based adaptive RIS control.","pith_inferences":["Editorial extension: a direct test of the angle-dependent model is to measure reflected power under capacitance-based versus phase-based codebooks across incidence angles; the paper's model predicts a growing performance gap that phase-only feedback cannot close.","Editorial extension: the shared-BS–RIS-channel structure suggests a natural collaborative or hierarchical feedback design where the network aggregates the common channel component across users, a direction the review gestures at but does not formalize.","Editorial extension: if multi-reflection paths are not negligible in dense indoor or wideband deployments, the factorization of feedback overhead that supports the structured-sparsity schemes collapses, implying the design guidelines implicitly target sparse outdoor scenarios.","Editorial extension: the suggested integration of multi-modal information could be recast as moving from channel reconstruction to scene-level feedback, potentially reducing feedback to a few high-level descriptors rather than channel parameters."],"forward_implications":["If structured sparsity holds, the shared BS–RIS channel lets one user transmit the common non-zero column indices, cutting multi-user feedback overhead roughly to the per-user RIS–user part.","Adaptive cascaded codebooks that allocate bits by path importance beat fixed RVQ codebooks as path count grows, since the path-gain vector dimension rises with the number of paths.","Deep-learning two-timescale feedback (exemplified by RIS-CsiNet) produces beamforming vectors and RIS phase shifts directly from compressed bitstreams, avoiding full CSI recovery and reducing both overhead and computation.","Because the RIS phase response depends on incidence angle, codebooks built from capacitance values are necessary; phase-only codebooks cannot be mapped to correct element settings.","Near-field and active-RIS architectures invalidate far-field feedback assumptions, so new codebooks and feedback protocols are needed."],"supporting_citations":[{"why":"Establishes single-structured sparsity in the hybrid-domain cascaded channel and the dimension-reduced feedback scheme that exploits it.","marker":"[21]"},{"why":"Identifies triple-structured sparsity with shared location offsets and cascaded path-gain ratios across users.","marker":"[22]"},{"why":"Supplies the adaptive cascaded codebook with dynamic bit partitioning that serves as the codebook-based channel-reconstruction baseline.","marker":"[27]"},{"why":"Introduces channel customization that reshapes the composite channel into sparse orthogonal paths, enabling SVD transceiver design with limited feedback.","marker":"[33]"},{"why":"Provides the RIS-CsiNet two-timescale deep-learning framework that directly generates beamforming vectors and RIS phase shifts from compressed CSI.","marker":"[34]"},{"why":"Grounds the capacitance-based codebook design under angle-dependent phase response with a learning-based adaptive control protocol.","marker":"[35]"},{"why":"Supports the user-to-RIS configuration path through convolutional autoencoder-based phase shift feedback compression.","marker":"[29]"},{"why":"Supplies the angle-dependent phase shifter model that motivates capacitance-based codebooks for RIS configuration.","marker":"[38]"}],"fun_headline_variants":["Three channel features should drive RIS feedback design","Position, dimension, sparsity: the triple key to RIS feedback","RIS feedback: three channel traits anchor codebook design","Let three RIS channel features choose your feedback scheme","RIS feedback design hinges on position, dimension, sparsity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The design guidelines assume that the cascaded BS–RIS–user channel is well approximated by single-reflection paths with a shared, slowly varying BS–RIS channel and sparse beamspace structure; if deployments have rich scattering, fast-moving RISs, or near-field conditions, the overhead reductions are not guaranteed.","fun_headline_variants_meta":{"raw":{"variants":["Three channel features should drive RIS feedback design","Position, dimension, sparsity: the triple key to RIS feedback","RIS feedback: three channel traits anchor codebook design","Let three RIS channel features choose your feedback scheme","RIS feedback design hinges on position, dimension, sparsity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0007,"raw_usage":{"total_tokens":3164,"prompt_tokens":954,"completion_tokens":2210,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":570,"completion_tokens_details":{"reasoning_tokens":2133}},"tokens_in":570,"tokens_out":2210,"duration_ms":17790,"temperature":1.0,"reasoning_tokens":2133,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:54:22.778978+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the non-zero column support of the beamspace cascaded channel across many users in a rich-scattering environment with multiple interacting RISs: if the shared column indices and offset/ratio structure disappear, the structured-sparsity feedback schemes lose their overhead advantage. Equally direct is comparing CSI recovery accuracy under a capacitance-based codebook versus a phase-shift codebook while sweeping the incidence angle at the RIS; the angle-dependent model predicts a growing gap that phase-only feedback cannot close.","supporting_citations":[{"cited_title":"Dimension reduced chan- nel feedback for reconfigurable intelligent sur- face aided wireless communications[J]","cited_arxiv_id":null,"evidence_quote":"Establishes single-structured sparsity in the hybrid-domain cascaded channel and the dimension-reduced feedback scheme that exploits it."},{"cited_title":"Triple-structured sparsity-based channel feedback for RIS-assisted MU-MIMO system[J]","cited_arxiv_id":null,"evidence_quote":"Identifies triple-structured sparsity with shared location offsets and cascaded path-gain ratios across users."},{"cited_title":"Adaptive bit partitioning for reconfigurable intelligent sur- face assisted FDD systems with limited feedback [J]","cited_arxiv_id":null,"evidence_quote":"Supplies the adaptive cascaded codebook with dynamic bit partitioning that serves as the codebook-based channel-reconstruction baseline."},{"cited_title":"Channel cus- tomization for limited feedback in RIS-assisted FDD systems[J]","cited_arxiv_id":null,"evidence_quote":"Introduces channel customization that reshapes the composite channel into sparse orthogonal paths, enabling SVD transceiver design with limited feedback."},{"cited_title":"Deep learning-based two-timescale CSI feedback for beamforming design in RIS-assisted communi- cations[J]","cited_arxiv_id":null,"evidence_quote":"Provides the RIS-CsiNet two-timescale deep-learning framework that directly generates beamforming vectors and RIS phase shifts from compressed CSI."},{"cited_title":"Learning-based adaptive IRS control with limited feedback codebooks[J]","cited_arxiv_id":null,"evidence_quote":"Grounds the capacitance-based codebook design under angle-dependent phase response with a learning-based adaptive control protocol."},{"cited_title":"Convolutional autoencoder-based phase shift feedback com- pression for intelligent reflecting surface-assisted wireless systems[J]","cited_arxiv_id":null,"evidence_quote":"Supports the user-to-RIS configuration path through convolutional autoencoder-based phase shift feedback compression."},{"cited_title":"Angle- dependent phase shifter model for reconfigurable China Communications 13 intelligent surfaces: Does the angle-reciprocity hold?[J]","cited_arxiv_id":null,"evidence_quote":"Supplies the angle-dependent phase shifter model that motivates capacitance-based codebooks for RIS configuration."}],"review_version":1}