{"id":"94a34ad4-a142-4e6f-95c2-e951773ded55","arxiv_id":"2506.22702","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Connected-RIS groups RIS elements whose phase shifts stay within a threshold across steering angles, sharing control signals to cut power and hardware complexity.","lead":"This paper proposes a reconfigurable intelligent surface (RIS) design, called Connected-RIS, where groups of elements with correlated phase shifts share a single control signal. The authors report 86-92% power savings and 83-98% fewer control lines versus fully controlled RIS, while claiming similar beamforming gain and data rate.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Power-model inconsistency in §V-B makes the 86-92% savings figure unverifiable; Table II contradicts its own stated parameters.","rationale":"We focused on the power model because the abstract's central claim is a quantitative savings figure. The reader's weakest_assumption also flagged the elevation restriction; we checked the paper and it does state in Section IV that θBS, ϕBS, θUE are fixed, so the elevation issue is a stated limitation rather than a hidden assumption. The power model, by contrast, is demonstrably inconsistent: the reported Table II cannot be reproduced from the stated equations and parameters. For example, the min-gain RIS in Case 1 has 3249 elements; at 15 mW per unit cell the unit-cell power alone is 48.735 W, which exceeds the reported total of 43.815 W by ~5 W. This is not a subtle modeling choice; it is a numerical contradiction. Similarly, Eq. (17) predicts P_circuit ≈ Pdrive for any fully controlled RIS, contradicting the per-element shift-register description (4-bit 74LS194 at 75 mW per control chain) and the cited measurement [9] showing control power scales with element count. The Connected-RIS rows require P_circuit ≈ 0, which is below the floor set by Eq. (16). We therefore cannot verify the 86-92% savings. We are not claiming the architectural idea is wrong—only that this paper's quantitative support is not credible. A major revision with a corrected, reproducible power model, and a clear statement of the elevation scope, could change the verdict.","tokens_in":15976,"tokens_out":26298,"duration_ms":280495,"concrete_test":"Using the paper's stated constants (Pcontrol=4.8 W, Punit=15 mW, Pdrive=75 mW, N and Nz from Table I and §V-B), reconstruct every entry in Table II from Eqs. (15)-(17). If the min-gain row for Deployment Case 1 (57×57 elements) is not reproducible at 43.815 W, or if no single assignment of Nc/Ns makes all six connected-RIS and nine benchmark entries consistent, then the quantitative power-savings claim is not grounded in the model. This can be done by hand in minutes.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline quantitative claim (86-92% power savings) rests on the power model and Table II. That model is internally inconsistent. Equation (17) gives P_circuit = ceil(Nc/(N Ns))Pdrive; for any large N this is a constant (≈Pdrive), so fully-controlled power is dominated by Punits = N Punit. In Deployment Case 1 the fully controlled 3dB-RIS has N=4900 and Punit=15 mW, giving Punits=73.5 W; adding Pcontrol=4.8 W and Pdrive≈75 mW yields the reported 78.41 W. But the same Table lists the min-gain RIS at 43.815 W even though that configuration has N=3249 elements, so Punits alone is 48.735 W and the total cannot be below 53.535 W. Thus the table is not derived from the stated parameters. Moreover, for the Connected-RIS, the paper sets Nunits=Nz and effectively neglects P_circuit (to reach 5.86 W the drive term would be ≈0.01 W, below Pdrive). No consistent interpretation of Eqs. (16)-(17) reproduces all three rows of Table II. Since the 86-92% figure is computed from these numbers, the central claim is not supported. Whether a corrected model would still yield large savings is unknown; the paper as written does not provide it.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Connected-RIS, a reconfigurable intelligent surface in which elements whose required phase shifts remain within a threshold across all azimuth steering directions in [-80 degrees, 80 degrees] are connected and share one control signal. It derives a fair-coverage RIS sizing rule using a 3GPP path-loss model, performs a correlation analysis over azimuth angles in three deployment cases, and compares the Connected-RIS with fully controlled min-gain and gain-margin RIS baselines in terms of number of loads, control lines, power consumption, gain, and data rate. The headline claims are reductions of 86-92% in power and 83-98% in control signals while maintaining sufficient gain for fair coverage.","tokens_in":16305,"tokens_out":12554,"duration_ms":134434,"significance":"The architectural idea is timely and, if properly validated, could be useful for reducing RIS control complexity. The paper's strength is its concrete design flow: a coverage-based sizing method, a clear grouping criterion, and comparisons against two reasonable baselines. However, the main quantitative results currently rest on an unvalidated and internally inconsistent power model and on threshold values fitted to the data, so the claimed savings are not yet established. The fixed-elevation assumption also needs to be reconciled with the claim of three-dimensional beamforming.","major_comments":[{"comment":"The power-consumption model and Table II are internally inconsistent. For Deployment Case 1, the min-gain RIS has 57x57=3249 elements; with the stated Punit=15 mW, Punits alone is 48.735 W. Adding Pcontrol=4.8 W and the smallest possible Pcircuit from Eq. (17), namely Pdrive=75 mW, gives at least 53.61 W, but Table II reports 43.815 W. The Connected-RIS rows are also inconsistent with the stated 70/83 column counts: the reported 5.86 W and 6.045 W imply Nunits of about 66 and 78 under the stated Punit and Pcontrol values, not 70 and 83. More fundamentally, Eqs. (16)-(17) yield Pcircuit approximately equal to Pdrive for a large array regardless of N, so they do not capture the dependence of drive power on the number of control signals that the paper's own motivation in Refs. [9], [20], [21] emphasizes. Because the abstract's 86-92% savings figure is computed from Table II, the paper's central quantitative claim is not supported by the model as presented.","section":"V-B, Table II, Eqs. (16)-(17)"},{"comment":"The threshold psi_th is chosen per deployment case to maximize the number of correlated columns, e.g., psi_th=31 degrees for Case 1 with DeltaG=3 dB and psi_th=30 degrees for Case 1 with DeltaG=6 dB, with analogous choices for Cases 2 and 3. The reported control-line and power reductions are therefore obtained at threshold values fitted to the same data used to compute the savings. No a priori criterion for selecting psi_th and no sensitivity analysis around the chosen values are provided, so the 83-98% control-signal reduction is not established as a robust design prediction.","section":"IV-A, Fig. 6"},{"comment":"The entire grouping procedure assumes a fixed elevation angle theta_UE. Because columns are grouped as single control signals, any variation in elevation changes the vertical phase gradient across a column and can make the within-column phase differences exceed psi_th. The paper does not evaluate this case, so the abstract's claim of three-dimensional passive beamforming is not supported by the fixed-elevation analysis; either the design's scope should be explicitly limited to fixed-elevation deployments or a variable-elevation validation is needed.","section":"IV-A, V-C"},{"comment":"The model sets Nunits=Nz for the Connected-RIS, i.e., it counts unit-cell power only for the first row of each column. Grouping control lines does not by itself remove the PIN diodes or bias circuits of the remaining N-Nz elements unless a specific circuit topology is described in which an entire column is driven by a single load. Without such a topology, the Connected-RIS unit-cell power is undercounted by a factor of approximately N/Nz, which directly affects the claimed savings.","section":"V-B, Eq. (15), definition of Nunits"}],"minor_comments":[{"comment":"The symbol N_z is defined as the number of rows in Section II but as the number of columns/groups in Section V-A; please use a distinct symbol for the number of groups.","section":"Section II vs. Section V-A"},{"comment":"The use of N is inconsistent: Eq. (13) defines N as the square root of the linear gain, while the text speaks of \"N^2 >= 4799 elements\" and then reports sizes such as 70x70; please clarify whether N is the number of elements per side or the total number of elements.","section":"Eq. (13) and Section III-B"},{"comment":"The update rule \"phi_{q+1} = phi_q + DeltaPhi/2\" introduces an unexplained factor of one-half; please verify whether this factor is intended.","section":"Algorithm 1, line 13"},{"comment":"The threshold comparison should explicitly wrap the phase differences to [0, 2*pi) before comparison with psi_th, otherwise aliasing at the 2*pi boundary can produce false correlations.","section":"Algorithm 2"},{"comment":"The axis tick labels in the preprint appear corrupted, e.g., \"0 1 02 03 04 05 06 07 08 09 0\"; the published version should use the original uncorrupted figures.","section":"Figures 6, 8, 9"}],"recommendation":"major_revision","confidential_remarks":"The power-model inconsistency in Table II is the most serious issue: the headline numbers cannot be reproduced from the stated parameters. This is fixable in principle, but the revision must provide a consistent power model and either reproduce Table II or replace it. The threshold-fitting issue also needs a principled treatment. If the authors cannot resolve these points, the paper should not be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper's real contribution is a concrete column-grouping scheme for RIS control based on phase-shift correlation across steering angles, plus a coverage-based minimum-size methodology. The beamforming and rate results are plausible and externally validated with a standard channel model. But the headline power savings are built on a power model that does not add up. Table II contradicts the paper's own parameters: the min-gain RIS in Case 1 has 3249 elements, each at 15 mW, so unit-cell power alone is 48.7 W, yet the table reports 43.8 W total. No consistent reading of Eqs. (16)-(17) reproduces all three rows. That is a load-bearing flaw: the 86-92% figure is computed from these numbers.\n\nWhat is genuinely useful: the insight that, for a fixed elevation angle, the phase profile across each RIS column is nearly constant across a wide azimuth range, so one control line per column suffices. That is stated up front (fixed θUE in Section IV), though the implications for variable elevation are not discussed. The correlation analysis itself is a reasonable engineering heuristic, and the number-of-control-lines reduction is real by construction. The authors also deserve credit for comparing against both a same-size fully controlled RIS and a minimum-gain RIS, and for including a rate analysis showing no major loss.\n\nThe soft spots, in order: (1) the power model is unverified and internally inconsistent; (2) the correlation threshold ψth is chosen to maximize grouping, so the savings are partly by construction, and the paper does not show robustness to other thresholds; (3) the free gain margin ΔG^dB is set to 3 or 6 dB but its role in fair coverage is not itself justified. These are fixable: a correct power model, a sensitivity sweep over ψth, and an explicit statement of the elevation range would do it.\n\nVerdict: not publishable as is, but not a throwaway. The architecture idea is worth pursuing, and the paper deserves a serious referee. I would send it back for major revision rather than desk reject, and would want to see the power analysis redone before trusting any of the quantitative claims.","headline":"The Connected-RIS grouping idea has merit, but the power model is internally inconsistent; the 86-92% savings figure is not supported.","tokens_in":16748,"tokens_out":2860,"would_cite":false,"duration_ms":29128,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Grouping correlated RIS elements into shared control lines cuts power use by up to 92% while preserving coverage.","keywords":["Reconfigurable intelligent surface","Connected-RIS","correlation analysis","passive beamforming","power consumption","control circuitry","fair coverage","3GPP channel model"],"falsifier":"Build or simulate a Connected-RIS prototype with column-shared controls and measure the beam pattern while the user's elevation angle moves away from the design value; if the gain falls by more than the 3–6 dB margin used to size the surface, the shared-column premise fails. Alternatively, recompute the panel power using the per-element supporting-power figure of about 3.7 mW cited from [9] instead of the per-column unit power; if the saving drops below the claimed 86–92%, the reported reduction is an artifact of the power model.","tokens_in":15754,"feed_emoji":"⚡","tokens_out":10877,"duration_ms":103695,"temperature":0.7,"pith_summary":"The paper claims that a reconfigurable intelligent surface (RIS) can be built with far fewer control signals if elements whose phase shifts are strongly correlated across all steering directions are connected to a common control line. By analyzing the phase-shift matrices needed to steer a beam across an azimuth range of $-80^\\circ$ to $80^\\circ$, the authors find that correlations align along the columns of the array, so an $83\\times 83$ surface only needs 83 control lines instead of 6,889. On this basis, their Connected-RIS reduces power consumption by 86–92% and control signals by 83–98% relative to fully controlled RIS designs, while maintaining gain above their fair-coverage threshold and delivering nearly the same data rates. If this holds, the design would make large RIS apertures practical for energy-constrained and large-scale wireless deployments.","feed_headline":"Shared control lines cut RIS power use by 92%","feed_subtitle":"Grouping correlated elements into shared control lines preserves beamforming gain while slashing hardware.","key_machinery":"The load-bearing mechanism is the correlation test applied to the phase-shift matrices. For each steering direction, Algorithm 1 generates a matrix $\\Psi_q$ from the channel phases; Algorithm 2 then computes the phase difference for every element pair across all $Q$ matrices, and a pair is marked correlated if the difference never exceeds the threshold $\\psi_{\\rm th}$ (about 28–31° in the simulations). Because the correlated pairs turn out to be the elements sharing a column, the design collapses the control structure: one load impedance and one DC control line per column, instead of per element. The same grouping also redefines the unit count in the power model, which is the source of the reported 86–92% saving.","core_discovery":"The central claim is that the phase-shift values needed to steer a RIS across the azimuth coverage region $-80^\\circ$ to $80^\\circ$ are correlated along columns: for each column, the phase difference between any two elements remains below a threshold $\\psi_{\\rm th}$ for every steering direction in that range. The paper defines two elements as correlated when this condition holds across all the phase-shift matrices $\\{\\Psi_q\\}$ generated for the coverage sector, and the correlation analysis for three deployment cases (RIS near the base station, midway, and near the user) shows the correlated clusters are entire columns. This leads to the Connected-RIS architecture, where only the first row of elements is independently controlled and each column is driven by one shared control signal. The paper shows that this cuts the number of load impedances from $N^2$ to $N_z$ — for instance, from 6,889 to 83 for an $83\\times 83$ surface — and, under its power model that charges per controlled unit rather than per physical element, reduces panel power from about 78–108 W to about 6 W across the deployment cases, with the RIS gain remaining sufficient for fair coverage and the achievable rate practically unchanged.","pith_inferences":["The column-correlation result is established for fixed elevation angles; if user elevation varies, the vertical phase gradient changes across a column, so the shared-control grouping would need to be re-derived for each elevation slice or abandoned.","The magnitude of the power saving is largely determined by the paper's decision to count one unit per column rather than per element; if per-element supporting circuitry (about 3.7 mW per element in the cited measurements) were charged, the saving would shrink toward the control-line reduction only.","A natural extension is to treat the correlation threshold $\\psi_{\\rm th}$ as a tunable design parameter that trades a small gain loss for even coarser grouping, instead of fixing it near 30° by simulation."],"forward_implications":["An $83\\times 83$ Connected-RIS can cover the whole $160^\\circ$ sector with 52 codewords, versus 80 for the min-gain fully controlled RIS, and the codeword storage shrinks to about 12,948 bits from 1,075,684 bits.","In the paper's case study, both a dynamically reconfigured panel and a fixed 52-panel layout consume about 314 W, compared with 5,629 W for the 6-dB fully controlled RIS and 2,279 W for the min-gain RIS.","Control signals drop by 83–98% across the deployment cases, so the DC wiring, FPGAs, and interfaces scale with the number of columns $N_z$ rather than the element count $N^2$, easing large-aperture implementation.","The fair-coverage methodology (equating the RIS-assisted received power to the direct BS-UE link, plus a 3 or 6 dB margin) gives a direct way to size the minimum number of elements for any deployment before applying the correlation grouping."],"supporting_citations":[{"why":"Supplies the phase-shift formula used to build every steering matrix.","marker":"[19]"},{"why":"Supplies the per-element PIN-diode power and the 3.7 mW per-element supporting-component figure that motivates reducing the number of control paths.","marker":"[9]"},{"why":"Provides the control-circuit power model and the 300 mW per configuration-update cost used in the case-study power comparison.","marker":"[10]"},{"why":"Defines the sub-connected (grouped) RIS architecture that the Connected-RIS extends with correlation-based grouping.","marker":"[14]"},{"why":"Provides the static power model and the FPGA board (4.8 W) used for the constant control power in Eqs. (15)-(17).","marker":"[20]"},{"why":"Establishes the unit-cell and drive-circuit power components on which the paper's power model is built.","marker":"[21]"},{"why":"Defines the 3GPP path-loss and Rician channel model used to compute the fair-coverage RIS sizes for the three deployment cases.","marker":"[16]"}],"fun_headline_variants":["Shared control lines cut RIS power use by 92%","Connected-RIS: 92% less power, 98% fewer control signals","RIS power use drops 92% with correlated element sharing","Correlated columns share controls, cut RIS power 92%","One control per column: RIS power drops 92%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole saving rests on the assumption that every element in a column can be driven by a single shared control signal without changing the beam, because their phase-shift differences stay below a fixed threshold for every steering direction in the coverage range — a property the paper verifies only for a fixed elevation angle.","fun_headline_variants_meta":{"raw":{"variants":["Shared control lines cut RIS power use by 92%","Connected-RIS: 92% less power, 98% fewer control signals","RIS power use drops 92% with correlated element sharing","Correlated columns share controls, cut RIS power 92%","One control per column: RIS power drops 92%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001338,"raw_usage":{"total_tokens":5481,"prompt_tokens":1031,"completion_tokens":4450,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":4363}},"tokens_in":647,"tokens_out":4450,"duration_ms":35521,"temperature":1.0,"reasoning_tokens":4363,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:00:52.580422+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build or simulate a Connected-RIS prototype with column-shared controls and measure the beam pattern while the user's elevation angle moves away from the design value; if the gain falls by more than the 3–6 dB margin used to size the surface, the shared-column premise fails. Alternatively, recompute the panel power using the per-element supporting-power figure of about 3.7 mW cited from [9] instead of the per-column unit power; if the saving drops below the claimed 86–92%, the reported reduction is an artifact of the power model.","supporting_citations":[{"cited_title":"Wireless communications through reconfigurable intel- ligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Supplies the phase-shift formula used to build every steering matrix."},{"cited_title":"Power consumption analysis of a reconfigurable intelligent surface for self-sustained operations,","cited_arxiv_id":null,"evidence_quote":"Supplies the per-element PIN-diode power and the 3.7 mW per-element supporting-component figure that motivates reducing the number of control paths."},{"cited_title":"Enhancing Energy Efficiency for Reconfigurable Intelligent Surfaces with Practical Power Models","cited_arxiv_id":"2310.15901","evidence_quote":"Provides the control-circuit power model and the 300 mW per configuration-update cost used in the case-study power comparison."},{"cited_title":"Active reconfigurable intelligent surface: Fully-connected or sub-connected?","cited_arxiv_id":null,"evidence_quote":"Defines the sub-connected (grouped) RIS architecture that the Connected-RIS extends with correlation-based grouping."},{"cited_title":"Static power consumption modeling and measurement of reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Provides the static power model and the FPGA board (4.8 W) used for the constant control power in Eqs. (15)-(17)."},{"cited_title":"Reconfigurable intelligent surface: Power consumption modeling and practical measurement validation,","cited_arxiv_id":null,"evidence_quote":"Establishes the unit-cell and drive-circuit power components on which the paper's power model is built."},{"cited_title":"5G: study on channel model for frequencies from 0.5 to 100 GHz (3GPP TR 38.901 version 14.3.0 release 14),","cited_arxiv_id":null,"evidence_quote":"Defines the 3GPP path-loss and Rician channel model used to compute the fair-coverage RIS sizes for the three deployment cases."}],"review_version":1}