{"id":"b6ba009c-76fe-4d77-944b-c1dc05a1255c","arxiv_id":"2505.01076","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A quasi-static IRS with manually or mechanically tuned passive elements is optimized via alternating DC-SCA to produce shaped 3D coverage beams with near-joint-optimization gain at lower complexity.","lead":"This paper proposes a quasi-static intelligent reflecting surface (QS-IRS) architecture with mechanically or manually tuned passive elements, and an alternating optimization algorithm that shapes 3D beams for area coverage while considering element radiation patterns. A smart generalist would read it to see whether low-cost, large-scale passive surfaces could replace expensive electronically controlled IRSs in long-term coverage applications.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sec. V-A claims adding sidelobe constraint (13) raises mainlobe gain; since (13) only shrinks the feasible set of (P1), that comparison must be a different baseline or an error.","rationale":"The reader's verdict is CONDITIONAL and I agree with that disposition: the paper's AO algorithm and its comparison to joint optimization in Table I are plausible, and the non-rectangular beam examples are interesting, but the text needs correction. My most load-bearing concern is not the ERP model identified in the reader's weakest_assumption, but the mathematically impossible claim that adding a constraint improves a maximum. The reader did flag this comparison as questionable in the rationale, but did not make it the central weakness. I regard it as more load-bearing because it is an internal inconsistency: no modeling assumption or channel detail can rescue the claim that (13) increases the maximum of 10 log10 ρ over a smaller feasible set. The comparison in Fig. 4 must be comparing different algorithms or a different objective, and the paper's causal attribution to the constraint is unsupported. This does not necessarily invalidate the AO algorithm's headline capability, which is why I would keep the CONDITIONAL verdict rather than reject, but the condition should explicitly require the authors to correct or re-label the Sec. V-A comparison and the related sentence in Sec. I. The ERP concern remains a legitimate practical limitation, but it is a modeling-accuracy issue rather than a direct contradiction within the paper's own optimization formulation.","tokens_in":10961,"tokens_out":8517,"duration_ms":94954,"concrete_test":"Run the same AO/DC-SCA solver on Case 1 with constraint (13) removed, keeping (12), (14), and all parameters identical. Since the feasible set expands, the optimal 10 log10 ρ must be at least as large as the constrained value in Fig. 4. If the unconstrained result is lower than the reported 'w/ sidelobe' value, then the original 'w/o sidelobe' curve came from a different algorithm or objective and must be re-labeled or recomputed. As a cross-check, feed the optimized constrained phase vector into the unconstrained objective; its minimum mainlobe gain is a lower bound on the unconstrained optimum, and the unconstrained optimum must reach or exceed that value.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is the claim in Sec. I and Sec. V-A that enforcing the sidelobe constraints (13) improves mainlobe power gain over the case without (13). In (P1), 10 log10 ρ is maximized subject to the mainlobe lower bounds (12) and the constant-modulus constraint (14). Adding the sidelobe upper bounds (13) can only remove feasible points: if F0 is the feasible set without (13) and F1 = F0 ∩ {(13)}, then F1 ⊆ F0, so max over F1 of 10 log10 ρ cannot exceed max over F0 of 10 log10 ρ. Therefore, Fig. 4 showing 'Case 1 w/ sidelobe' above 'Case 1 w/o sidelobe' cannot be an effect of the constraint itself. The likely explanation is that the 'w/o sidelobe' curve is not the optimum of (P1) without (13), but a suboptimal algorithm, possibly the decomposed method of [4]. That conflates solver quality with the value of the constraint. The paper's conclusion that sidelobe constraints 'reduce energy leakage in sidelobes and thus raise energy concentration in the mainlobe' is not established by this comparison. The numerical validation should either solve the unconstrained problem with the same solver, or explicitly label the baseline as a suboptimal benchmark rather than as the effect of adding (13).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a quasi-static IRS (QS-IRS) architecture that tunes passive element phases via mechanical adjustment or manual re-arrangement, targeting low-cost, large-scale deployment for long-term area coverage. To design the QS-IRS beamforming, the authors formulate an optimization problem (P1) that maximizes a common mainlobe power gain subject to shape masks, sidelobe constraints, and constant-modulus phase constraints, while explicitly accounting for the element radiation pattern (ERP). They propose a joint DC-SCA method and an alternating optimization (AO) method that exploits a Kronecker-product separability of the array steering vector and the phase profile. Numerical results show that the AO method achieves power gains close to those of the joint optimization for 4x4 and 8x8 arrays, scales to 48x48 elements in about two hours, can generate non-rectangular and non-flat beam shapes, and exhibits modest loss under phase quantization.","tokens_in":11278,"tokens_out":6636,"duration_ms":73494,"significance":"If the results hold, the QS-IRS concept combined with the low-complexity AO algorithm is a meaningful step toward practical large passive-array beamforming for static coverage applications. The explicit inclusion of the element radiation pattern and flexible shape masks goes beyond prior 1D-decomposed broadbeam designs, and the complexity reduction from O(M^6.5) joint SDP to O(M_y^6.5 + M_z^6.5) per AO step is substantial. The paper also provides useful empirical evidence on gain scaling and quantization loss. However, the central comparative claim about the benefit of sidelobe constraints is not supported by the presented baselines, and the optimality gap for the large-scale regime that motivates the QS-IRS is not quantified. These issues require correction before the main claims can be accepted.","major_comments":[{"comment":"The statement that explicitly enforcing the sidelobe constraints (13) 'achieves gain improvement over the case in [4] without (13)' is unjustified. Adding (13) to (P1) can only shrink the feasible set, so the maximum of 10 log10 ρ under (13) cannot exceed the maximum without (13) for the same problem formulation and solver. The comparison made in Fig. 4 (Case 1 w/ sidelobe vs. Case 1 w/o sidelobe) appears to compare the proposed method (with ERP, shape masks, and AO) against the different algorithm of [4], so the gain difference cannot be attributed to the sidelobe constraint. Please re-run the baseline without (13) using the same problem formulation and solver, or explicitly label the [4]-based curve as a suboptimal benchmark and revise the causal claims in Section IV-A and Section V-A accordingly.","section":"§IV-A and §V-A, Eq. (13), Fig. 4"},{"comment":"The claim that AO achieves power gains 'close to that of the joint optimization algorithm' is only demonstrated for 4x4 and 8x8 arrays in Table I; for 16x16 and larger, the joint-optimization entries are omitted. Since the paper's motivation is large-scale QS-IRS panels (e.g., 48x48), the closeness-to-optimality claim is unsupported in the regime of interest. The O(M^2) power scaling law in Fig. 6 is an empirical fit to the AO curves and is not validated against the joint optimum. Please provide a joint-optimization result at an intermediate size (e.g., 16x16 with a reasonable time limit) or clearly qualify the optimality claim as applying only to small arrays.","section":"§V-B, Table I and Fig. 6"},{"comment":"The AO method restricts the QS-IRS phase vector to w = w_y ⊗ w_z, which is a genuine restriction on the feasible set of (P1). The paper does not quantify the suboptimality of this Kronecker separability assumption for the large arrays that motivate the QS-IRS concept. The small-M comparison in Table I suggests the gap is small for M ≤ 64, but this cannot be extrapolated to M = 48×48 without further evidence. Please add a discussion of the potential performance loss due to separability or provide a numerical comparison for at least one larger case.","section":"§IV-B, Eq. (11), Algorithm 1"}],"minor_comments":[{"comment":"In the definition of r_mz, the z-axis element spacing is written as dy; it should be dz.","section":"Eq. (3)"},{"comment":"The pseudocode sets ζ←0 in the initialization while ζ is also the maximum number of AO iterations from the input; this overwrites the input and makes the loop condition false immediately. Please use a separate iteration counter, e.g., t←0, and retain ζ as the input limit.","section":"Algorithm 1"},{"comment":"The phrase 'to some extend' should be 'to some extent'.","section":"§II"},{"comment":"The caption says 'difference schemes'; it should be 'different schemes'.","section":"Fig. 4 caption"},{"comment":"The convergence claim that the AO objective value is non-decreasing until convergence is stated as a general guarantee, but the cited DC result applies to the DC penalty term, not directly to the alternating maximization of the original nonconvex objective. Please soften the statement to an empirical observation or supply a proof for the AO iteration.","section":"§IV-B"},{"comment":"The penalty weight σ is fixed in the numerical experiments; the sensitivity of the final beamforming gain to σ is not reported. Please include a brief sensitivity study or describe how σ was chosen.","section":"Eqs. (16)-(17), §V"}],"recommendation":"major_revision","confidential_remarks":"The main technical issue is the attribution in §IV-A and §V-A of the gain improvement to the sidelobe constraints; this is fixable by a fair baseline comparison or by explicit re-labeling of the [4] baseline as a different solver. The lack of a joint-optimization baseline for large arrays is also important given the paper's central scalability claim. If the authors can provide a fair comparison and qualify the optimality claims appropriately, the paper could be suitable for publication. The QS-IRS architecture itself builds on existing electromechanical reflectarray hardware, so the novelty is primarily in the system formulation and the scalable AO design rather than in the hardware concept; this should be positioned carefully."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Take on 2505.01076: this is a legitimate, useful extension of the region-oriented IRS beamforming line, with a practical low-cost architecture and a real complexity win. The QS-IRS concept is not brand-new—it's reflectarray hardware plus electromechanical tuning—but the paper's specific contributions are solid: explicit element radiation pattern in the optimization, shape masks that allow non-rectangular and non-flat beam footprints, and an AO algorithm that knocks the per-iteration complexity down from O(M^6.5) to O(My^6.5+Mz^6.5), making 48x48 arrays tractable in about two hours. That matters for anyone thinking about large passive IRS deployment.\n\nThe good parts: the problem (P1) is clean, the DC-SCA/AO derivation is coherent, and the numerical results (Table I, Figs. 3-6) support the main claims. The empirical O(M^2) power scaling is naturally presented as an observation, not a fitted law. Convergence is argued via a cited proposition plus simulation, which is typical for this literature. I also appreciate the DnA assembly discussion; it gives the architecture concreteness.\n\nThe soft spots, in proportion: the headline claim that sidelobe constraints (13) improve mainlobe gain is not supported by the V-A comparison. Adding constraints cannot increase the maximum of (P1); the \"w/o sidelobe\" curve must be a suboptimal solver (likely [4]'s decomposition). The authors should either solve the unconstrained problem with the same AO solver or label that baseline as a suboptimal benchmark. This is fixable, but as written it conflates solver quality with constraint benefit. Minor: no code or data are released, and the ERP model F(Π) = (sinθ cosφ)^{G/2-1} is idealized (no mutual coupling, pattern asymmetry, or fabrication tolerance), which they don't overclaim but it means the shaped beams are simulation-level. Also, the O(M^2) scaling is empirical; fine as a heuristic.\n\nWho this is for: IRS/reflectarray researchers, coverage planning, 6G low-cost deployment. I'd suggest a serious referee, with the requested baseline fix and ideally code release. I wouldn't cite it in my own work until the comparison is corrected, but after that it's citable.\n\nMy recommendation: engage with it, but ask for the revision.","headline":"A useful, incremental IRS coverage paper with a real complexity win and a clean formulation; the main flaw is an apples-to-oranges sidelobe comparison that needs fixing before I'd trust the headline claim.","tokens_in":11791,"tokens_out":2850,"would_cite":false,"duration_ms":28935,"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":"Hand-tuned passive panels shape 3D wireless coverage beams","keywords":["quasi-static IRS","intelligent reflecting surface","shaped beamforming","area coverage","element radiation pattern","alternating optimization","difference-of-convex","phase quantization"],"falsifier":"Fabricate a small QS-IRS panel with the designed phases and measure its far-field 3D gain pattern. If the measured mainlobe gain, sidelobe gap, or beam shape deviates from the simulated profile by more than the reported margins (about 0.4 dB relative to joint optimization, or 2 dB at 2-bit quantization), the central claim that the optimized phases deliver the predicted shaped coverage is refuted.","tokens_in":10769,"feed_emoji":"📡","tokens_out":4825,"duration_ms":45414,"temperature":0.7,"pith_summary":"The paper introduces quasi-static IRS (QS-IRS), a reflective surface whose element phases are set by mechanical adjustment or manual rearrangement instead of electronic control, so panels can be mass-produced from purely passive parts. It claims that a QS-IRS can be optimized to cast a shaped 3D beam over a target area while suppressing sidelobes, with the element radiation pattern explicitly included in the design. The proposed alternating-optimization algorithm reaches power gains close to joint optimization but at far lower computational cost, which is what makes large panels with thousands of elements practical. A sympathetic reader would take the central claim to be that low-cost, manually reconfigurable reflectors can deliver long-term area coverage that was previously expected only from electronically controlled intelligent surfaces.","feed_headline":"Hand-tuned passive panels shape 3D wireless coverage beams","feed_subtitle":"A quasi-static IRS with thousands of passive elements achieves near-optimal shaped beams with just mechanical adjustment.","key_machinery":"The load-bearing object is the Kronecker structure of the full steering vector, $\\mathbf{a}(\\Pi_i,\\Pi_r)=\\mathbf{a}_y\\otimes\\mathbf{a}_z$, which makes the power gain the product of a y-axis factor and a z-axis factor, each affine in its own rank-one matrix $\\mathbf{W}_y$ or $\\mathbf{W}_z$. This product is non-convex jointly but convex in one factor alone, enabling alternating optimization. The rank-one requirement is handled by the DC identity $\\mathrm{rank}(\\mathbf{W})=1 \\iff \\|\\mathbf{W}\\|_*-\\|\\mathbf{W}\\|_2=0$, added as a penalty and linearized by SCA. The shape mask $d(\\Pi_p)$ and sidelobe gap $\\delta$ enter as linear inequalities in $\\mathbf{W}$, and the QS-IRS phase vector is recovered from the dominant singular vectors of $\\mathbf{W}_y$ and $\\mathbf{W}_z$.","core_discovery":"The central discovery is a tractable design recipe for QS-IRS shaped beamforming. For a uniform planar array with element pattern $F(\\Pi)=(\\sin\\theta\\cos\\varphi)^{G/2-1}$, the cascaded channel power gain factorizes as $\\eta^2\\,|\\mathbf{a}_y^{T}\\mathbf{w}_y|^2\\,|\\mathbf{a}_z^{T}\\mathbf{w}_z|^2$, which lets the authors split the full phase optimization into alternating one-dimensional subproblems. Each subproblem is solved by a difference-of-convex penalty for the rank-one constraint and successive convex approximation, subject to mainlobe shape masks and a sidelobe gain gap. In simulation, the algorithm forms square, trapezoidal, and parabolic shaped beams, keeps gains within about 0.4 dB of joint optimization at $8\\times8$, attains 69.73 dB with a $48\\times48$ panel, and shows a power scaling law close to $O(M^2)$. The paper further reports that 4-bit phase quantization costs almost nothing and 2-bit costs about 2 dB.","pith_inferences":["The same manual-reconfiguration idea could be carried to millimeter-wave bands if modular subarray patterns are designed once and reassembled, a direction the paper leaves open.","Because reconfiguration is slow, QS-IRS suits deployment environments whose blockage and coverage needs change on long timescales, such as indoor parking or building surfaces, rather than fast-moving user tracking.","A testable extension would be to optimize the number and set of distinct mass-produced patterns jointly with the phase quantization level, since the assembly cost depends on pattern diversity, not just bit depth.","The $O(M^2)$ gain scaling, if it survives real element coupling, implies that very large passive panels could compete with active arrays for static coverage at a fraction of hardware cost."],"forward_implications":["Large QS-IRS panels of $48\\times48$ passive elements can be designed offline in roughly two hours on a single CPU core, making network planning for long-term coverage feasible.","Shaped beams need not be rectangular or flat: trapezoidal and parabolic masks can be enforced to match irregular target areas.","Power gain grows approximately as $O(M^2)$, so doubling linear array dimensions quadruples coverage gain under the considered model.","Including sidelobe constraints improves mainlobe gain by concentrating energy, compared with designs that ignore them.","Four-bit phase quantization is essentially lossless, so only a handful of distinct element patterns are needed for mass production."],"supporting_citations":[{"why":"Establishes the area-illumination setting and decouples illumination from channel estimation.","marker":"[2]"},{"why":"Motivates 3D beam flattening and subarray-based beam broadening for aerial IRS.","marker":"[3]"},{"why":"Supplies the baseline DC-SCA passive beamforming method and the complexity analysis that the alternating optimization improves on.","marker":"[4]"},{"why":"Provides the difference-of-convex technique and the convergence guarantee used for the rank-one penalty.","marker":"[9]"},{"why":"Supplies the element radiation pattern model $F(\\Pi)$ used throughout the formulation.","marker":"[10]"},{"why":"Provides the reflectarray phase-tuning mechanisms that QS-IRS inherits.","marker":"[16]"},{"why":"Demonstrates hybrid polarization-phase tuning with 360-degree phase coverage, the basis of the divide-and-assemble element patterns.","marker":"[18]"},{"why":"Shows an electromechanical IRS with rotatable patches that supports quasi-static reconfiguration.","marker":"[19]"},{"why":"Shows an electromechanical IRS with movable element ground, another practical QS-IRS implementation.","marker":"[20]"}],"fun_headline_variants":["Mechanical tuning of passive panels shapes 3D coverage beams","Quasi-static IRS achieves shaped beams with mechanical tuning","Low-cost passive arrays: near-optimal shaped beams via mechanical tuning","Shaped 3D beams from passive panels at 0.4 dB from optimal","Mechanical adjustment shapes beams with 4-bit phase at almost no loss"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire optimization rests on the element radiation pattern model $F(\\Pi)=(\\sin\\theta\\cos\\varphi)^{G/2-1}$ accurately representing the angle-dependent gain of every real element; mutual coupling, fabrication tolerances, and pattern asymmetry would shift the actual beam away from the simulated shape.","fun_headline_variants_meta":{"raw":{"variants":["Mechanical tuning of passive panels shapes 3D coverage beams","Quasi-static IRS achieves shaped beams with mechanical tuning","Low-cost passive arrays: near-optimal shaped beams via mechanical tuning","Shaped 3D beams from passive panels at 0.4 dB from optimal","Mechanical adjustment shapes beams with 4-bit phase at almost no loss"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00108,"raw_usage":{"total_tokens":4526,"prompt_tokens":958,"completion_tokens":3568,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":574,"completion_tokens_details":{"reasoning_tokens":3476}},"tokens_in":574,"tokens_out":3568,"duration_ms":27237,"temperature":1.0,"reasoning_tokens":3476,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:27:06.536885+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fabricate a small QS-IRS panel with the designed phases and measure its far-field 3D gain pattern. If the measured mainlobe gain, sidelobe gap, or beam shape deviates from the simulated profile by more than the reported margins (about 0.4 dB relative to joint optimization, or 2 dB at 2-bit quantization), the central claim that the optimized phases deliver the predicted shaped coverage is refuted.","supporting_citations":[{"cited_title":"Aerial intelligent reflecting surface: Joint place- ment and passive beamforming design with 3D beam flattening,","cited_arxiv_id":null,"evidence_quote":"Motivates 3D beam flattening and subarray-based beam broadening for aerial IRS."},{"cited_title":"Passive beamforming for 3-D coverage in IRS-assisted communications,","cited_arxiv_id":null,"evidence_quote":"Supplies the baseline DC-SCA passive beamforming method and the complexity analysis that the alternating optimization improves on."},{"cited_title":"Reconfigurable intelligent surface empowered downlink non-orthogonal multiple access,","cited_arxiv_id":null,"evidence_quote":"Provides the difference-of-convex technique and the convergence guarantee used for the rank-one penalty."},{"cited_title":"Nayeri, F","cited_arxiv_id":null,"evidence_quote":"Provides the reflectarray phase-tuning mechanisms that QS-IRS inherits."},{"cited_title":"Hybrid polarization-phase tuning methodology for reflectarray antennas,","cited_arxiv_id":null,"evidence_quote":"Demonstrates hybrid polarization-phase tuning with 360-degree phase coverage, the basis of the divide-and-assemble element patterns."},{"cited_title":"Angular-adaptive reconfigurable spin-locked metasurface retroreflector,","cited_arxiv_id":null,"evidence_quote":"Shows an electromechanical IRS with rotatable patches that supports quasi-static reconfiguration."},{"cited_title":"An electromechanically reconfigurable intelligent surface for enhancing sub-6g wireless communication signal,","cited_arxiv_id":null,"evidence_quote":"Shows an electromechanical IRS with movable element ground, another practical QS-IRS implementation."}],"review_version":1}