{"id":"fd25a83e-4785-460b-b47e-50acfc64e96a","arxiv_id":"2501.13403","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"ROMA arrays combining panel rotation and antenna movement can improve average spectral efficiency in multi-user MIMO systems.","lead":"The paper proposes a new antenna array design called ROMA that lets each antenna panel rotate in 3D while its elements also move, and reports that this flexibility raises average spectral efficiency in multi-user MIMO downlinks. It is relevant because flexible antenna architectures are a candidate 6G technology, and rotation adds a new degree of freedom beyond existing movable antenna designs.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The printed panel-rotation map in Eq. (1) is not a rigid rotation: at α=β=0 it sends (X,0,Z) to (X,X,Z) and changes inter-antenna spacings. If this map is used in the simulations, the ROMA gains are an artifact of non-physical array stretching.","rationale":"The reader's weakest assumption (LIA/ALR) targets the proof of Theorem 1, but the simulation's SE objective is stated as Eq. (5) with the exact channel (3), so the empirical curves may survive even if Theorem 1 is not proved. The rotation map, by contrast, enters every channel phase used in the optimization and in the baseline comparisons. A single wrong trig term changes the array geometry in a way that is not a rigid-body motion; because the paper's contribution is precisely the rotation degree of freedom, the correctness of Eq. (1) is load-bearing. The internal evidence from Corollary 1 suggests the intended formula is y = X sin α + Z sin β cos α, so the flaw is likely a typo; nevertheless, without code or a rerun, the published result is unverified. This supports the reader's CONDITIONAL verdict, with the condition sharpened to include correcting Eq. (1) and re-running the simulations. I do not see a reason to reject outright because the correction is simple and the proposed architecture is plausible.","tokens_in":9611,"tokens_out":22530,"duration_ms":172170,"concrete_test":"Obtain the simulation code (or re-run the exact-channel AO optimizations) with Eq. (1) corrected to ∆y = X sin α + Z sin β cos α, and verify that the map is orthonormal for all α and β (e.g., pairwise antenna distances remain constant). Then regenerate Fig. 3 and Fig. 4 for ROMA, MA, RO, AS, and FPA. If the ROMA advantage disappears or shrinks substantially, the headline result is due to the non-rigid coordinate transform; if it persists, the typo is benign and the empirical claim stands.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central simulation claim depends on the coordinate model for rotated and movable panels. As printed, Eq. (1) maps the unrotated element (X,0,Z) to (X,X,Z) and does not preserve distances: ||r||^2 = 2X^2 cos^2 α + Z^2(cos^2 β + sin^2 β sin^2 α) ≠ X^2+Z^2. Since P1 includes minimum-distance constraints (11)-(12) and the channel phases in Sec. II and Appendix A are built from these coordinates, an array using Eq. (1) is stretched or skewed when 'rotated', which can create artificial channel gains. The rest of the paper implicitly uses a different formula: the σs,i and ςs,i definitions in Corollary 1 correspond to y = X sin α + Z sin β cos α, not to the printed X cos α. No code is released, so it is impossible to tell whether the Fig. 2-4 curves were generated with the printed or the intended map; if the printed map was used, the reported ROMA-over-baseline gains are not physically meaningful.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This letter introduces ROMA, an architecture in which uniform planar arrays at both the BS and the users can be rotated about two axes while the individual antenna elements are repositioned within a bounded region. The authors model the downlink channel as a sum of L multipath components, approximate each channel entry by a single absolute-sum phase term using the LIA/ALR approximation from [11], and derive an upper bound on the average SE under MR precoding (Theorem 1 and Corollary 1). They then formulate a joint optimization problem P1 over the precoder, element positions, and panel rotation angles, and propose an alternating optimization (AO) algorithm based on penalty decomposition and gradient descent. Section V reports simulations showing that ROMA outperforms FPA, MA, RO, and AS baselines in average SE.","tokens_in":1456,"tokens_out":1435,"duration_ms":72724,"significance":"If the geometric model and the reported gains are correct, ROMA is a plausible extension of movable-antenna and rotary-antenna ideas, and the comparison against four baselines is a useful first benchmark. The manuscript does not provide code, does not give an error bound for the core LIA/ALR approximation, and does not formally prove convergence of the AO algorithm, so the numerical results are the main evidence for the central claim. The contribution is therefore significant but conditional: the idea is worth publishing if the geometry is corrected and the simulations are made reproducible, but the current manuscript cannot be validated as written.","major_comments":[{"comment":"Equation (1) does not describe a rigid rotation of the panel. Setting alpha=beta=0 maps (X,0,Z) to (X,X,Z), and the Euclidean norm is not preserved for general angles. The Corollary 1 definitions sigma_{s,i}=d_sh cos(alpha) gamma + d_sh sin(alpha) eta and ς_{s,i}=d_sv cos(beta) vartheta + d_sv sin(beta) cos(alpha) eta - d_sv sin(beta) sin(alpha) gamma imply that r_{tm,y} has coefficient X sin(alpha), not X cos(alpha) as printed in (1). Since the channel phases, constraints (11)-(12), and all subsequent optimization and simulation results are built on these coordinates, this inconsistency is load-bearing. The authors must state the correct rotation matrix, confirm which map was actually used for Figures 2-4, and either rerun the simulations with the correct map or justify the printed map physically.","section":"Section II, Eq. (1)"},{"comment":"The upper bound in Theorem 1 rests on the approximation in (6), taken from [11, Theorem 1] under the LIA and ALR conditions. The manuscript gives no quantitative error bound, and the simulations use L=15 multipath components with a movable region A=2.5λ, a setting in which the 'arbitrarily large region' condition cannot hold. The authors should either prove the approximation for the simulated regime or validate it numerically, and they should state how a failure of (6) affects the bound (7) and the convergence argument that relies on boundedness.","section":"Theorem 1 and Eq. (6)"},{"comment":"The convergence statement following Algorithm 1 ('monotonically non-increasing... ensuring convergence to a stationary point') is not proven. Subproblem P1-b is solved by a greedy circle-intersection selection, and the claim that 'this intersection point is the optimal z_m' is not established for the coupled constraints (16) involving all antennas simultaneously. The penalty update with increasing rho also lacks a formal argument that stationary points of the penalized problem track stationary points of P1. A proof or a precise reference is needed before the convergence claim can be accepted.","section":"Algorithm 1 and Section IV"},{"comment":"The simulation description omits several parameters needed for reproducibility: the number of channel paths for Figures 2 and 4, the angular distribution of the scatterers, the initial configuration, the minimum spacing D, the penalty schedule rho, and the number of channel realizations or Monte Carlo drops. Without these details and without code, the comparisons in Figures 3-4 cannot be checked. A complete parameter table and ideally the simulation code should be provided.","section":"Section V"}],"minor_comments":[{"comment":"In Eq. (8), the second factor has denominator sin^2(pi/lambda (ς_{s,u}-ς_{s,j})) while the numerator uses ς_{s,j}-ς_{s,u}; the square makes the sign immaterial, but the notation should be made consistent.","section":"Corollary 1, Eq. (8)"},{"comment":"The abstract and introduction say '3D rotation angles', but the model uses two rotation angles per panel; please clarify the axis convention and the meaning of 'tilt relative to the z-axis'.","section":"Introduction and Section II"},{"comment":"The QoS constraint SE_u >= SE_min is mentioned in footnote 3 but is not incorporated into P1; either include it in the formulation or state explicitly that it is deferred to future work.","section":"Footnote 3"},{"comment":"The caption of Figure 2 says 'different transmit power p' but the curves are not labeled with the corresponding values of p; please add the values.","section":"Figure 2"},{"comment":"Reference [6] is cited as 'our previous study' but is an arXiv preprint; please update the reference if a peer-reviewed version is available.","section":"Reference [6]"}],"recommendation":"major_revision","confidential_remarks":"The geometric inconsistency in Eq. (1) is likely a typo, because the Corollary 1 definitions appear to use the intended rotation matrix with r_{tm,y} = X sin(alpha) + Z sin(beta) cos(alpha). However, as submitted, the paper cannot be validated without correcting Eq. (1), confirming which map generated Figures 2-4, and addressing the LIA/ALR validity in the simulated regime. I do not see grounds for rejection if the authors can supply the corrected geometry, a numerical check of the approximation, and reproducible simulation details; otherwise the reported ROMA gains cannot be trusted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The idea is simple and worth a look: combine panel rotation with movable elements, and jointly optimize both for multi-user MIMO. That is genuinely a new twist relative to MA, 6DMA, and rotate-only baselines. The paper shows consistent SE gains over FPA, MA, RO, and AS across region size and transmit power. I believe the gains are probably real, but there is one landmine and one approximation issue.\n\nFirst, Eq. (1) is not a rigid rotation as printed. At α=β=0 it maps (X,0,Z) to (X,X,Z), and it does not preserve distances. The rest of the paper—specifically Corollary 1 and Appendix A—implicitly uses the correct formula with y = X sin α + Z sin β cos α. So the printed coordinate transform is a typo, likely a transcription error. But without a code release, I cannot verify which map drove the simulation curves. If the printed map was actually used, the ROMA gains could be an artifact of array stretching. That is the first thing a referee must ask for.\n\nSecond, the SE upper bound in Theorem 1 is built on the LIA/ALR approximation from [11], with a generic δ and no error bound. For A=2.5λ and L=15, ALR is not satisfied, so the 'upper bound' may not be an upper bound. That said, the optimization objective in P1 is the exact SE (5), and the simulations report exact SE, so the central numerical claim does not collapse if the bound is loose. Still, the paper's theoretical framing overreaches.\n\nMinor: the AO convergence claim relies on monotonicity and boundedness, which is plausible but not proven; simulation details are thin (number of channel drops, user distribution, computational cost). These are standard letter-level omissions.\n\nThe paper is for people working on movable/fluid antennas and 6G flexible arrays. It provides a useful positional contribution and a simulation data point. It deserves a serious referee, primarily to check the rotation-map issue and request code. I would send it out rather than desk reject.","headline":"A plausible new flexible-array idea with a serious typo in the rotation map; the simulation gains are believable but the paper needs a corrected Eq. (1) and a code release before I'd trust the numbers.","tokens_in":10391,"tokens_out":5134,"would_cite":true,"duration_ms":667211,"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":"ROMA-aided multi-user MIMO, with panels that rotate in 3D while elements move within a region, beats fixed, movable-only, rotation-only, and selection baselines in average spectral efficiency via joint geometric optimization.","keywords":["rotary and movable antenna","multi-user MIMO","spectral efficiency","alternating optimization","antenna position optimization","array rotation","maximum ratio precoding"],"falsifier":"Re-simulate Figs. 3 and 4 using the exact multipath channel of Eq. (3) instead of the approximation in Eq. (6), at the same settings ($A=2.5\\lambda$, $L=15$); if the optimized ROMA configuration no longer beats the MA and RO baselines, or if the gap between the exact SE and the Theorem 1 upper bound is large, the central claim is falsified. A simpler check: at fixed transmit power, increase $L$ while keeping $A$ small—the approximation should degrade and the predicted SE gain should shrink.","tokens_in":9463,"feed_emoji":"📡","tokens_out":7242,"duration_ms":62274,"temperature":0.7,"pith_summary":"The paper is trying to establish that letting antenna panels both rotate in three dimensions and reposition their individual elements—a configuration it calls rotary and movable antenna (ROMA)—can substantially raise the average spectral efficiency of a downlink multi-user MIMO system without adding antennas or bandwidth. It develops a geometric model in which each panel's two rotation angles and each element's position enter the channel phases, derives an upper bound on spectral efficiency under maximum-ratio precoding, and proposes an alternating-optimization algorithm that alternately updates transmit-side positions, user-side positions, and rotation angles. Simulation results show the ROMA design outperforming fixed-position, movable-antenna, rotation-only, and antenna-selection baselines, with the gain growing as the movable region and transmit power increase. If correct, this makes antenna geometry itself a resource that can be optimized jointly with signal processing, with implications for future MIMO deployments.","feed_headline":"Rotating + moving antennas lifts multi-user MIMO spectral efficiency","feed_subtitle":"Jointly optimizing panel rotation and element positions beats position-only, rotation-only, and fixed-array designs.","key_machinery":"The central object is the ROMA panel, a uniform planar array whose plane can be rotated around the x-axis by angle $\\alpha$ and tilted relative to the z-axis by angle $\\beta$, while each element's position on the plane can be shifted within a region. The argument is carried by writing every channel entry as a sum of phase terms that depend on these rotation angles and positions, then invoking the linearly independent angle (LIA) condition and the arbitrarily large region (ALR) assumption to replace each entry by a single phase term $\\|\\mathbf{b}_{umn_u}\\|_1 e^{j\\pi v_u}$. That replacement yields the channel-gain quantity $G_u$ and the SE upper bounds in Theorem 1 and Corollary 1, which in turn define the objective for the optimization. The optimization itself is an alternating algorithm: it updates transmit-side and user-side element positions by variable splitting with a penalty term, enforces minimum-distance constraints through a geometric alternating-optimization step, and updates the two rotation angles by gradient descent using automatic differentiation.","core_discovery":"On its own terms, the paper's discovery is that joint control of two geometric degrees of freedom—panel rotation and element translation—yields spectral-efficiency gains that either degree alone cannot provide. For a system of ROMA panels at both base station and users, the paper shows that the average spectral efficiency under MR precoding is bounded by a closed-form expression involving the channel gain $G_u$ and inter-user interference, and that under line-of-sight propagation the bound takes an explicit trigonometric form depending on rotation angles and antenna spacings. It then establishes that maximizing the average SE over rotation angles and element positions can be carried out by an alternating algorithm that converges to a stationary point, and verifies by simulation that the resulting configuration outperforms fixed-position arrays, movable antennas without rotation, rotatable arrays without moving elements, and antenna selection.","pith_inferences":["The mechanism behind the gain is likely angular separation: rotating a panel changes the effective direction of the user's signal in the array manifold, which for users with overlapping arrival angles can reduce inter-user interference more than moving elements alone; this suggests ROMA's advantage will be largest in crowded angular scenarios and near-zero when users are already well separated.","The small-regime inconsistency between the ALR assumption and the simulated $A=2.5\\lambda$ region implies the reported gains may be partly an artifact of the approximation; a natural test is to run the same AO algorithm on the exact channel and compare, which the paper does not do.","The framework suggests a design spectrum: at one end pure movable antennas (translation only), at the other pure rotation; ROMA's joint optimization indicates there is an efficient frontier between translation and rotation that a hardware designer could trade off against motor cost.","In near-field or strong-scattering settings where the LIA/ALR conditions fail, a robust extension would replace the single-phase-term approximation with the full sum-of-phases channel and optimize the same geometry by sampling or surrogate models."],"forward_implications":["For a fixed number of antennas, ROMA can be treated as a software-reconfigurable geometry layer: the same hardware can be steered toward user clusters as their spatial distribution changes, improving average SE without additional spectrum or power.","The closed-form SE bound makes the geometric design problem differentiable, so gradient-based tools can be used for real-time reconfiguration once CSI is available.","Gains grow with the movable-region size and with transmit power, so ROMA is most valuable in high-power, spatially constrained deployments where fixed-position, rotation-only, and antenna-selection designs saturate.","Because the boundedness argument extends to other precoding schemes, the same alternating-optimization framework can be applied beyond MR precoding, for instance with the zero-forcing precoding already used in the algorithm's precoding subproblem."],"supporting_citations":[{"why":"Supplies the multipath channel model and the L=15 simulation parameters against which ROMA is evaluated.","marker":"[7]"},{"why":"Provides the LIA/ALR conditions and the approximation theorem used to turn each channel entry into one phase term, the basis of Theorem 1.","marker":"[11]"},{"why":"Defines the movable-antenna baseline whose position optimization ROMA extends with panel rotation.","marker":"[3]"},{"why":"Defines the rotation-only baseline with fixed element positions and panel rotation.","marker":"[14]"},{"why":"Defines the antenna-selection baseline of choosing 9 of 12 fixed antennas.","marker":"[15]"},{"why":"Provides the geometric alternating optimization used to enforce minimum-distance constraints between elements.","marker":"[12]"},{"why":"Supplies the high-SNR SE reformulation that converts the SE expression into the bound used in Theorem 1.","marker":"[13]"},{"why":"Gives the per-user spectral-efficiency expression in terms of the interference matrix used as the objective.","marker":"[10]"},{"why":"Supplies the differential-evolution baseline to which the alternating-optimization algorithm's convergence is compared.","marker":"[6]"}],"fun_headline_variants":["Rotating plus moving antennas jointly beats either alone","Joint optimization of antenna rotation and position boosts MIMO spectral efficiency","Combining antenna rotation and movement improves MIMO performance","Joint rotation and movement of antennas enhances MIMO spectral efficiency"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the multipath arrivals have sufficiently independent angles and the antenna movable region is large enough that every channel entry can be replaced by one phase term; the paper's own simulations use a small region ($A=2.5\\lambda$) with 15 paths, a setting where that premise is not strictly guaranteed, and if it fails the SE upper bound and the optimization's theoretical justification do not hold.","fun_headline_variants_meta":{"raw":{"variants":["Rotating plus moving antennas jointly beats either alone","Joint optimization of antenna rotation and position boosts MIMO spectral efficiency","Combining antenna rotation and movement improves MIMO performance","Joint rotation and movement of antennas enhances MIMO spectral efficiency"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000897,"raw_usage":{"total_tokens":3845,"prompt_tokens":908,"completion_tokens":2937,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":524,"completion_tokens_details":{"reasoning_tokens":2872}},"tokens_in":524,"tokens_out":2937,"duration_ms":19333,"temperature":1.0,"reasoning_tokens":2872,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:59:05.079614+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-simulate Figs. 3 and 4 using the exact multipath channel of Eq. (3) instead of the approximation in Eq. (6), at the same settings ($A=2.5\\lambda$, $L=15$); if the optimized ROMA configuration no longer beats the MA and RO baselines, or if the gap between the exact SE and the Theorem 1 upper bound is large, the central claim is falsified. A simpler check: at fixed transmit power, increase $L$ while keeping $A$ small—the approximation should degrade and the predicted SE gain should shrink.","supporting_citations":[{"cited_title":"Rayleigh fading model ing and channel hardening for reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Supplies the multipath channel model and the L=15 simulation parameters against which ROMA is evaluated."},{"cited_title":"Performance analys is and optimization for movable antenna aided wideband communica tions,","cited_arxiv_id":null,"evidence_quote":"Provides the LIA/ALR conditions and the approximation theorem used to turn each channel entry into one phase term, the basis of Theorem 1."},{"cited_title":"Movable-an tenna position optimization: A graph-based approach,","cited_arxiv_id":null,"evidence_quote":"Defines the movable-antenna baseline whose position optimization ROMA extends with panel rotation."},{"cited_title":"Spatial multiplexing in near-field line-o f-sight MIMO communications: Paraxial and non-paraxial deployments,","cited_arxiv_id":null,"evidence_quote":"Defines the rotation-only baseline with fixed element positions and panel rotation."},{"cited_title":"Energy efficiency of large- scale multiple antenna systems with transmit antenna selection,","cited_arxiv_id":null,"evidence_quote":"Defines the antenna-selection baseline of choosing 9 of 12 fixed antennas."},{"cited_title":"Handling distance con straint in movable antenna aided systems: A general optimization fram ework,","cited_arxiv_id":null,"evidence_quote":"Provides the geometric alternating optimization used to enforce minimum-distance constraints between elements."},{"cited_title":"The ca pacity of wire- less networks: Information-theoretic and physical limits ,","cited_arxiv_id":null,"evidence_quote":"Supplies the high-SNR SE reformulation that converts the SE expression into the bound used in Theorem 1."},{"cited_title":"Op timal bilinear equalizer for cell-free massive MIMO systems over correlat ed Rician channels,","cited_arxiv_id":null,"evidence_quote":"Gives the per-user spectral-efficiency expression in terms of the interference matrix used as the objective."}],"review_version":1}