{"id":"6aa51aa0-3cd4-480e-a0af-9899c19ef022","arxiv_id":"2506.00972","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A joint beamforming, RIS phase, and movable-antenna position design for near-field directional modulation is reported to raise secrecy rate by 28% while cutting antenna count by 37.5% in simulation.","lead":"This paper designs a secure wireless transmission scheme that combines movable antennas, a reconfigurable intelligent surface, and near-field directional modulation to maximize secrecy rate under hardware impairments. Simulations report a 28% secrecy-rate gain while using 37.5% fewer antennas than a conventional fixed-position array.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm 1 uses α inside L4,k to optimize wk, but α is only computed afterward from wk via (80), leaving an unaddressed circular dependency that makes the reported 28% SSR gain unreproducible.","rationale":"The reader's weakest-assumption analysis identified the SLNR-to-SSR surrogate equivalence as the primary risk and the α–wk coupling as a secondary point. I agree that the surrogate equivalence is unproven, but the more concrete and more immediately falsifiable problem is the α circular dependency: L4,k in (49) depends on α, Algorithm 1 neither receives nor updates α, and (80) only computes α after wk is fixed. This is an internal-consistency gap, not merely a question of surrogate tightness. It directly threatens the reproducibility of the simulation-based headline claim, because the reported SSR values may depend on an unspecified α initialization. The issue is addressable by specifying the initialization and adding an outer fixed-point loop, so the appropriate verdict remains CONDITIONAL rather than REJECT or ACCEPT. I do not claim the authors acted improperly; the manuscript simply omits a required algorithmic step.","tokens_in":22979,"tokens_out":2775,"duration_ms":30218,"concrete_test":"Modify Algorithm 1 so that the initial α is an explicit input, then run the full pipeline (Algorithm 1 plus the α update in (80) plus MMSE receive beamformer updates) for two very different initial values, e.g., α = 0.3 and α = 0.9, under otherwise identical settings. If the final SSR after convergence differs materially between the two runs, the algorithm is initialization-dependent and the reported 28% gain is not robust. A positive control would be to add an outer loop that alternates (80) with Algorithm 1 until |α^{(i+1)} − α^{(i)}| < 10⁻³ and then report the converged SSR and the number of outer iterations required.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central performance claim depends on the correctness of Algorithm 1's output. In Sec. 4.2, the SLNR surrogate L4,k in (49) contains the power-allocation factor α through the imperfect-CSI penalty (ε̂eP0/K + ε̂k(1−α)P0)‖θ̂‖. This α is therefore needed to form the generalized Rayleigh quotient in P5 and to solve for wk. However, Algorithm 1's input list does not include α, and no α update appears inside the loop. Section 4.4 computes α only after wk has been optimized, using (80): α = K‖u_k^H Q_k T w_k‖² / (P0 N_a² ‖T Q_k^H u_k‖²). This creates a circular dependence: the wk used to compute α was itself obtained with an unspecified value of α. If α is initialized to an arbitrary constant, the final (wk, α) pair is generally not a fixed point of the coupled equations. Consequently, the SSR values plotted in Figs. 5–10, including the headline 28% enhancement over FPA with 37.5% fewer antennas, may correspond to an inconsistent operating point rather than a valid solution of P1. The manuscript does not state how α is initialized, whether it is updated jointly with wk, or whether the reported results use a fixed point. Without this, the simulation cannot be reproduced and the central claim is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper formulates a secrecy sum rate (SSR) maximization problem for a near-field RIS-assisted MIMO system with movable antennas, hardware impairments, imperfect CSI, discrete MA placement, and artificial noise. The authors decompose the non-convex joint problem and propose three algorithms: an iterative leakage-theory/phase-alignment method for transmit beamforming and RIS phases, a uniform-grouping discrete MA placement algorithm, and a compressed-sensing-based non-uniform grouping algorithm. Receive beamforming is derived by an MMSE criterion. Simulation results claim a 28% SSR enhancement with 37.5% fewer antennas relative to fixed-position-antenna systems. The paper is a useful extension of directional modulation to MA/RIS near-field settings, but two load-bearing issues—the asserted equivalence of the SLNR surrogate to SSR and the circular dependence on α—need to be resolved before the central quantitative claim can be accepted.","tokens_in":23303,"tokens_out":4892,"duration_ms":55732,"significance":"If the reported gains are reproducible, the paper would be among the first to combine near-field directional modulation, movable antennas, RIS, hardware impairments, and imperfect CSI in a single design, with low-complexity discrete MA positioning. The explicit modeling of active-RIS power constraints, the use of leakage theory for beamforming, and the two MA-grouping algorithms are concrete contributions. The paper is not parameter-fitted to a target output: the optimization is formulated with channel and impairment constraints and then evaluated by simulation. However, the significance is tempered by the lack of a proof or numerical validation that the SLNR surrogate tracks the true SSR, by an unspecified treatment of α in the main iterative loop, and by the absence of any statistical characterization of the simulation results.","major_comments":[{"comment":"The paper asserts that the SLNR ratio in (44), and its later forms (49) and (53), is 'equivalent to the SSR maximization problem,' but no proof of a monotonic relationship is provided. The original SSR in (20) is a sum over users of log(1 + SINR_bob) - log(1 + SINR_eve), whereas (53) defines Rs(wk) as a single per-user Rayleigh quotient. Maximizing such a quotient is not generally equivalent to maximizing a sum of logarithms of SINR differences, especially under the HWIs and imperfect-CSI terms folded into L4,k and L5,k. Since every subsequent optimization step (P3-P6, Algorithms 1-3) is built on this surrogate, the reported 28% SSR gain may reflect the surrogate rather than the true SSR. The authors should either prove the equivalence under the stated constraints or provide a numerical comparison showing that the surrogate-optimized solution closely tracks the true SSR (20) across the simulated parameter range.","section":"Sec. 4.2, Eqs. (44)-(53)"},{"comment":"The factor α enters the beamforming optimization through L4,k in (49) via the term (ε̂e P0/K + ε̂k(1−α)P0)||θ̂|| I_N, but α is not part of Algorithm 1's input list and is not updated inside the iteration; it is computed only afterward from wk using (80). This creates a circular dependency: the wk obtained from Algorithm 1 depends on an unspecified value of α, and the value of α obtained from (80) will not, in general, be consistent with the α used in that wk computation. In addition, (80) is k-dependent, while α in (1) must be a single global scalar; the manuscript does not say how the per-user expressions are reconciled. The authors must specify how α is initialized, whether it is updated jointly with wk in a fixed-point loop, and how a consistent (wk, α) pair is obtained for the reported simulations.","section":"Sec. 4.2, Eq. (49) and Sec. 4.4, Eq. (80); Algorithm 1"},{"comment":"The simulation section reports no error bars, no number of channel realizations, and no description of how the random channel estimation errors in (21)-(24) are sampled for each plotted point. The headline numbers—28% SSR enhancement and 37.5% antenna reduction—are single-point comparisons extracted from these figures. Since the imperfect-CSI errors and the channel realizations are stochastic, the reader cannot assess whether the claimed gains are statistically significant or specific to one realization. The authors should specify the Monte Carlo protocol, the number of averaged realizations, and confidence intervals or at least standard-error markers for the main comparisons.","section":"Sec. 5, Figs. 5-10"},{"comment":"The transition from (48) to (49) multiplies the imperfect-CSI penalty by ||θ̂|| in the denominator, but the derivation of (37)-(38) already uses the inequality ||θ̂ Âe T w_k|| ≤ ε̂e ||θ̂|| ||T w_k||, and the resulting bound is then inserted as an additive term inside the Rayleigh quotient. It is not shown that this additive substitution preserves the monotonic ordering of the original SLNR or that the subsequent eigenvector solution of P5 is a valid approximation of the constrained problem with C9 and C10. A formal statement of the approximation error, or a comparison against the original constrained problem for a small instance, would be needed to justify the use of (49) and (53) as the objective in P3-P5.","section":"Sec. 4.2, Eqs. (48)-(49) and (53)"}],"minor_comments":[{"comment":"The variable N_z is used in the mapping (5) but is never defined; it appears to be N_v or N_h. Please define it explicitly.","section":"Sec. 3.1, Eq. (5)"},{"comment":"The sum notation in (9) and (10) is rendered as '∑ 4 i=1' and '∑ 3 i=1', which is unclear; it should be ∑_{i=1}^{4} and ∑_{i=1}^{3}.","section":"Sec. 3.1, Eqs. (9)-(10)"},{"comment":"The complexity expression 'approximately Nt n0^3 / (S N^3) that of the exhaustive search algorithm' is grammatically and mathematically unclear; please state whether it means 'N_t n_0^3 / (S N^3)' times the exhaustive-search complexity, and define all symbols.","section":"Sec. 4.3.1"},{"comment":"In the passive-RIS initial state, the bound (55) uses ||θ||^2 = M, but the preceding line writes the inequality as if ||θ||^2 appears without identifying that θ is the all-ones vector scaled by unit modulus; please clarify the initial-state assumptions.","section":"Sec. 4.2, Eq. (55)"},{"comment":"In the MMSE derivation, the expression for E_k(u_k) is written with both u_k and u_k^H dependencies, but the derivative in (85) is taken with respect to u_k^*; it would be helpful to state the Wirtinger-calculus convention used.","section":"Sec. 4.4, Eq. (82)"},{"comment":"The text contains several spacing artifacts, such as 'T ABLE 1' and 'achieve' split across lines; a careful copyedit would improve readability.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"This is a meaningful extension of directional modulation to near-field MA/RIS systems, and the paper has a substantial amount of original formulation and algorithm design. However, the two main load-bearing points—the surrogate-to-SSR equivalence and the treatment of α—are not adequately addressed, and the simulation section lacks the statistical detail needed to support the quantitative headline claim. These issues are fixable within the manuscript's scope: a proof or strong numerical validation of the surrogate, a clear fixed-point update for α, and Monte Carlo error characterization would make the results reproducible and convincing. I would not reject the paper at this stage, but I would require these revisions before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my candid read of arXiv:2506.00972. The paper does something genuinely new: it is the first to put near-field directional modulation together with RIS, movable antennas, hardware impairments, and imperfect CSI in one system, and the two discrete-position grouping schemes (uniform and CS-based non-uniform) are not in the cited literature. The system model is coherent, the complexity reduction relative to exhaustive search is real, and the simulation campaign is broad. Credit where it is due: this is a plausible decomposition of a hard problem, and the CS-based grouping idea is inventive.\n\nThe soft spots are exactly where the reader's report and the stress-test note point. Most load-bearing is the α–wk cycle. Algorithm 1 optimizes wk using L4,k, which contains the power allocation factor α through the imperfect-CSI penalty. But α is computed only later, in (80), from the same wk. The algorithm neither initializes α nor updates it inside the loop. So the final (wk, α) pair is generally not a fixed point of the coupled equations, and the manuscript does not state whether the reported numbers come from a consistent operating point. This makes the headline 28% SSR gain unreproducible as written. A simple fix—iterate α and wk, or state that α is held fixed and justify that choice—would address it. As it stands, the central result is not established.\n\nSecond, the SLNR surrogate in (44)–(53) is asserted to be \"equivalent\" to SSR maximization without proof. That is a strong claim. The surrogate may be a reasonable engineering heuristic, but no monotonic relationship is shown, and the entire optimization leans on this equivalence. Third, the simulations have no error bars and no number of channel realizations. Everything is a single simulated scenario, so the 28% and 37.5% numbers could be geometry-specific.\n\nThe citation pattern looks fine; the self-citations are to the authors' prior DM work and do not create a circular dependency. The math is mostly standard generalized Rayleigh quotient and MMSE, so I see no fatal error beyond the α gap.\n\nWho gets value from this paper: wireless security researchers working on MA or RIS, who will want to see the algorithm corrected. It deserves serious refereeing—the combination is important if it holds up—but it should come back with a clear α-update rule, a sensitivity check on the initial α, and statistical confidence intervals for the claimed gains.","headline":"A genuinely new combination of near-field DM, RIS, MAs, HWIs, and imperfect CSI, but the central 28% gain claim rests on an unproven SLNR-to-SSR surrogate and an unaddressed α–wk feedback loop.","tokens_in":23809,"tokens_out":3083,"would_cite":false,"duration_ms":31683,"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":"Jointly tuning movable-antenna positions, RIS phases, beamforming, and artificial noise raises the secrecy sum rate by 28 percent with 37.5 percent fewer antennas than fixed-position arrays, even with hardware impairments and imperfect…","keywords":["movable antennas","directional modulation","reconfigurable intelligent surface","near-field communication","physical layer security","hardware impairments","imperfect channel state information","compressed sensing MA positioning"],"falsifier":"Compute the true secrecy sum rate in (20) at the beamforming and phase-shift solutions produced by the SLNR-based Algorithm 1, then at several deliberately perturbed alternatives that score lower on the surrogate; if any perturbed design achieves a higher true SSR, the claimed surrogate-to-SSR equivalence is broken. A simpler check is to re-run the simulation with alpha re-updated inside the iteration loop; if the secrecy sum rate changes by more than the simulation's tolerance, the fixed-point assumption behind the reported 28% gain is violated.","tokens_in":2077,"feed_emoji":"📡","tokens_out":9511,"duration_ms":155877,"temperature":0.7,"pith_summary":"This paper is the first to design near-field directional modulation for a movable-antenna (MA) base station aided by a reconfigurable intelligent surface (RIS), and it asks whether the added mechanical freedom pays for itself in secrecy. The central claim is that a joint optimization of MA positions, RIS phase shifts, confidential and artificial-noise beamforming, the power split, and receiver combining achieves a 28% gain in secrecy sum rate (SSR) while using 37.5% fewer antennas than a traditional fixed-position-antenna (FPA) system, all under hardware impairments (HWIs) and imperfect channel state information (CSI). A sympathetic reader would care because physical-layer security usually demands more antennas, power, or RIS elements; this paper argues the same security can be bought with fewer, movable elements. It supports the claim with three low-complexity algorithms: an iterative leakage-theory design of beamforming and RIS phases, a uniform grouping strategy, and a compressed-sensing-based non-uniform grouping strategy for discrete MA placement.","feed_headline":"RIS-aided movable antennas lift secrecy rate 28%, cut antennas 37.5%","feed_subtitle":"Movable antennas plus RIS beat fixed arrays on secrecy rate, even with imperfect hardware and channel knowledge.","key_machinery":"The load-bearing object is the leakage-theoretic SLNR surrogate of Sec. 4.2: for each user k, the objective is the ratio of the desired received power at that user to the sum of leakage power to all other users, noise terms, the hardware-impairment virtual channel, and an imperfection penalty that grows with the channel-estimation error bounds. Maximizing this ratio is a generalized Rayleigh quotient; the paper solves it by taking the leading eigenvector after a null-space projection that enforces the artificial-noise orthogonality constraints. Two supporting mechanisms carry the rest: a phase-difference variable with monotonic-gain constraints that turns the RIS phase optimization into a convex program, and a compressed-sensing group-sparsity problem that approximates the antenna-selection constraint by an ell-one norm over per-position beamforming weights, generating non-uniform position groups at far lower complexity than exhaustive search.","core_discovery":"The paper's discovery, stated on its own terms, is that the secrecy sum rate maximization for a near-field RIS-assisted MA downlink, a non-convex problem that couples beamforming vectors, RIS phases, MA positions, power allocation, and receive filters, can be decomposed and solved without exhaustive search over antenna positions. In the decomposition, the SSR objective is replaced by a signal-to-leakage-noise ratio (SLNR) built from leakage matrices, a virtual channel that absorbs the transmitter and receiver hardware impairments, and channel-estimation error bounds, turning each beamforming subproblem into a generalized Rayleigh quotient solvable by an eigenvector computation with null-space projection to keep the artificial noise out of Eve's subspace. The RIS phases are then updated through a convex phase-alignment step that guarantees Bob's received power does not decrease between iterations, and the discrete MA positions are chosen by grouping candidate positions uniformly or by a compressed-sensing group-sparsity formulation that prunes redundant locations. Simulations then show the combined design delivering the claimed 28% SSR enhancement with a 37.5% antenna reduction against FPA baselines, plus larger gains from active RIS than passive RIS and saturation behavior in antenna count, candidate positions, and transmit power.","pith_inferences":["The SLNR construction that absorbs HWIs and CSI error bounds into a virtual channel is portable: the same trick could convert secrecy-rate or covert-rate optimizations in other architectures (such as cell-free networks, STAR-RIS, or terahertz links) into Rayleigh-quotient problems, with the same unproven surrogate-to-SSR gap to watch.","The alpha-w_k fixed-point loop is incomplete: alpha is computed from w_k in (80) while w_k depends on alpha through L_4,k in (49), and Algorithm 1 never re-updates alpha. A natural testable extension is to alternate an alpha-update into the iteration and check whether SSR rises; if it does, the reported 28% gain is a lower bound.","The discrete MA-position selection is structurally a sparse-array selection problem, so the CS grouping could be sharpened by proving that the ell-one relaxation preserves the SSR ordering of position groups; absent such a proof, the grouping is a heuristic whose worst-case gap to exhaustive search is unknown.","Because the near-field boundary shifts with the array aperture as antennas move, the channel model's Fresnel-zone assumption should be re-checked when the movable region is large; a testable extension would vary the region size and verify the NF steering vectors against a full spherical-wave model."],"forward_implications":["If the central claim holds, antenna count is a tradeable resource: a base station can give up roughly three of every eight antennas and recover the lost secrecy performance by letting the remaining antennas move within a small region, cutting RF-chain and hardware cost.","Active RIS should be preferred over passive RIS for near-field secrecy when the reflection power budget allows, since the simulations show substantially larger SSR gains, while adding RIS elements beyond a point yields diminishing returns in MA-aided systems.","The CS-based non-uniform grouping (Algorithm 3) is the better choice when the movable region or candidate-position count is small, and it approaches the best-of-10000-random-placements benchmark as candidate positions grow.","Secrecy performance degrades measurably with hardware impairment levels (up to 24% at mu_t = mu_r = 0.01 versus the impairment-free baseline), so improving transceiver hardware quality is a direct route to secrecy."],"supporting_citations":[{"why":"Supplies the hardware-impairment model: transmitter distortion noise z_t and receiver/AGC distortion z_r,k with ratios mu_t, mu_r, used in the signal model of Sec. 3.1.","marker":"[44]"},{"why":"Supplies the bounded channel-estimation-error model (norm of Delta H <= epsilon) used to build the imperfect-CSI constraints and the SLNR penalty term.","marker":"[48]"},{"why":"Justifies the near-field (Fresnel-zone) channel model via its eq. (9), the criterion the paper relies on when placing the RIS close to Bob.","marker":"[47]"},{"why":"Provides the movable-antenna channel modeling and performance analysis that this paper extends from rate to secrecy-oriented directional modulation.","marker":"[22]"},{"why":"Establishes MA-position optimization for multiuser communication, the baseline positioning method the discrete grouping algorithms must beat or approximate.","marker":"[23]"},{"why":"Provides the normalization method used to scale the optimized RIS phase vector so that the active-RIS reflection power constraint C8 is satisfied.","marker":"[6]"},{"why":"Supplies the directional-modulation secrecy-rate-maximization technique over RIS that the leakage-theory beamforming design builds on.","marker":"[2]"},{"why":"Originates the artificial-noise idea that the AN component of the transmitted signal and the zero-space projection constraints implement.","marker":"[8]"}],"fun_headline_variants":["28% secrecy gain and 37.5% fewer antennas with RIS and movable arrays","Movable antennas and RIS: 28% secrecy boost, 37.5% fewer antennas","Near-field RIS-MA system: 28% secrecy rate gain, 37.5% antenna savings","Secrecy rate up 28%, antennas down 37.5% via RIS and movable arrays"],"cache_read_input_tokens":25984,"weakest_assumption_plain":"Everything hinges on the paper's signal-to-leakage-noise-ratio substitute in Section 4.2 rising and falling with the true secrecy sum rate; the paper asserts that equivalence but never proves it, so the claimed 28 percent gain is only as solid as that match.","fun_headline_variants_meta":{"raw":{"variants":["28% secrecy gain and 37.5% fewer antennas with RIS and movable arrays","Movable antennas and RIS: 28% secrecy boost, 37.5% fewer antennas","Near-field RIS-MA system: 28% secrecy rate gain, 37.5% antenna savings","Secrecy rate up 28%, antennas down 37.5% via RIS and movable arrays"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000806,"raw_usage":{"total_tokens":3589,"prompt_tokens":1044,"completion_tokens":2545,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":660,"completion_tokens_details":{"reasoning_tokens":2445}},"tokens_in":660,"tokens_out":2545,"duration_ms":18128,"temperature":1.0,"reasoning_tokens":2445,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:54:12.019362+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the true secrecy sum rate in (20) at the beamforming and phase-shift solutions produced by the SLNR-based Algorithm 1, then at several deliberately perturbed alternatives that score lower on the surrogate; if any perturbed design achieves a higher true SSR, the claimed surrogate-to-SSR equivalence is broken. A simpler check is to re-run the simulation with alpha re-updated inside the iteration loop; if the secrecy sum rate changes by more than the simulation's tolerance, the fixed-point assumption behind the reported 28% gain is violated.","supporting_citations":[{"cited_title":"Secure communication in active IRS-assisted miso systems with hard ware impairments,","cited_arxiv_id":null,"evidence_quote":"Supplies the hardware-impairment model: transmitter distortion noise z_t and receiver/AGC distortion z_r,k with ratios mu_t, mu_r, used in the signal model of Sec. 3.1."},{"cited_title":"Robus t beamforming design for IRS-aided cell-free systems with csi un- certainties and capacity-limited backhaul,","cited_arxiv_id":null,"evidence_quote":"Supplies the bounded channel-estimation-error model (norm of Delta H <= epsilon) used to build the imperfect-CSI constraints and the SLNR penalty term."},{"cited_title":"Multi-beam design for near - field extremely large-scale IRS-aided wireless communicati ons,","cited_arxiv_id":null,"evidence_quote":"Justifies the near-field (Fresnel-zone) channel model via its eq. (9), the criterion the paper relies on when placing the RIS close to Bob."},{"cited_title":"Modeling and performance an aly- sis for movable antenna enabled wireless communications,","cited_arxiv_id":null,"evidence_quote":"Provides the movable-antenna channel modeling and performance analysis that this paper extends from rate to secrecy-oriented directional modulation."},{"cited_title":"Movable-antenna en - hanced multiuser communication via antenna position optim iza- tion,","cited_arxiv_id":null,"evidence_quote":"Establishes MA-position optimization for multiuser communication, the baseline positioning method the discrete grouping algorithms must beat or approximate."},{"cited_title":"Three high-rate beamforming methods for active IR S- aided wireless network,","cited_arxiv_id":null,"evidence_quote":"Provides the normalization method used to scale the optimized RIS phase vector so that the active-RIS reflection power constraint C8 is satisfied."},{"cited_title":"Enhanced secrecy rate maximization for directional modul ation networks via IRS,","cited_arxiv_id":null,"evidence_quote":"Supplies the directional-modulation secrecy-rate-maximization technique over RIS that the leakage-theory beamforming design builds on."},{"cited_title":"Guaranteeing secrecy using artificia l noise,","cited_arxiv_id":null,"evidence_quote":"Originates the artificial-noise idea that the AN component of the transmitted signal and the zero-space projection constraints implement."}],"review_version":1}