{"id":"3534ff3c-d456-48bb-9d0d-0bbea3995295","arxiv_id":"2507.14804","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Movable element positions at a STARS, jointly optimized with active and passive beamforming, increase simulated sum secrecy rate in a full-space eavesdropping scenario.","lead":"A movable-element simultaneously transmitting and reflecting surface is proposed to improve wireless physical layer security against eavesdroppers on both sides of the surface. Jointly optimizing the element positions, the surface coefficients, and base station beamforming raises the simulated sum secrecy rate above fixed-element surfaces.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central ME-STARS gain assumes exact eavesdropper CSI, unstated and unjustified for passive adversaries; the secrecy rates shown are upper bounds, not achievable secure rates.","rationale":"The paper's central numerical claims are only meaningful if the optimized rates are achievable secrecy rates. Eq. (10) uses exact g_{k_e}(R), and the optimization enforces constraints (12g)-(12h) on eavesdropper rates; this requires the transmitter to know the passive eavesdroppers' channels, which is exactly what a passive adversary hides. The paper contains no statement of this assumption and no robust/worst-case treatment. This is the most load-bearing concern because it affects both the objective and the constraints used in every figure, including the headline 25% ME gain. If the premise fails, the curves are upper bounds against a genie-aided adversary, and the ME-STARS versus FPE-STARS comparison is not a statement about realizable secure communications. The reader's weakest_assumption identifies the same issue, so I agree. Secondary issues also deserve attention: the numerical curves appear to come from single channel realizations without Monte Carlo averaging, and the rank-one proof of Theorem 1 is deferred to an external reference rather than shown. These do not change the verdict. The paper would be acceptable as a conditional contribution if it explicitly frames the results as perfect-Eve-CSI upper bounds or adds a legitimate-CSI-only/robust benchmark; the current unstated premise leaves the headline claim unsupported for the intended passive-eavesdropper scenario.","tokens_in":20489,"tokens_out":7823,"duration_ms":90749,"concrete_test":"Ask the authors to state whether Fig. 4 assumes genie-known g_{k_e}(R); then re-run the ME-STARS versus FPE-STARS comparison under a model without eavesdropper CSI, e.g., using only legitimate-user CSI and a worst-case expectation over eavesdropper channels within the same angular/spatial uncertainty set, or a robust version of (12) with g_{k_e} varying over an uncertainty region. If the ME-STARS gain over FPE-STARS vanishes or is not consistently positive across realizations, the abstract's first claim is conditional on perfect eavesdropper CSI.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The optimization problem (12) directly imposes constraints (12g)-(12h) involving the eavesdropper rates R^e_{k_e,k_l}, which by (10) depend on the exact eavesdropper channels g_{k_e}(R). The objective (12a) also contains the term max_{k_e} R^e_{k_e,k_l}. Nowhere in the paper is it stated that eavesdropper CSI is available, nor is any estimation or robust design provided for a passive adversary. Since a passive eavesdropper does not feed back CSI, the computed beamformers and ME positions are not implementable as a secure communication scheme; the simulated sum secrecy rate is an upper bound achieved against known eavesdropper channels. The claimed ME-STARS gain over FPE-STARS is therefore not established for the intended threat model: both schemes may benefit differently when eavesdropper channels are unknown, and the relative gain could disappear or reverse. This is not a criticism of the optimization math; it is a missing premise that the abstract and numerical sections never disclose.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper considers a downlink MISO secure communication system in which a simultaneously transmitting and reflecting surface (STARS) is equipped with movable elements (MEs). The authors formulate a sum secrecy rate maximization problem that jointly optimizes the ME positions, the STARS transmission/reflection coefficients, and the BS active beamforming, subject to user rate and eavesdropper leakage constraints. They propose an alternating optimization (AO) algorithm: a penalty-based gradient ascent method for ME positions, and successive convex approximation (SCA) / semi-definite relaxation (SDR) methods for the passive and active beamforming subproblems. Convergence and complexity analyses are provided, and numerical results compare the proposed ME-STARS with fixed-position STARS, ME-RIS, and random-position STARS baselines.","tokens_in":20671,"tokens_out":11253,"duration_ms":132238,"significance":"If the central claims hold, the paper provides a useful quantification of the physical-layer security gains offered by moving STARS elements and, importantly, shows that the secrecy-rate benefit saturates as the movable region grows, which is practically relevant for aperture sizing. The system model and optimization machinery are mostly standard and carefully derived; in particular, the gradient derivation in Appendix A and the complexity accounting are explicit, and the comparison baselines are appropriate. The paper does not suffer from the circularity concern that sometimes arises in optimization papers: it optimizes the same secrecy-rate objective that it evaluates, which is the normal procedure for benchmarking an algorithm. However, the quantitative claims are currently under-supported because the threat model and numerical methodology are incompletely specified.","major_comments":[{"comment":"The optimization problem (12) requires exact eavesdropper channels: constraints (12g)-(12h) and the objective (12a) depend on the eavesdropper rates R^e_{ke,kl} defined in (10), which in turn require exact knowledge of g_{ke}(R). The manuscript never states whether eavesdropper CSI is available, nor does it discuss how a passive adversary's channel would be obtained. Without this premise, the computed beamformers and ME positions are not implementable as a secure communication scheme, and the simulated secrecy rates are upper bounds; the claimed ME-STARS gain over FPE-STARS is therefore not established for the intended threat model. Please state the CSI/threat-model assumption explicitly and either justify it (e.g., an active or untrusted eavesdropper whose channel can be estimated) or add a robust formulation and clearly interpret the current results as benchmark upper bounds.","section":"Section II-A and II-B, Eqs. (10) and (12)"},{"comment":"The numerical results are presented as single curves without any statement of the number of random channel realizations, Monte Carlo averaging, or confidence intervals. Since the users and eavesdroppers are 'randomly scattered' and the channel path responses are random, the quantitative claim in Fig. 4 of 'around 25%' improvement over FPE-STARS is not statistically supported. Please provide the averaging procedure, error bars or standard deviations, and results over multiple random initializations of the non-convex algorithm so that the abstract claims can be assessed.","section":"Section IV, Figs. 3-7"},{"comment":"The convergence argument for Algorithm 4 is incomplete. Inequality (a) in (50) asserts that the ME-position update increases the original sum secrecy rate R_sum, but Algorithm 1 performs gradient ascent on the penalized surrogate G in (21), which contains nonzero penalty and smoothing terms whenever constraints are active; maximizing G need not increase R_sum. The outer loop in Algorithm 1 terminates upon feasibility, not necessarily with a monotone increase of R_sum. Please provide a rigorous monotonicity argument, or alternatively state convergence to a stationary point of a penalized problem, and verify the monotonic behavior empirically.","section":"Section III-D, Eq. (50)"},{"comment":"Theorem 1 is load-bearing for the claim that the SDR relaxation of problem (48) is tight, since it justifies recovering rank-one beamforming vectors through Cholesky decomposition. However, the proof is omitted with only a pointer to [14, Appendix A]. For a self-contained journal paper, please include the proof or state explicitly that this is an adaptation of a known result; without this, the rank-recovery step is not verified within the manuscript.","section":"Section III-C, Theorem 1"}],"minor_comments":[{"comment":"The phrase 'the the active' appears in the abstract and again in the introduction to Section III; it should read 'the active'.","section":"Abstract and Section III"},{"comment":"The text says constraint (12f) ensures that distances 'do not exceed D0', but the inequality in (12f) is a minimum-separation constraint. Please reword as 'are no smaller than d0' and unify the notation d0 (used in the equations) with D0 (used in the numerical section).","section":"Section II-B, Eq. (12f) and Section IV"},{"comment":"The baseline list jumps from 'Baseline scheme 2 (ME-RIS)' to 'Baseline scheme 4 (RPE-STARS)'; there is no Baseline scheme 3. Please renumber the baselines.","section":"Section IV, baseline list"},{"comment":"The title of Algorithm 3 reads 'BE active beamforming subproblem'; it should be 'BS active beamforming subproblem'.","section":"Algorithm 3 title"},{"comment":"The dependence of the mode index \\vartheta on the link side is not made explicit: for a given legitimate user and a given eavesdropper, the correct coefficient matrix is \\Phi_t or \\Phi_r depending on which side of the STARS the terminal lies. Please make this dependence explicit in (25)-(32) so that the formulation is unambiguous.","section":"Section III-B, Eqs. (25)-(32)"},{"comment":"The noise variance appears as \\sigma_l in the denominator of (A.7) but is defined as \\sigma_l^2 in Section II-A. Please use the squared notation consistently.","section":"Appendix A, Eq. (A.7)"},{"comment":"In the third row of the matrix in (13), the subscript 'S,p' should presumably be 'S,k' for consistency with the other rows.","section":"Eq. (13)"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of the journal and the algorithmic framework appears coherent, but the missing eavesdropper-CSI premise and the absence of any statistical support in the numerical section are substantial for the paper's central claims. Both issues are fixable in revision, so I do not recommend rejection. The convergence and Theorem 1 gaps should also be addressed for completeness."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a solid, incremental optimization paper. The genuinely new part is applying the authors' own ME-STARS concept to physical layer security under full-space eavesdropping; that combination is not in the cited literature. The AO/SDR machinery is standard and the derivations look coherent. The saturation of the secrecy gain with movable-region size is a useful practical insight.\n\nThe soft spot that matters: the whole formulation assumes perfect knowledge of the eavesdropper channels. Constraints (12g)-(12h) and the secrecy rate (11) depend on exact g_{k_e}(R), and the paper never says where that comes from. A passive eavesdropper doesn't feed back CSI, so the simulated sum secrecy rate is an upper bound, not an achievable secure rate. The stress-test note is right about this. It doesn't break the optimization math, but it changes the strength of the abstract's claim. The 25% ME gain over FPE-STARS is established only against known eavesdropper channels; under uncertainty the relative gain could shrink or even reverse. The authors should state the assumption, add a robust variant or an estimation discussion, or at minimum label the curves as upper bounds.\n\nTwo smaller issues. First, the numerical section shows single secrecy-rate curves with no Monte Carlo averaging or error bars; the 25% figure may be one channel realization. Second, Theorem 1's proof is deferred to [14, Appendix A]; that's acceptable if the cited result covers this exact setup, but it should be verified rather than taken on faith. No code or data is provided, which makes the single-curve plots harder to check.\n\nIf the authors fix the CSI disclosure and add statistical averaging, this becomes a reasonable subfield contribution. I'd send it to review: the problem is well-posed, the approach is competent, and the saturation insight is worth having on record. I'd want the referee to push on the CSI point specifically.","headline":"Competent incremental optimization paper; the 25% ME-STARS secrecy gain is an upper bound until the perfect-eavesdropper-CSI assumption is disclosed and tested.","tokens_in":21216,"tokens_out":2016,"would_cite":true,"duration_ms":23582,"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":"This paper claims that letting the elements of a simultaneously transmitting and reflecting surface move inside a small region measurably improves secrecy rates against full-space eavesdropping, with the gain saturating once the movable…","keywords":["movable elements","STARS","physical layer security","sum secrecy rate","position optimization","passive beamforming","full-space eavesdropping","alternating optimization"],"falsifier":"Simulate the proposed optimization with eavesdropper channels known only up to an estimation error, or with only their statistical distribution available, and compare the resulting achievable secrecy rate against the fixed-position STARS baseline. If the ME-STARS advantage disappears or reverses under imperfect eavesdropper channel knowledge, the paper's central claim of a significant implementable secrecy gain would be falsified.","tokens_in":20286,"feed_emoji":"📡","tokens_out":5109,"duration_ms":56330,"temperature":0.7,"pith_summary":"This paper tries to establish that allowing the elements of a simultaneously transmitting and reflecting surface (STARS) to move inside a small region materially improves physical-layer security in a full-space eavesdropping scenario, and that the gain saturates once the movable region reaches a modest size. The authors formulate a sum secrecy rate maximization problem that jointly decides where the surface elements sit, how the surface splits and phase-shifts transmitted and reflected signals, and how the base station beams to two legitimate users, while constraining information leakage to two eavesdroppers on both sides of the surface. They solve it by alternating a gradient-ascent position update with successive-convex-approximation beamforming updates. If the claim holds, the practical message is that movable elements are worth deploying for secrecy even though the elements only travel a few wavelengths, and that enlarging the aperture beyond a few wavelengths yields little additional secrecy benefit.","feed_headline":"Movable STARS elements lift secrecy rate by 25 percent","feed_subtitle":"Jointly moving and tuning a smart surface beats fixed-element designs, and the gain plateaus after a small aperture.","key_machinery":"The carrying mechanism is the element position matrix $\\mathbf{R}$ together with a field-response channel model in which only the phases of the BS-STARS and STARS-user/eavesdropper channels change as elements move, while angles and gains stay constant. The algorithm optimizes $\\mathbf{R}$ by a gradient ascent that maps unconstrained variables through a hyperbolic tangent into the confined region, handles minimum-distance and rate constraints by a penalty method with log-sum-exp smoothing, and alternates this with SCA-based updates of the STARS transmission/reflection coefficients and the base station beamforming. This decomposition turns the intractable coupled nonconvex problem into three solvable blocks whose successive improvement drives the overall secrecy rate upward.","core_discovery":"The paper's central claim is that a STARS whose elements can be repositioned within a confined region ('ME-STARS') significantly outperforms a conventional STARS with fixed-position elements in secrecy rate, and that the sum secrecy rate saturates within a limited movable-region size. In the simulated setup, ME-STARS beats fixed-position STARS by about 25 percent at the same transmit power, also outperforming two separate movable-element RISs and a random-position STARS. The saturation is explained by the limited number of locally optimal element positions available inside a small region: once the region is large enough to accommodate those positions, extra aperture gives no further gain. The paper therefore argues both for the value of position optimization at the surface and for a design rule on how large the movable region needs to be.","pith_inferences":["The exact-eavesdropper-CSI premise is likely the main gap between the computed rates and a deployable scheme; a statistical or robust formulation over the eavesdropper channel distribution is a natural extension.","The saturation result suggests a rule of thumb for aperture sizing that could be tested in other MA/STARS settings: choose the region size where the marginal gain flattens rather than maximizing aperture.","Because the gain is tied to small-scale fading richness, the advantage over fixed-position STARS may shrink in strongly line-of-sight environments; a LoS-only simulation would clarify how much of the gain is positional diversity.","The same alternating framework could be extended to secrecy outage probability as the objective, which would convert the claimed rate gain into a reliability statement without assuming eavesdropper CSI."],"forward_implications":["A fixed number of surface elements can yield a secrecy-rate gain without adding RF chains, simply by repositioning elements within a few wavelengths.","The movable region can be designed small: once its size reaches a few wavelengths, further expansion adds little secrecy rate, which guides aperture sizing in practice.","The ME gain grows with the number of propagation paths, because richer scattering creates more local optima for element positions to exploit.","Optimized positions, not just mobility, are what matter: ME-STARS beats random-position STARS, so the gradient-ascent position update carries real value."],"supporting_citations":[{"why":"Supplies the field-based channel model for movable-antenna systems that the position-dependent channel representation builds on.","marker":"[8]"},{"why":"Defines the field response vector used to express channels as functions of the movable element positions.","marker":"[9]"},{"why":"Introduces the STARS concept of simultaneous transmission and reflection that this system extends to secure communications.","marker":"[4]"},{"why":"Prior ME-STARS communication framework that this paper extends into the physical-layer security setting.","marker":"[32]"},{"why":"Provides the rank-one recovery argument used in the active beamforming subproblem and the STAR-RIS security baseline.","marker":"[14]"},{"why":"Supplies the penalty method used to handle nonconvex constraints in both position and passive beamforming subproblems.","marker":"[33]"},{"why":"Supplies the backtracking line search used for step-size selection in the gradient-ascent position update.","marker":"[34]"}],"fun_headline_variants":["Movable STARS elements boost secrecy without extra power","Position-tuning a smart surface yields 25% secrecy gain","Secrecy rate saturates: key insight from movable STARS","Small aperture moves enough: ME-STARS secrecy gain plateaus","Moving STARS elements beats fixed ones but only to a point"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the transmitter knows the eavesdroppers' channels exactly when optimizing; without exact eavesdropper channel state information, the computed secure beamforming and the claimed movable-element gain are an upper bound rather than an achievable system performance.","fun_headline_variants_meta":{"raw":{"variants":["Movable STARS elements boost secrecy without extra power","Position-tuning a smart surface yields 25% secrecy gain","Secrecy rate saturates: key insight from movable STARS","Small aperture moves enough: ME-STARS secrecy gain plateaus","Moving STARS elements beats fixed ones but only to a point"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000665,"raw_usage":{"total_tokens":3015,"prompt_tokens":904,"completion_tokens":2111,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":2026}},"tokens_in":520,"tokens_out":2111,"duration_ms":16366,"temperature":1.0,"reasoning_tokens":2026,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:48:00.520296+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the proposed optimization with eavesdropper channels known only up to an estimation error, or with only their statistical distribution available, and compare the resulting achievable secrecy rate against the fixed-position STARS baseline. If the ME-STARS advantage disappears or reverses under imperfect eavesdropper channel knowledge, the paper's central claim of a significant implementable secrecy gain would be falsified.","supporting_citations":[{"cited_title":"STAR: Simultaneous transmission and reflection for 360° coverage by intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Introduces the STARS concept of simultaneous transmission and reflection that this system extends to secure communications."},{"cited_title":"Exploiting Movable-Element STARS for Wireless Communications","cited_arxiv_id":"2412.19974","evidence_quote":"Prior ME-STARS communication framework that this paper extends into the physical-layer security setting."},{"cited_title":"Security enhancement for coupled phase-shift STAR-RIS networks,","cited_arxiv_id":null,"evidence_quote":"Provides the rank-one recovery argument used in the active beamforming subproblem and the STAR-RIS security baseline."},{"cited_title":"Nocedal and S","cited_arxiv_id":null,"evidence_quote":"Supplies the penalty method used to handle nonconvex constraints in both position and passive beamforming subproblems."},{"cited_title":"Boyd and L","cited_arxiv_id":null,"evidence_quote":"Supplies the backtracking line search used for step-size selection in the gradient-ascent position update."}],"review_version":1}