{"id":"e6a475c4-b0c6-46f3-96b6-24d8267f0a4a","arxiv_id":"2505.07035","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Movable-antenna position optimization under imperfect channel knowledge reduces to maximizing the estimated channel norm, solvable optimally by a graph algorithm.","lead":"This paper designs where to place movable antennas when the wireless channel is not known perfectly, and shows that the best strategy is still to point the signal along the estimated channel. This matters because movable antennas are a candidate for next-generation wireless, and real systems never have perfect channel knowledge.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Random-case MRT optimality is only conditional; for small outage probabilities with ymax<y1, the approximated problem's optimum is not MRT, contradicting the abstract's unconditional claim.","rationale":"The reader flagged the Bernstein-type inequality's tightness as the weakest assumption. The present review finds a more specific, internally documented gap: the paper's own analysis shows that MRT optimality in the random-error case holds only when ymax≥y1. For small outage probabilities (0<ρ<e^{-1/2}), which is the practically relevant regime, F(y) is convex with a decreasing-then-increasing shape, so the maximum over [0,ymax] need not be at ymax. The paper acknowledges this and states that if ymax≤y1 the maximum R0 is negative due to the approximation, but it does not bound the approximation error, does not quantify when ymax≥y1, and still claims in the abstract and conclusion that MRT is optimal in both scenarios. This makes the headline claim conditional in a way that is not disclosed. However, the underlying graph-based channel-norm maximization may still be the right algorithm for the true outage problem, because with MRT the true outage probability is stochastically increasing in ||h_est||²; the paper just does not prove this and instead relies on a conservative bound. Therefore the verdict should remain conditional, not clean accept, but the concern is about proof completeness and overclaiming rather than a demonstrated false result.","tokens_in":9110,"tokens_out":17266,"duration_ms":180146,"concrete_test":"Use the paper's own condition: for the setup of Fig. 4 with ρ=0.01, compute y1=σ²/2·(2ln(1/ρ)−1) and, for each Monte Carlo realization, ymax=||h_est||². If any realization has ymax<y1, compare F(0)=σ²Pmax(1−√(2ln(1/ρ))) with F(ymax). If F(0)>F(ymax), the optimal beamformer for the approximated problem (17) is orthogonal to h_est, not MRT, directly contradicting the unconditional MRT claim. If all reported realizations satisfy ymax≥y1, then the conditional claim holds and the concern does not land in the paper's numerical regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section IV-B derives F(y) and shows it is convex. For 0<ρ<e^{-1/2}, F'(0)<0, so F first decreases on [0,y0] and then increases. Maximizing F over [0,ymax] therefore places the optimum at an endpoint. Only when ymax≥y1, where y1 is the positive zero of F, is the optimum attained at ymax with MRT. If ymax<y1, the optimal y is either 0 or ymax, and F(ymax) may be negative, an artifact of the conservative Bernstein-type bound. The paper does not prove ymax≥y1 for the considered scenarios, does not state this condition in the abstract or conclusion, and gives no algorithm for the complementary case. Thus the central claim that MRT is optimal for both CSI error models and that the graph-based channel-norm maximization optimally solves the true random-error problem (6) is not established as stated. Moreover, Lemma 1 is only a sufficient condition, so no gap bound links the solution of the approximated problem (17) to the true outage-constrained problem (6).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This letter considers robust movable-antenna (MA) position optimization for a MISO system under imperfect channel state information (CSI). Two CSI error models are treated: norm-bounded errors and Gaussian-distributed random errors. For the norm-bounded model, the worst-case received signal power is derived in closed form, and it is shown that maximum-ratio transmission (MRT) is optimal; the problem then reduces to maximizing the squared norm of the estimated channel, which is solved by a graph-based algorithm from the authors' prior work. For the random-error model, a Bernstein-type inequality is used to approximate the outage constraint by a deterministic lower bound, and it is argued that, under certain conditions, MRT remains optimal and the same graph-based channel-norm maximization applies. Numerical results show that the proposed MA scheme can outperform some fixed-antenna benchmarks operating under perfect CSI. The paper's core norm-bounded derivation is sound, but the random-error part contains overclaimed optimality and a gap between the approximated and true problems.","tokens_in":9322,"tokens_out":8932,"duration_ms":87575,"significance":"If the claims are properly qualified, the paper makes a useful contribution by extending MA position optimization to imperfect CSI, an important practical issue. The norm-bounded worst-case analysis (Section III) is clean: the lower bound (8) is tight with e = -δĥ/‖ĥ‖, and the reduction to channel-norm maximization via MRT is valid. The use of the graph-based solver from [12] gives an efficient and exact solution for the norm-bounded problem. The random-error part depends on an approximation, and the paper's unconditional statements about MRT optimality and optimal MA placement are not currently justified. With a revised presentation that spells out the conditions under which MRT is optimal and clearly separates the approximated problem from the true outage-constrained problem, the letter would be a solid contribution; as written, the central claim is overstated.","major_comments":[{"comment":"The claim that MRT is optimal for randomly distributed CSI errors is only conditional and contradicts the unconditional wording in the abstract ('we show the optimality of the MRT for imperfect CSI in both scenarios') and conclusion ('MRT is optimal for both types of CSI errors in general'). For 0 < ρ < e^{-1/2}, F(y) is convex and decreases on [0, y0] before increasing; hence the maximum of F(y) over [0, ymax] is attained at an endpoint, not necessarily at ymax. The paper itself states that if ymax < y1, the optimal y is either 0 or ymax, and the maximum R0 may be negative. In that case MRT is not optimal for the approximated problem (17), and the graph-based channel-norm maximization solving (12) is not the solution. The paper provides no proof that ymax ≥ y1 for the considered numerical scenarios and no algorithm for the complementary case. The abstract, conclusion, and Section V statements must be qualified, or an explicit condition (and a way to verify it) must be added.","section":"Section IV-B, Eqs. (20)-(22); Abstract and Conclusion"},{"comment":"The Bernstein-type inequality is only a sufficient condition for the outage constraint (15), so the optimal solution of the approximated problem (17) is not guaranteed to be optimal, or even feasible, for the true non-outage problem (6). The paper does not bound the gap between the two problems. Since the abstract and Section I describe the proposed scheme as obtaining 'the optimal MA positions' for the random-error case, this is a load-bearing gap. The authors should explicitly state that (17) is an approximation, present any available tightness guarantees from [14], or verify through numerical experiments that the approximation is tight in the considered operating regimes.","section":"Section IV-A, Lemma 1 and Eq. (16); Section IV-B, Eq. (17)"},{"comment":"There is an internal inconsistency between Lemma 1 as printed and the subsequent derivation. Equation (16) reads Tr(Q) − √(2 ln(1/ρ))(‖Q‖2 + 2‖r‖2) + s ≥ 0, while Eq. (18) defines ‖Q‖2 and ‖r‖2 as squared norms. Substituting those definitions into (16) gives σ²Pmax − √(2 ln(1/ρ))(σ⁴Pmax² + 2σ²Pmax²|ĥ^H ω0|²) + Pmax|ĥ^H ω0|² ≥ R0, which is not what Eq. (19) states. The standard Bernstein-type concentration inequality (as in [14]) involves the square root of ‖Q‖_F² + 2‖r‖², which is what Eq. (19) appears to use. This needs to be corrected in Lemma 1, and the notation for norms should be made unambiguous; as printed, the stated Lemma does not imply the convex approximation used in the rest of the paper.","section":"Section IV-A, Eq. (16) and Section IV-B, Eq. (18)-(19)"}],"minor_comments":[{"comment":"The notation ‖Q‖2 is overloaded: Eq. (18) computes ‖Q‖2 = σ⁴Pmax², which is the squared Frobenius norm, not a standard norm. Please use ‖Q‖_F² and ‖r‖² explicitly to avoid confusion.","section":"Eq. (18)"},{"comment":"The abstract claims the proposed scheme can outperform 'other benchmark schemes implemented under perfect CSI conditions,' but the numerical results in Fig. 4 only demonstrate outperformance relative to FPAs without antenna selection under perfect CSI. Please clarify the scope of the claim.","section":"Abstract and Section V, Fig. 4"},{"comment":"The sentence 'the MA position optimization helps enhance the channel norm ‖h‖², as well as the robustness against the CSI error for a given y1' is unclear; y1 is not a user-specified parameter but a function of σ² and ρ. Rephrase to state that a larger channel norm makes it more likely that ymax ≥ y1.","section":"Section IV-B, final paragraph"}],"recommendation":"major_revision","confidential_remarks":"The norm-bounded part is technically sound and the paper builds on a polynomial-time graph-based solver from the authors' prior work, which is a positive feature. However, the random-error part is currently overclaimed: the unconditional MRT-optimality statement is not supported by the paper's own analysis, and the approximation gap between problems (6) and (17) is not addressed. These issues are fixable within the scope of the manuscript, so I recommend major revision rather than rejection. I would also encourage the authors to make the auxiliary notation in Eq. (16) consistent with the standard Bernstein inequality used in Eq. (19)."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nHere's my read on arXiv:2505.07035. The norm-bounded CSI error part is genuinely good: the worst-case objective reduces to channel norm maximization via a tight triangle inequality, and MRT is optimal for that reduction. That is a clean, useful result, and the derivation is easy to follow.\n\nThe random-error part is more mixed. The Bernstein-type lower bound is standard and correctly applied, and the simplification to the scalar function F(y) is correct. But the claim that MRT is optimal for the random-error case only holds for the approximated problem, and only when the estimated channel norm is large enough relative to the error variance and outage probability. In Section IV-B the authors actually spell out the complementary case: for rho < e^{-1/2} and ymax < y1, the Bernstein bound goes negative and MRT is not the maximizer. That is an honest acknowledgment in the body, but it is missing from the abstract and conclusion, which say \"MRT is optimal for both scenarios\" without qualification. That is an overclaim that should be fixed.\n\nThe other soft spot is that Lemma 1 is only a sufficient condition, so there is no gap bound between the approximated problem and the true outage constraint. They partly address this by simulating actual outage performance, which looks reasonable, but the theoretical claim is limited.\n\nOverall, this is a reasonable incremental contribution to the movable-antenna literature. It extends the authors' prior graph-based position solver to imperfect CSI, and the numerical results make a plausible case that MA's spatial selection gain can offset CSI error. Nothing here is groundbreaking, but it is competently done, and the norm-bounded result is solid.\n\nWho is this for? Researchers working on movable or fluid antennas who need robust positioning under estimation error. For them it is worth reading. I would support sending this to peer review, with a required revision to qualify the MRT optimality claim and state the ymax >= y1 condition in the abstract.\n\nBest,\n[You]","headline":"Norm-bounded part is clean; random-error MRT optimality is conditional and the abstract overstates it.","tokens_in":9850,"tokens_out":5736,"would_cite":false,"duration_ms":54198,"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":"Under imperfect CSI, robust movable-antenna placement reduces to maximizing the estimated channel norm.","keywords":["movable antennas","robust optimization","imperfect CSI","Bernstein-type inequality","maximum-ratio transmission","antenna position optimization","outage probability","MISO systems"],"falsifier":"For a small instance (say $N=3$, $M=20$), enumerate all position subsets satisfying the minimum-distance constraint, compute the exact outage probability of the MRT beamformer for the Gaussian error model by Monte Carlo, and compare against the Bernstein-approximate optimum; if any subset yields a higher true non-outage SNR than the graph-based solution, the optimality claim for the random-error case fails.","tokens_in":8907,"feed_emoji":"📡","tokens_out":7397,"duration_ms":69637,"temperature":0.7,"pith_summary":"This letter asks whether movable-antenna (MA) systems keep their advantage when the channel estimates used to place the antennas are imperfect. For norm-bounded CSI errors it derives the worst-case received power in closed form, and for Gaussian-distributed errors it uses a Bernstein-type inequality to obtain a tractable lower bound on the non-outage received power. It then shows that in both cases maximum-ratio transmission (MRT) is optimal, so the robust placement problem collapses to the same task as in the perfect-CSI case: maximize the squared norm of the estimated channel over the allowed antenna positions. Because that task has an exact polynomial-time graph-based solution, the robust problem inherits it. A direct consequence argued in the paper is that optimized MA positions under imperfect CSI can outperform fixed-antenna arrays running under perfect CSI.","feed_headline":"Robust movable-antenna placement reduces to channel-norm maximization","feed_subtitle":"Even imperfect CSI lets movable antennas beat fixed ones by maximizing the channel norm.","key_machinery":"The load-bearing identity is the triangle-type lower bound $|\\omega^H(\\hat{\\mathbf h}+\\mathbf e)|^2 \\ge (|\\omega^H\\hat{\\mathbf h}|-|\\omega^H\\mathbf e|)^2$, with equality attainable by choosing $\\mathbf e=-\\delta\\hat{\\mathbf h}/\\|\\hat{\\mathbf h}\\|$, which converts the semi-infinite worst-case constraint into a one-dimensional shrinkage term. For the Gaussian model, the Bernstein-type inequality in Lemma 1 converts the probabilistic constraint into the convex deterministic inequality (16), which in turn becomes the scalar function $F(y)$ of the beamformed estimated-channel power $y$. The reduction works by showing $F$ is monotone in $y$ in the relevant regime, forcing MRT and leaving only the channel-norm maximization that the graph-based algorithm solves.","core_discovery":"The central claim is that robust MA position optimization in a MISO downlink has a common reduced form for the two standard imperfect-CSI models. For norm-bounded error with $\\|\\mathbf e\\|\\le\\delta$, the worst-case received power is $(|\\omega^H\\hat{\\mathbf h}|-\\delta\\|\\omega\\|)^2$ when positive and zero otherwise; with MRT this becomes $P_{\\max}(\\|\\hat{\\mathbf h}\\|-\\delta)^2$, so positions should maximize $\\|\\hat{\\mathbf h}\\|^2$. For Gaussian error, replacing the outage constraint by the Bernstein-type bound of Lemma 1 turns the problem into maximizing a scalar function $F(y)$ of $y=|\\hat{\\mathbf h}^H\\omega_0|^2$; whenever the feasible region is in the regime where $F$ is increasing (in particular for $\\rho\\ge e^{-1/2}$, or for small $\\rho$ when the estimated channel norm is large enough), the optimum again uses MRT and maximizes $\\|\\hat{\\mathbf h}\\|^2$. Thus the robust placement problem is optimally solved by the same graph-based algorithm used for perfect CSI, applied to estimated channels, and the resulting placement can offset the estimation error enough to beat fixed antennas under perfect CSI.","pith_inferences":["Going beyond the paper, if the Bernstein bound is loose, a non-MRT beamformer could improve the true outage performance in low-channel-norm regimes; evaluating the exact outage probability at the returned positions would test this.","Going beyond the paper, any CSI error model whose lower bound on received power is an increasing function of $\\|\\hat{\\mathbf h}\\|$ would inherit the same channel-norm placement rule, extending the graph algorithm's reach.","Going beyond the paper, the reduction is unlikely to survive multi-user or MIMO extensions, where beamformers interact; robust placement there would require a joint beamformer-position search."],"forward_implications":["For norm-bounded CSI errors, the worst-case SNR is computable in closed form and the optimal transmit vector is MRT; the only remaining decision is which sampling points the antennas occupy.","For Gaussian CSI errors under the stated conditions, the same MRT-plus-channel-norm-maximization structure holds for the Bernstein approximation, so the graph algorithm returns the optimal approximate positions in $O(NM^2)$ time.","Because MA position optimization enlarges the estimated channel norm, the robustness margin $\\|\\hat{\\mathbf h}\\|-\\delta$ is larger than for fixed antennas, which is why MA systems can beat perfect-CSI fixed antennas when the error threshold is moderate.","When the outage probability or CSI variance is large enough that the channel-norm regime condition fails, the approximation can yield a negative non-outage SNR bound, indicating the system should not rely on that placement and beamformer pair."],"supporting_citations":[{"why":"Supplies the graph-based fixed-hop shortest-path algorithm that optimally solves the channel-norm maximization, the core subproblem after the robust reductions.","marker":"[12]"},{"why":"Provides the Bernstein-type inequality used in Lemma 1 to convert the probabilistic outage constraint into the convex deterministic bound (16).","marker":"[14]"},{"why":"Supplies the field-response channel model used in the numerical evaluations to generate channel maps for MA positions.","marker":"[11]"}],"fun_headline_variants":["Robust MA placement: maximize channel norm despite CSI errors","Imperfect CSI? Movable antennas still maximize channel norm","MA position optimization: norm maximization is robust to CSI","Even with imperfect CSI, MA placement is norm maximization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Bernstein-type bound in Lemma 1 is tight enough that optimizing the approximate problem (17) still gives genuinely optimal, or near-optimal, MA positions and MRT for the true outage-constrained problem (6); the paper proves only a sufficient condition and does not bound the gap.","fun_headline_variants_meta":{"raw":{"variants":["Robust MA placement: maximize channel norm despite CSI errors","Imperfect CSI? Movable antennas still maximize channel norm","MA position optimization: norm maximization is robust to CSI","Even with imperfect CSI, MA placement is norm maximization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000225,"raw_usage":{"total_tokens":1483,"prompt_tokens":985,"completion_tokens":498,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":601,"completion_tokens_details":{"reasoning_tokens":431}},"tokens_in":601,"tokens_out":498,"duration_ms":5607,"temperature":1.0,"reasoning_tokens":431,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:28:06.935070+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a small instance (say $N=3$, $M=20$), enumerate all position subsets satisfying the minimum-distance constraint, compute the exact outage probability of the MRT beamformer for the Gaussian error model by Monte Carlo, and compare against the Bernstein-approximate optimum; if any subset yields a higher true non-outage SNR than the graph-based solution, the optimality claim for the random-error case fails.","supporting_citations":[{"cited_title":"Movable-an tenna po- sition optimization: A graph-based approach,","cited_arxiv_id":null,"evidence_quote":"Supplies the graph-based fixed-hop shortest-path algorithm that optimally solves the channel-norm maximization, the core subproblem after the robust reductions."},{"cited_title":"Outage constrained robust transmit o ptimization for multiuser MISO downlinks: Tractable approximations by conic optimization,","cited_arxiv_id":null,"evidence_quote":"Provides the Bernstein-type inequality used in Lemma 1 to convert the probabilistic outage constraint into the convex deterministic bound (16)."}],"review_version":1}