{"id":"4bad76dd-c716-4d12-8868-402a11fffb90","arxiv_id":"2607.23249","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A Markov-mixture cross-sparsity prior plus posterior-weighted movable-antenna placement is claimed to reduce multi-target angular RMSE in multipath sensing.","lead":"This paper combines a specialized Bayesian prior with movable-antenna placement to find multiple targets when radar echoes bounce between targets before returning. The authors claim this cuts angular location error in multipath settings that ordinary fixed-antenna arrays handle poorly.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Taylor-ellipse mainlobe constraint and Dquad projection are used outside their region of validity; a single analytic consistency check can settle whether Proposition 1 controls the actual 6 dB contour.","rationale":"The reader's weakest_assumption exactly matches the most load-bearing vulnerability: Proposition 1's Taylor-ellipse approximation is used as both a hard constraint and a projection target, yet its validity at the optimized operating point is never verified. The paper has independent value in the proposed prior and ambiguity-function framework, and no internal inconsistency is evident beyond the omitted terms explicitly acknowledged in Appendix A. The concern is testable with a simple numerical comparison, so the verdict should remain CONDITIONAL — it identifies a gap in evidence, not a demonstrated flaw. Agreement: the reader and I independently converge on the same weakest assumption; my contribution is to name a specific quantitative check that would resolve it.","tokens_in":22886,"tokens_out":1418,"duration_ms":11797,"concrete_test":"For the optimized arrays reported in Figs. 9-10, numerically compute the true 6 dB contour of |chi(Delta f_t, Delta f_r)|^2 from Eq. (21) (no Taylor truncation) and compare it with the ellipse of Eq. (22) using the reported Mt=Mr=8, Dt=Dr=8 lambda, Q=16 settings. Compute the maximum relative error in Wmain (minor axis) and the maximum contour mismatch over angles. If the error exceeds a threshold (e.g., 10% in Wmain or visible mismatch in the 6 dB contour), Proposition 1 is not valid at the optimized operating point; rerun the optimization with an exact mainlobe-width constraint to see if the RMSE/ambiguity results change materially.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claim is that optimizing MA positions via the posterior-weighted ambiguity function and DPGD reduces multi-target angular RMSE in multipath environments. The load-bearing assumption identified by the reader is the use of the local second-order Taylor ellipse for the 6 dB mainlobe contour (Proposition 1, Eq. (22), Appendix A). This approximation is not merely descriptive: it is enforced as a hard constraint Wmain >= 2√2/Q and as the basis for the projection ProjDquad in Section V-B.2. The proof in Appendix A drops fourth-order terms Delta f_t^4, Delta f_r^4, and the coupling term sigma_t^2 sigma_r^2 Delta f_t^2 Delta f_r^2. After optimization, the arrays may have large aperture, producing a narrow mainlobe; the operating region may violate the small-offset assumption. If the approximation breaks down, the constraint and projection do not control the actual mainlobe width, and the feasible-set geometry used in DPGD is not what is claimed. The paper provides no numerical or analytic check that the second-order approximation reproduces the true 6 dB contour at the optimized positions; only Fig. 9 shows a qualitative ambiguity function comparison, with no overlay of the predicted contour. This is a concrete, correctness-relevant gap: a wrong constraint could bias the optimized positions and degrade the claimed RMSE gains.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes an MA-enabled multi-target wireless sensing scheme for multipath environments. It introduces a cross-sparsity Markov mixture prior to estimate coarse target locations via the SF-TVBI Bayesian inference framework, constructs a posterior-probability-weighted 2D ambiguity function as a surrogate for inter-target interference, and optimizes the MA positions to suppress sidelobes at the target-offset bins and to control the mainlobe width. The resulting nonconvex problem is solved by a Dykstra-based projected gradient descent (DPGD) algorithm. The central claims are that the mixture prior improves RMSE/detection probability by about 5 dB over the cross-sparsity baseline in the low-to-moderate SNR regime, that the posterior-weighted ambiguity function outperforms the conventional one for MA optimization, and that DPGD approaches the RMSE of an interior-point-method PGD with substantially lower complexity. These claims are supported by Monte Carlo simulations (Figs. 7–11) without an analytical bound connecting the surrogate objective to estimation error.","tokens_in":23375,"tokens_out":6713,"duration_ms":61590,"significance":"If the claims hold, the paper makes a useful contribution to MA-enabled sensing by showing that coarse posterior target-location information can be injected into an ambiguity-shaping criterion and optimized with a low-complexity projected-gradient method. The system model explicitly includes both LoS and first-order NLoS paths, which is more realistic than single-path models. The paper also provides a closed-form mainlobe-width relation (Proposition 1) and a cross-sparsity Markov mixture prior that appears to give a consistent estimation gain in simulation. However, the load-bearing assumptions are not fully verified: the objective function and its discretized gradient are inconsistent; the Taylor-ellipse approximation of the mainlobe is used as a hard constraint without a validity check at the optimized configurations; and the posterior weights are produced and evaluated by the same inference pipeline, so the source of the reported gain is not isolated. These issues are addressable but require substantive revision.","major_comments":[{"comment":"The objective and the gradient are internally inconsistent. From Eq. (29), χp is equal to p(P_{k,f})·χ on each bin (k,f). Therefore |χp|² = p(P_{k,f})²|χ|², and the continuous objective in Eq. (32) is Σ_{k,f} p(P_{k,f})² ∫_{C_{k,f}} |χ|². However, the Riemann-sum approximation in Eq. (34) inserts an additional p(P_{k,f}) factor, yielding p(P_{k,f})³ weighting, and the derivatives in Eqs. (41)–(42) contain only one p(P_{k,f}) factor, matching an objective proportional to p(P_{k,f})|χ|² rather than p(P_{k,f})²|χ|². As written, Algorithm 1 is not minimizing the stated J. This needs to be corrected and the implemented objective clearly specified.","section":"§IV-C and §V-A, Eqs. (29), (32), (34), (41)"},{"comment":"The 6 dB mainlobe contour is obtained by dropping the fourth-order terms Δf_t^4, Δf_r^4, and σ_t²σ_r²Δf_t²Δf_r² in Appendix A. This approximation is load-bearing because it defines the hard constraint Wmain ≥ 2√2/Q and the Dquad projection. The paper gives no check that the second-order ellipse reproduces the true 6 dB contour after optimization. For the simulation parameters (Q=16, initial half-wavelength ULA with Mt=Mr=8), the ellipse half-width is roughly √((1-ρ6)/σ²) ≈ 0.07, but evaluating the true 1D array response at that offset gives |S(Δ)|² ≈ (sin(4πΔ)/(8sin(πΔ/2)))² ≈ 0.77, close to 0.25. Thus the ellipse can substantially under-estimate the actual contour. Please include a direct numerical overlay of Eq. (22) on the true 6 dB contour at the optimized positions, or replace the constraint with a more robust characterization.","section":"Proposition 1 and Appendix A; constraint in §IV-C and projection in §V-B.2"},{"comment":"The offset-bin weights p(P_{k,f}) in Eq. (28) are derived from the posterior q(s) of the SF-TVBI stage, and the same SF-TVBI pipeline is then used to evaluate the final RMSE in Figs. 7–11. The paper does not provide a test with mismatched or corrupted posterior weights, nor an independent coarse-localization method, so the reported gain of DPGD with the proposed ambiguity function (DPGD-PAF) over DPGD with the traditional one (DPGD-TAF) could come in part from the posterior information itself rather than from the ambiguity-shaping objective. A sensitivity study or a decoupled validation would make the central claim more convincing.","section":"§III–§IV, posterior weighting and final evaluation"}],"minor_comments":[{"comment":"Typos: 'ProjDploy' should be 'ProjDpoly' (in §V-B and Algorithm 1); 'DGPD' in Section VII should be 'DPGD'.","section":"Throughout"},{"comment":"The sign in the definition Δf_t ≜ f_t − f'_t in Eq. (22) is opposite to the definition Δf_t ≜ f'_t − f_t used in Eqs. (20)–(21). The magnitude is unchanged, but the notation should be made consistent.","section":"Eq. (20)–(22)"},{"comment":"The complexity comparison with IPM-PGD depends on the inner iteration count K2 of the Dykstra projection. Since the actual K2 used in simulations is not reported, the 'substantially lower complexity' claim is not fully supported. Please give typical K2 values or a measured runtime comparison.","section":"Section V-C (complexity)"},{"comment":"The core inference algorithm (SF-TVBI) is cited from reference [14] by overlapping authors, and no code or data is released. Given that the proposed prior modifies that algorithm, pseudo-code for the prior updates or a code release would help reproducibility.","section":"Reproducibility"}],"recommendation":"major_revision","confidential_remarks":"The central idea is promising and fits the journal's scope, but the objective/gradient inconsistency and the unvalidated Taylor-ellipse constraint are load-bearing correctness issues that must be resolved before publication. The posterior-circularity concern is also important because the headline gains could be an artifact of the evaluation protocol rather than the optimization. These are fixable in a revision, so I do not recommend rejection at this stage."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper has two genuinely new pieces — the Markov mixture extension of the cross-sparsity prior to suppress false NLoS-to-LoS coupling, and the posterior-probability weighting of the 2D ambiguity function so MA placement suppresses sidelobes at the actual target separations. Both are clearly explained, and the reported gains (about 5 dB over the prior cross-sparsity model at low/moderate SNR, and a few dB RMSE improvement over ULA) are consistent with the mechanism rather than suspicious. The DPGD algorithm is a reasonable engineering choice: replacing the projection onto D = Dpoly ∩ Dquad with Dykstra iterations, and using PAV for the isotonic part, is sound and the complexity reduction is plausible.\n\nThe soft spots are real. First, the objective is not what it says. In Eq. (29), chi_p already contains the posterior weight p(P_{k,f}). Since the indicator functions partition the offset domain, |chi_p|^2 inside bin (k,f) is p(P_{k,f})^2 |chi|^2. Eq. (32) therefore equals sum p^2 ∫|chi|^2, but the Riemann sum in Eq. (34) and the gradients in Eqs. (41)-(42) carry an extra outer p(P_{k,f}), effectively optimizing p^3|chi|^2. That changes the weighting. This is not a typo-level concern because the whole point is to emphasize posterior-likely offsets; the current equations do not match the stated objective.\n\nSecond, Proposition 1's second-order Taylor ellipse is used as a hard feasibility constraint and in the Dquad projection, but the paper never checks whether the dropped fourth-order terms are small at the optimized positions. After optimization the aperture is wide and the mainlobe narrow, so the operating region may be outside the quadratic regime. A single plot overlaying the predicted 6 dB ellipse on the true |chi|^2 contour at the optimized arrays would settle this. Without it, the constraint is a claim, not a verified control on mainlobe width.\n\nThird, the empirical claims are simulation-only: no error bars, no released code or full parameter settings, and the core inference relies on arXiv:2511.14051 from the same group, whose algorithm is not independently verified. That does not sink the paper, but it makes the 5 dB prior gain conditional.\n\nThis is not a manufactured-flaw situation. The central idea is plausible and the paper reads as honest. It deserves a serious referee, but I would not accept it as is. The referee should ask for (a) a consistent objective/gradient derivation, (b) a numerical check of the mainlobe ellipse at optimized positions, and (c) error bars or released code. The likely audience is people working on MA/ISAC and multipath estimation; the posterior-weighted ambiguity function is the idea most worth carrying forward.","headline":"Incremental but real extension of the authors' cross-sparsity framework to movable antennas; the posterior-weighted ambiguity function is worth attention, but the optimized objective has a weight inconsistency and the mainlobe-ellipse constraint is used without a validity check.","tokens_in":23778,"tokens_out":3443,"would_cite":true,"duration_ms":32219,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A12","90C26"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that repositioning movable antennas to suppress the specific sidelobes where actual target separations fall can cut multi-target angle-estimation error in multipath environments, and it supplies a low-complexity algorithm","keywords":["movable antennas","wireless sensing","multi-target estimation","multipath propagation","ambiguity function","antenna position optimization","structured sparsity","Bayesian inference"],"falsifier":"Take the optimized antenna positions and compute the exact 6 dB contour of |χ|² from Eq. (21) without Taylor truncation. If the minor-axis width differs materially from 2√((1−ρ₆)/σ²) or falls below 2√2/Q, the mainlobe constraint in problem (31) is not enforcing what the paper claims. A second check: compare RMSE for target separations inside one grid cell versus outside; the method's predicted advantage should vanish when the weighted sidelobes are numerically zero.","tokens_in":22757,"feed_emoji":"📡","tokens_out":4111,"duration_ms":46014,"temperature":0.7,"pith_summary":"The paper is trying to establish that in a multipath wireless sensing scene with several targets, the positions of movable transmit and receive antennas can be optimized against a posterior-probability-weighted two-dimensional ambiguity function, so that the array stops scattering ambiguity energy at the exact angular offsets separating the targets. A sympathetic reader would care because this turns antenna placement into a targeted interference-suppression tool instead of a generic sidelobe-reduction exercise, and the paper's simulations show the resulting RMSE gain grows as targets become more numerous. The paper also argues that a Markov-mixture extension of the cross-sparsity prior fixes a specific failure mode of cross-sparsity modeling: false alarms on first-order non-line-of-sight paths corrupting the direct-path estimates. If these claims hold, movable antennas are not just an extra knob for existing sensing metrics; they let the array adapt its geometry to the particular target constellation it is trying to resolve.","feed_headline":"Move antennas to silence the sidelobes that corrupt multipath sensing","feed_subtitle":"Posterior-weighted ambiguity shaping lowers multi-target angle error, and a cheap algorithm matches an expensive baseline.","key_machinery":"The load-bearing object is the 2D posterior-probability-based ambiguity function, defined as a bin-wise weighting of the conventional two-dimensional ambiguity function by posterior probabilities of target-separation offsets. Each offset bin carries the probability that at least one pair of coarse-estimated targets has that relative spatial frequency offset, so the optimization concentrates on sidelobes that actually cause inter-target leakage. Around it sits a second piece: the mainlobe-width characterization, which approximates the 6 dB contour of the ambiguity function as an ellipse whose axes are set by the variances of the transmit and receive antenna position distributions, and which i","core_discovery":"The central claim is that multi-target sensing accuracy in multipath is governed less by the full ambiguity surface than by the ambiguity values at the offsets corresponding to actual pairwise target separations. The paper constructs a 2D posterior-probability-based ambiguity function, where each offset bin is weighted by the posterior probability that some pair of detected targets is separated by that offset. It then optimizes antenna positions to minimize the weighted integrated sidelobe energy while preserving a minimum mainlobe width tied to the grid spacing. The paper shows, via Monte Carlo simulation, that this targeted sidelobe suppression lowers angular RMSE relative to both uniform","pith_inferences":["Editorial inference: the posterior-weighting scheme inherits its own coarse estimates, so at very low SNR the weights could emphasize wrong offsets and steer antennas incorrectly; a two-pass refinement that re-optimizes after a second estimation round would test this and could make the scheme self-correcting.","Editorial inference: the same variance-based mainlobe formula could be ported to other reconfigurable-antenna architectures (e.g., fluid antennas or sparse subarray selection), since the constraint depends only on the resulting spatial distribution, not on the mechanism that produces it.","Editorial inference: the paper compares RMSE but not against a multi-target Cramér-Rao bound; a bound-based comparison would separate how much of the gain comes from sidelobe suppression versus how much from mainlobe narrowing, which the current ambiguity-function objective cannot distinguish.","Editorial inference: because the ambiguity function factorizes into transmit and receive sums, the optimization decouples in structure; a testable extension would be to optimize transmit and receive arrays independently or to extend the weighting to higher-order NLoS paths when their offsets are also identifiable."],"forward_implications":["If the central claim is right, movable-antenna placement for multi-target sensing should be formulated per scenario: the optimal aperture is the one that not only narrows the mainlobe but also kills sidelobes at the detected target separations.","The proposed prior fixes a concrete failure mode of cross-sparsity modeling, so multi-target estimators that rely on LoS-NLoS coupling become less fragile to false alarms in low-SNR multipath.","The explicit ellipse relation links mainlobe width directly to the variance of antenna positions, giving a simple design rule: spread antennas about a centroid while keeping the spatial variance within the grid-resolution constraint.","The complexity advantage of the proposed algorithm means antenna-position optimization becomes feasible in near-real-time or resource-constrained sensing nodes.","As target number grows, the gain over fixed arrays widens, pointing to movable antennas as especially valuable in dense multi-target scenarios."],"fun_headline_variants":["Sidelobe-weighted antenna moves sharpen multipath target sensing","Movable antennas suppress weighted sidelobes for multi-target accuracy","Posterior-weighted ambiguity shaping cuts multipath angle errors","Antenna position optimization reduces sidelobe blur in multipath sensing"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof that the 6 dB mainlobe boundary is an ellipse keeps only second-order Taylor terms and drops fourth-order terms and the coupling term; if that local quadratic model stops holding after antennas spread to a large aperture, the mainlobe-width constraint and projection step would not control the actual mainlobe as claimed.","fun_headline_variants_meta":{"raw":{"variants":["Sidelobe-weighted antenna moves sharpen multipath target sensing","Movable antennas suppress weighted sidelobes for multi-target accuracy","Posterior-weighted ambiguity shaping cuts multipath angle errors","Antenna position optimization reduces sidelobe blur in multipath sensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000215,"raw_usage":{"total_tokens":1279,"prompt_tokens":774,"completion_tokens":505,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":518,"completion_tokens_details":{"reasoning_tokens":437}},"tokens_in":518,"tokens_out":505,"duration_ms":5280,"temperature":1.0,"reasoning_tokens":437,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T23:58:13.266750+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the optimized antenna positions and compute the exact 6 dB contour of |χ|² from Eq. (21) without Taylor truncation. If the minor-axis width differs materially from 2√((1−ρ₆)/σ²) or falls below 2√2/Q, the mainlobe constraint in problem (31) is not enforcing what the paper claims. A second check: compare RMSE for target separations inside one grid cell versus outside; the method's predicted advantage should vanish when the weighted sidelobes are numerically zero.","supporting_citations":[],"review_version":1}