{"id":"5ae76a34-12b4-4d36-a9f5-8486d63ad1ec","arxiv_id":"2502.09023","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A fractional-programming BCD algorithm for jointly optimizing beamformers, receive filters, movable-antenna positions, and RIS coefficients raises radar SINR by about 5 dB over fixed-position antennas in simulations.","lead":"This paper designs a joint beamforming, antenna-position, and RIS-phase optimization algorithm for a radar-and-communication base station with movable antennas, aiming to maximize radar signal quality while keeping communication links working. The reported gains over fixed antennas are roughly 5 dB in radar SINR, which matters for 6G systems that combine sensing and communication in blocked environments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Position-update SCA relies on unstated Hessian-bound constants δ0 and δk; without them the monotone-convergence and QoS-feasibility claims in Algorithm 2 are not verified, so the reported 5 dB gain is not reproducible.","rationale":"Good-faith reading: the FP transform is exact, the subproblems are convex, and the BCD structure would be monotone if the SCA surrogates are valid. The reader's far-field concern, however, does not seem to land in the simulated regime: A=2λ=0.2 m, BS-RIS distance is about 30 m, so the Fraunhofer distance is about 0.8 m and the plane-wave phase model is appropriate. The load-bearing risk is instead the unstated SCA constants in the antenna-position block. The position update is the only block that directly moves the antennas, and the claimed 5 dB MA gain comes from this block; if (31) and (33) are not global bounds, the monotonicity and feasibility arguments collapse. Since the paper gives no values for δ0, δk and no code, the simulation result cannot be checked from the text alone. The proposed test would settle whether the missing constants are easy to choose and whether the headline gain survives an independent implementation. This keeps the reader's CONDITIONAL verdict unchanged: the concern does not disprove the claim, but it does mean the central numerical result is not yet verified.","tokens_in":10480,"tokens_out":24927,"duration_ms":261781,"concrete_test":"Take the simulation setup of Section IV (A=2λ, N=8, M=32, L=4, Q=2, with the same angles and power levels). Numerically evaluate the Hessians ∇²Γ̂r(t̃) and ∇²f_k(t̃) over a fine grid of the moving region C_t, using W, Λ, V from the first BCD iteration, and set δ0 and δk to the largest eigenvalues plus a small margin. Re-run Algorithm 2 with those constants and verify three things: (i) the radar-SINR sequence is nondecreasing; (ii) every returned point satisfies |v_m|=1 and Γ_k ≥ γ_k for all users; (iii) the gap to the FPA baseline in Fig. 3 remains about 5 dB. If any check fails, the reported gain is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III-D approximates the radar objective and QoS constraints in the antenna-position block using quadratic surrogates (31) and (33). These surrogates are only valid global lower/upper bounds if δ0 I ⪰ ∇²Γ̂r(t̃) and δk I ⪰ ∇² f_k(t̃) hold on the entire moving region C_t. The paper specifies neither δ0 nor δk nor a computable procedure for them, stating only that the calculation is 'similar to [12]'. If either constant is too small, the SCA step (35) may not be a bound: the objective can decrease and constraint (35d) can be satisfied while the true QoS constraint (32) is violated. The claim that Algorithm 2 converges to a feasible point and yields ~5 dB radar-SINR gain over FPA therefore depends on an unverified numerical ingredient. This is more decisive than the far-field assumption, which is reasonable here since A=2λ=0.2 m and link distances are about 30 m, giving a Fraunhofer distance of roughly 0.8 m.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies an RIS-enhanced dual-functional radar-communication (DFRC) system in which the base station antennas are movable within local regions. It formulates a radar SINR maximization problem under per-user communication QoS constraints, minimum antenna separation constraints, transmit power, and unit-modulus RIS constraints, and proposes to solve it by block coordinate descent that alternates between the fractional-programming auxiliary variable, transmit beamformers, RIS phase coefficients, and transmit/receive antenna positions. The antenna-position block uses successive convex approximation with quadratic surrogates, and the RIS block uses a penalty method. Simulation results claim roughly 5 dB radar SINR gain over a fixed-position-antenna (FPA) baseline and monotone convergence within a few iterations.","tokens_in":10752,"tokens_out":14510,"duration_ms":144883,"significance":"If validated, the paper would provide a useful design framework for exploiting movable-antenna spatial degrees of freedom in RIS-aided DFRC systems, and the comparison against RPA, random-RIS, and GAS baselines is informative. The fractional-programming reformulation is algebraically standard, the block structure is clearly organized, and the simulation setup is mostly specified. However, the SCA surrogates for the antenna-position block rely on constants that are neither specified nor computable from the text, a conjugation error appears in the RIS-phase block, and the convergence/feasibility claims are not backed by constraint-violation or unit-modulus-violation metrics. These issues must be resolved before the reported radar SINR gain can be considered reproducible and attributable to the proposed algorithm.","major_comments":[{"comment":"The constants delta_0 and delta_k are never specified. For (35d) to imply the true QoS constraint (32), the surrogate (33) must be a global upper bound on f_k over the whole feasible region, which requires a verified Hessian bound; for (31) to be a global lower bound on the objective, the relevant condition is a bound on the spectral norm (or the lower end of the Hessian spectrum) of the Hessian, not merely delta_0 I ⪰ grad^2 Gamma_r as written. The statement that the calculation is 'similar to [12]' is not sufficient. If these constants are too small, the t-update can satisfy (35d) while violating the true communication constraint (32), and the monotone convergence reported in Fig. 2 is not guaranteed. Please report the actual constants or a backtracking procedure to determine them, and add a per-iteration feasibility check.","section":"Section III-D, Eqs. (31), (33), (35)"},{"comment":"There is a conjugation error in the transformation of the RIS-phase constraint. With v = [v_1, ..., v_M]^H and V = diag(v_m), for \\tilde h_{k,j} = diag(h_k^H) H(\\tilde t) w_j one obtains h_k^H V H(\\tilde t) w_j = v^T \\tilde h_{k,j}, so |h_k^H V H(\\tilde t) w_j|^2 = v^H \\tilde h_{k,j}^* \\tilde h_{k,j}^T v, not v^H \\tilde h_{k,j} \\tilde h_{k,j}^H v. Unless \\tilde h_{k,j} is redefined with a conjugate, Eqs. (22)-(25) and Algorithm 1 maximize a different function from the actual user SINR. Please correct the definition or the quadratic form and re-check the numerical results.","section":"Section III-C, Eq. (22)"},{"comment":"Convergence to a feasible point is not established. The RIS block is solved by a penalty method with only |v_m| <= 1 enforced at each inner step, and no projection or post-processing is described to recover exact unit-modulus phases; the final point may therefore not correspond to physically implementable RIS coefficients. In addition, BCD updates use the fractional-programming auxiliary variable fixed from the previous block, so the equivalence in (15) is only valid at the optimum; the monotone objective curve in Fig. 2 does not imply that all constraints in (14) are satisfied. Please report constraint-violation metrics, final unit-modulus error, and, if needed, a feasibility-restoration step.","section":"Algorithm 2 and Section IV"},{"comment":"The FPA and RPA baselines are described only by antenna geometry ('uniform planar arrays' and 'randomly distributed in the moving region'), without stating whether the same beamforming and RIS optimization is applied to them. If FPA simply uses a fixed UPA without the same W/V/RIS optimization, then the 5 dB gain in Fig. 3 conflates movable-antenna gains with gains from the optimization framework itself. Please state explicitly that all baselines use the same per-block optimization with only the position degrees of freedom removed, or provide an ablation isolating the MA contribution.","section":"Section IV, baseline descriptions"}],"minor_comments":[{"comment":"There are numerous typos, including 'INTR ODUTION' in the section header, 'assmue', 'expessed', 'refomulate', and 'interation' in the Fig. 2 caption; these should be corrected.","section":"Throughout"},{"comment":"The values of the SCA constants delta_0 and delta_k, the penalty parameters rho_1 and tau, the thresholds xi_1 and xi_2, and the initialization of all blocks are not reported; please include them for reproducibility.","section":"Section IV"},{"comment":"The far-field assumption is reasonable for the simulated geometry (A = 2 lambda = 0.2 m versus link distances around 30 m), but stating the Fraunhofer distance check explicitly would help readers assess when the geometric channel model is valid.","section":"Section II-A"},{"comment":"The inequality in (24) requires positive values of eta^(l) and z_k^(l); this positivity condition should be stated explicitly.","section":"Section III-C, Eq. (24)"}],"recommendation":"major_revision","confidential_remarks":"The paper has a plausible high-level framework and the simulations are suggestive, but the missing SCA constants, the conjugation error in the RIS block, and the unverified feasibility of the final point are load-bearing for the claimed 5 dB gain. These issues appear fixable within the manuscript's scope, so I recommend major revision rather than rejection. I would also ask the editor to ensure the authors clarify the FPA baseline so the headline gain is not an artifact of comparing against an unoptimized benchmark."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this one with a conditional acceptance hat on. The problem in (14)—jointly optimizing beamforming, receive filter, MA positions, and RIS coefficients under QoS and clutter—is new relative to [10] and [11], and the FP-based framework is a sensible assembly of known tools. That is the core value.\n\nWhat the paper does well: the derivations are self-contained, the FP transform is an exact equivalence, and each BCD subproblem is convex. The simulations compare against FPA, RPA, random RIS, and GAS, which is the right benchmark set. The ~5 dB radar SINR gain over FPA in Fig. 3 is plausible. The far-field assumption is not a problem here: A=2λ=0.2 m and link distances around 30 m put the Fraunhofer distance near 0.8 m, so the geometric channel model holds.\n\nThe soft spot is Section III-D. The SCA surrogates in (31) and (33) require constants δ0 and δk such that δ0I ⪰ ∇²Γ̂r(t̃) and δkI ⪰ ∇²f_k(t̃) over the entire moving region. The paper never gives these constants or a computable procedure to find them; it just says \"similar to [12]\". Without them, the SCA step (35) is not guaranteed to be a lower/upper bound. The objective could decrease, and constraint (35d) could hold while the true QoS constraint (32) is violated. That undermines the monotone-convergence and feasibility claims in Algorithm 2. The paper also does not prove convergence to a stationary point, and the penalty method only gives approximate unit modulus for the RIS. These are real gaps, but they are standard for this literature—the paper is incomplete, not wrong.\n\nOn citations: the reference list is appropriate, and [12] is the proper source for the gradient machinery. No red flags on self-citation.\n\nWho benefits: researchers working on MA-ISAC or RIS-DFRC optimization will find the formulation useful and the algorithm template handy. It is not a breakthrough, but it is a legitimate step in an active direction. I would not desk-reject it. Send it to review, and ask the authors to supply the missing constants, a convergence proof or at least a clear feasibility check after each BCD block, and error bars on the simulations. Without those, the 5 dB result stays a single-run observation.","headline":"Novel formulation, sound framework, but the omitted SCA constants and missing convergence proof make the 5 dB gain conditional rather than established.","tokens_in":11267,"tokens_out":3046,"would_cite":true,"duration_ms":28894,"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 optimizing movable-antenna positions, RIS reflection coefficients, beamforming, and receive filtering can raise radar SINR by roughly 5 dB over fixed-position arrays while holding communication QoS, according to the paper's…","keywords":["movable antenna","dual-functional radar-communication","reconfigurable intelligent surface","radar SINR maximization","joint transceiver design","block coordinate descent","fractional programming","successive convex approximation"],"falsifier":"Run the same joint design with a near-field (spherical-wavefront) channel model instead of the far-field response model; if the movable-antenna SINR gain over fixed positions shrinks toward zero or reverses as the moving-region size approaches the BS-RIS distance, the 5 dB claim rests on the far-field assumption.","tokens_in":10314,"feed_emoji":"📡","tokens_out":7550,"duration_ms":66764,"temperature":0.7,"pith_summary":"Movable antennas give a dual-functional radar-communication array extra spatial degrees of freedom, and this paper tries to show that those degrees of freedom can be spent on radar sensing without sacrificing communication. The design jointly optimizes transmit beamforming, the receive filter, antenna positions, and RIS reflection coefficients to maximize radar SINR under per-user QoS constraints. Because the problem is non-convex and coupled, the paper develops a block coordinate descent algorithm built on fractional programming, successive convex approximation, and a penalty method for unit-modulus RIS phases. In simulation the proposed design gains about 5 dB in radar SINR over a fixed-position antenna array and keeps that gain nearly constant as the communication QoS threshold rises. A sympathetic reading takes this as evidence that antenna-position optimization plus RIS tuning is a practical route to better integrated sensing and communication.","feed_headline":"Movable antennas add 5 dB radar gain in DFRC test","feed_subtitle":"Jointly tuning antenna positions, RIS phases, and beamforming lifts radar SINR while keeping user links on target.","key_machinery":"The argument runs on the field-response phase model for movable antennas, $g(t_n)$, whose entries are phases $e^{j\\frac{2\\pi}{\\lambda}\\rho(t_n,\\psi^e_j,\\psi^a_j)}$ depending on each antenna's in-region position. This phase model converts antenna-position variables into phases inside the radar steering matrices, letting the optimization move antenna locations. On top of that, the paper applies the fractional-programming transformation of the radar SINR into $\\hat{\\Gamma}_r(W,\\tilde{r},\\tilde{t},\\Lambda) = \\zeta_0^2 \\mathrm{tr}(2\\Re\\{W^H A_0(\\tilde{r},\\tilde{t})^H \\Lambda\\} - \\Lambda^H(\\Xi+\\sigma_r^2 I_N)\\Lambda)$, with the optimal auxiliary $\\Lambda^\\star$ in closed form; then SCA surrogate functions handle the non-convex QoS and minimum-separation constraints, and a penalty method enforces the RIS unit-modulus constraints.","core_discovery":"The central claim is that radar SINR in an RIS-enhanced DFRC system can be maximized by moving the transceiver antennas as part of the optimization, and that doing so pays off. Using a far-field geometric channel model in which antenna motion only changes link phases, the paper reformulates the radar SINR through fractional programming, then alternates updates of the auxiliary matrix, beamformers, RIS coefficients, and antenna positions. The numerical result singled out by the paper is a radar SINR improvement of about 5 dB over the fixed-position-antenna benchmark at the same transmit power, with convergence typically within six iterations and little sensitivity to the communication SINR threshold because the movable antennas supply extra spatial degrees of freedom.","pith_inferences":["The 5 dB figure is computed under far-field assumptions; a near-field or position-dependent-blockage model would likely shrink the gain, so an experimental benchmark with reconfigurable or stepper-motor antennas would be a more demanding test.","The framework suggests a dual objective in which positions are optimized to maximize communication rate under a radar SINR floor, which the paper does not run but the variable structure supports.","Quantized position grids (as in the greedy antenna selection baseline) capture only part of the gain; this implies that mechanical positioning accuracy on the order of a wavelength fraction may matter for realizing the reported advantage.","A robust version that accounts for angle uncertainty in target and clutter steering vectors would be a natural next step, since the current SINR expression assumes known angles."],"forward_implications":["Combining movable-antenna position optimization with RIS phase tuning yields roughly a 5 dB radar SINR improvement over a fixed-position array at equal transmit power.","The proposed BCD-FP-SCA-penalty procedure converges in about six iterations for the tested antenna counts, so the joint design is computationally plausible for online use.","Raising the communication SINR threshold barely lowers the achieved radar SINR, indicating that movable antennas absorb the QoS burden with their extra spatial degrees of freedom.","The same framework extends MA transceiver design to RIS-covered dead zones, where the RIS provides the only communication link between BS and users."],"supporting_citations":[{"why":"It supplies the far-field field-response phase model used for movable antennas.","marker":"[12]"},{"why":"It provides the geometric L-path channel model with constant angles and amplitudes for all links.","marker":"[13]"},{"why":"It introduces the fractional-programming transformation and the closed-form auxiliary-variable update.","marker":"[17]"},{"why":"It supplies the successive-convex-approximation surrogate technique used on non-convex constraints.","marker":"[18]"},{"why":"It gives the radar SINR expression for DFRC under QoS-aware precoding.","marker":"[14]"},{"why":"It gives the minimum variance distortionless response receive filter that the radar SINR derivation starts from.","marker":"[15]"},{"why":"It derives the simplified radar SINR form used after optimal filtering.","marker":"[16]"},{"why":"It provides the RIS lower-bound SINR feasibility reformulation that drives the penalty update.","marker":"[9]"}],"fun_headline_variants":["Movable antennas lift radar SINR by 5 dB in DFRC","Moving antennas add 5 dB to radar SINR in DFRC","RIS plus movable antennas: 5 dB radar SINR gain in DFRC","Antenna movement yields 5 dB radar SINR in DFRC","Moving antennas in RIS-DFRC: 5 dB radar SINR"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes the far-field condition: as antennas move, each path's angle and amplitude stay constant, so only the phase of each link changes.","fun_headline_variants_meta":{"raw":{"variants":["Movable antennas lift radar SINR by 5 dB in DFRC","Moving antennas add 5 dB to radar SINR in DFRC","RIS plus movable antennas: 5 dB radar SINR gain in DFRC","Antenna movement yields 5 dB radar SINR in DFRC","Moving antennas in RIS-DFRC: 5 dB radar SINR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000892,"raw_usage":{"total_tokens":3794,"prompt_tokens":842,"completion_tokens":2952,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":458,"completion_tokens_details":{"reasoning_tokens":2853}},"tokens_in":458,"tokens_out":2952,"duration_ms":19408,"temperature":1.0,"reasoning_tokens":2853,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T22:52:33.424609+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same joint design with a near-field (spherical-wavefront) channel model instead of the far-field response model; if the movable-antenna SINR gain over fixed positions shrinks toward zero or reverses as the moving-region size approaches the BS-RIS distance, the 5 dB claim rests on the far-field assumption.","supporting_citations":[{"cited_title":"MIMO capacity characterization for movable antenna systems,","cited_arxiv_id":null,"evidence_quote":"It supplies the far-field field-response phase model used for movable antennas."},{"cited_title":"Fractional programming for communication systems—part I: Power control and beamforming,","cited_arxiv_id":null,"evidence_quote":"It introduces the fractional-programming transformation and the closed-form auxiliary-variable update."},{"cited_title":"QoS-aware precoder optimization for radar sensing and multiuser communications under per-antenna power constraints,","cited_arxiv_id":null,"evidence_quote":"It gives the radar SINR expression for DFRC under QoS-aware precoding."},{"cited_title":"High-resolution frequency-wavenumber spectrum analysis,","cited_arxiv_id":null,"evidence_quote":"It gives the minimum variance distortionless response receive filter that the radar SINR derivation starts from."},{"cited_title":"Generalized transceiver beamforming for DFRC with MIMO radar and MU-MIMO communication,","cited_arxiv_id":null,"evidence_quote":"It derives the simplified radar SINR form used after optimal filtering."},{"cited_title":"An overview of signal processing techniques for RIS/IRS- aided wireless systems,","cited_arxiv_id":null,"evidence_quote":"It provides the RIS lower-bound SINR feasibility reformulation that drives the penalty update."}],"review_version":1}