{"id":"400dd012-bbcb-423c-8465-47d7977f4e53","arxiv_id":"2504.18843","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A DMA reception design is proposed that minimizes a Cramér-Rao bound over an area of interest while enforcing uplink SNR constraints for multiple users.","lead":"This paper shows how a Dynamic Metasurface Antenna (DMA) can be tuned to sense a whole region while still receiving data from multiple phones. The authors derive a mathematical bound on location accuracy and use it to design beam weights.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The advertised QoS guarantee depends on an unverified SDR rank-one/projection step: if q_i^opt violates the Lorentzian codebook or SNR constraints, the central dual-functional claim is unsupported.","rationale":"I focused on the promise of guaranteed multi-user uplink communication performance. For that promise to hold, the SDR output must be convertible to a feasible Lorentzian DMA configuration that satisfies the SNR thresholds. This is the least secure step in the argument: the only support is a general citation to [22], and no numerical feasibility audit appears in Section IV. The reader's weakest_assumption identifies this same point. I also noted that the FIM derivation conditions on data symbols and reflection coefficients, and that the SNR constraint ignores multiuser interference; these are real modeling caveats, but they affect how one interprets the sensing CRB and the exact QoS metric rather than the immediate feasibility of the optimized weights. Thus the decisive check is to audit the SDR projection. If the audit passes, the paper's conditional verdict should stand or be upgraded; if it fails, the claimed guarantee is unsupported. Because the reader already made the verdict conditional on exactly this check, I recommend no change to the verdict.","tokens_in":8733,"tokens_out":10021,"duration_ms":115863,"concrete_test":"Re-run every P1/P2 CVX instance used for Fig. 1 (all Pmax and γu values, 500 Monte Carlo trials), record the eigenvalues of each optimized Q_i, and form q_i^opt = exp(j∠[u_i]_{1:NE}). Then compute the original per-user received SNR Γ_u = (1/σ²) Σ_i Tr{H_u,i q_i^opt (q_i^opt)^H} and verify each w_i,n^RX = 0.5(j + e^{j∠[q_i^opt]_n}) lies in W with phase in [-π/2, π/2]. Report the fraction of rank-one solutions, the maximum relative SNR shortfall max_u max(0, γ_u - Γ_u)/γ_u, and the fraction of codebook violations. If the shortfall or violation fraction is nonzero, the projected solutions are not feasible and the communication guarantee in the abstract is not delivered.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III-B relaxes each rank-one Q_i in P1/P2 to PSD and then recovers q_i^opt = exp(j∠[u_i]_{1:NE}) from the principal singular vector u_i. The paper justifies rank-one recovery by citing [22] ('as established in [22], the SDR-relaxed problem yields solutions that satisfy the necessary rank-one constraints'), but [22]'s tightness conditions are not shown to hold for this problem, which couples NRF PSD matrices through U SNR constraints and 3|A| Schur-complement LMIs. The projection step is a heuristic: even if Q_i is rank-one, phase-only extraction can move q_i outside the Lorentzian codebook W in (1) and can reduce the left side of Tr{H_u,i Q_i} below γ_u. Section IV reports PEB/RMSE curves but never verifies the post-projection SNR constraints or codebook membership for any Monte Carlo trial. Since the abstract and conclusion promise guaranteed multi-user uplink communication performance, the claim is currently supported only by an unexamined assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies receive beamforming design for a dynamic metasurface antenna (DMA) that simultaneously supports multi-user uplink communications and wide-area radar sensing. The authors derive a Cramér-Rao Bound (CRB) and the corresponding Position Error Bound (PEB) for estimating the positions of active users and passive targets in the near field, and then use the PEB as the design objective for the DMA's Lorentzian-constrained analog weights under per-user SNR constraints. The nonconvex design problem is relaxed via semidefinite relaxation (SDR) into problems P1 and P2, and a closed-form solution is also proposed. Numerical results over an area of interest indicate that the proposed designs outperform the CRB-minimizing DMA benchmark of [12]. The paper also includes a complexity analysis of the proposed algorithms.","tokens_in":8954,"tokens_out":12781,"duration_ms":137407,"significance":"The paper addresses a timely problem—ISAC with extremely large DMA apertures—and contains several useful ingredients: a closed-form CRB tailored to the partially-connected DMA structure, a convex SDR formulation that exploits the block structure of the codebook, and a low-complexity alternative. If the technical issues below are fixed, the area-wide PEB design can be a meaningful contribution to DMA-based ISAC. The stated guarantees, however, currently depend on assumptions that are not proven or tested: the tightness of the SDR and the treatment of unknown data symbols and reflection coefficients in the CRB. The simulation study is honest in comparing against a benchmark with full target knowledge, but it does not currently validate the QoS claims after the projection step.","major_comments":[{"comment":"The SDR is not closed under the actual variable Q_i=[q_i;1][q_i;1]^H. As written, the relaxed constraints only impose Q_i⪰0 and do not include the unit-modulus constraints on q_i (equivalently the diagonal constraints on Q_i) that follow from q_{i,n}=e^{jφ_{i,n}} in the Lorentzian codebook (1). With only PSD constraints, the problems are unbounded below: scaling each Q_i by a common factor t scales the FIM I_η by t and the SNR left-hand sides by t, so PEB tends to 0 while the SNR constraints remain feasible. A normalization constraint (e.g., diag(Q_i)=1) must be included and the optimization problem restated.","section":"Section III-B, P1 and P2"},{"comment":"The recovery q_i^opt = exp(j∠[u_i]_{1:N_E}) from the principal singular vector u_i of the SDR solution is asserted to be exact by citing [22], but the general rank-one guarantee of [22] is not verified for this problem, which couples N_RF blocks through U SNR constraints and 3|A| Schur-complement LMIs. The phase-only projection can leave the Lorentzian codebook W in (1) and can reduce the per-user SNR below the threshold. Section IV reports PEB and RMSE but not the post-projection SNR or codebook membership; thus the 'guaranteed' multi-user QoS conclusion is currently unsupported. Please provide either a correctness argument, numerical verification of the constraints after projection, or a qualification of the claim.","section":"Section III-B, recovery of q_i^opt"},{"comment":"The FIM is computed from the conditional mean µ = vec{W^H P^H ∑_u h_u s_u}, which treats the uplink data symbols s_u and the reflection coefficients β_k (via h_R,u) as known. In the actual uplink ISAC scenario these are unknown nuisance parameters; ignoring them yields a PEB that is a lower bound for an estimator with additional side information. The manuscript neither includes these nuisance parameters in the FIM nor states this caveat. The sensing performance claims in Section IV should be re-evaluated or explicitly restricted to the pilot-assisted/data-aided case.","section":"Section III-A, Eqs. (4)–(7)"},{"comment":"The objective PEB(WRX;η) is not well defined as a straightforward substitution of the 3|A| grid coordinates into the 3K-dimensional FIM of Section III-A. The received signal in (4) depends only on the K actual targets/UEs; for a grid point at which no target is present, the derivatives ∂h_u/∂[η]_i vanish, making the corresponding rows and columns of I_η zero and the 'PEB' singular. If a virtual target is postulated at every grid point, the signal model in (4) and h_R,u must be modified accordingly, and the evaluation of the PEB in the simulations must be aligned with that model. Please clarify the definition of the area-wide PEB and how it is used in the simulations.","section":"Section III-B, P1"}],"minor_comments":[{"comment":"There is a typo in 'Quality-of-Sevice' in the third paragraph; it should be 'Quality-of-Service'.","section":"Introduction"},{"comment":"The word 'Relexation' in the first paragraph of Section III-B should be 'Relaxation'.","section":"Section III-B"},{"comment":"The sentence 'where Bi ∈ C^{(N_E+1)×(N_E+1)} has a similar structure to Hu,i, but is defined with respect to the the matrix B' contains the duplicated word 'the the'.","section":"Section III-C"},{"comment":"The notation 'Iη ea' in the LMI constraint is hard to read; it should be written as [I_η e_a; e_a^T b_a] ⪰ 0 with clear subscripts.","section":"Section III-B, P1"},{"comment":"Reference [23] is a paper on the physics of the near-field and does not appear to be the source of the harmonic-geometric mean inequality used in Section III-C; please replace it with a standard matrix inequality reference.","section":"References"},{"comment":"It would help the reader if the figures reported the feasibility rate of the projected solutions (e.g., the fraction of Monte Carlo trials in which the projected q_i^opt satisfies both the codebook and the SNR constraints), since this directly bears on the QoS claim.","section":"Section IV"}],"recommendation":"major_revision","confidential_remarks":"The manuscript relies heavily on the authors' own prior work [10] and [12] for the channel model and the benchmark; this is acceptable but the novelty relative to [12] should be made sharper in the revised version. I have no concerns about the integrity of the numerical experiments, but the unresolved SDR normalization and the unstated conditioning assumptions in the CRB need to be fixed before the paper can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a legitimate extension of the authors' own DMA localization work. The area-wide PEB objective over an AoI grid, with per-user SNR constraints and a low-complexity closed-form alternative, is the new piece. The model and CRB derivation are standard but carefully laid out, and the simulations are plausible: P1 beats the benchmark [12] even when the benchmark knows all target positions, which makes the case that area-wide sensing is not just a gimmick.\n\nWhat I trust: the partially-connected DMA structure is exploited to turn the Lorentzian phase constraint into a rank-one PSD form; the trace expressions are consistent; and the CFS from the principal singular vectors of the B_i matrices gives a sensible fallback with much lower complexity. The benchmark choice is fair—it gives the reference method an advantage and the proposed method still wins. The RMSE-to-PEB convergence in Fig. 1a is the kind of sanity check that makes the simulation credible.\n\nThe weak spot is exactly the one flagged in the stress test. Section III-B recovers q_i^opt = exp(j∠[u_i]_{1:NE}) from the SDR solution and claims via [22] that the relaxation is rank-one tight. No argument maps the general tightness conditions in [22] to this problem, which couples NRF matrices through U SNR constraints and Schur-complement LMIs. The projection is also a heuristic: even if each Q_i is rank-one, phase-only extraction need not stay in the Lorentzian codebook (1), and it can reduce the SNR term. The paper never checks, after projection, that Γ_u ≥ γ_u holds over any Monte Carlo trial. Since the abstract and conclusion promise 'guaranteed' communication performance, that promise currently rests on an unexamined assumption. The FIM is a second, softer issue: ξ contains only positions, not the passive-target reflection coefficients β_k, so the CRB is optimistic if those are truly unknown.\n\nBottom line: conditional. The central idea is sound and the simulation trend is believable; the missing feasibility check for the projection is removable by adding a constraint-recovery step plus a Monte Carlo histogram of post-projection SNR. I'd send it to peer review, and ask the authors to verify or repair the rank-one/feasibility claim or soften the guarantee language. A reader working on metasurface ISAC will get value; I'd likely cite the area-wide PEB formulation.","headline":"Solid area-wide PEB extension of the authors' DMA work; the SDR rank-one/projection step needs verification before the QoS guarantee is credible.","tokens_in":9460,"tokens_out":2433,"would_cite":true,"duration_ms":26156,"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":"A single DMA receiver can be tuned to sense a whole region while keeping uplink users connected, and it beats localization-first designs in simulation.","keywords":["dynamic metasurface antennas","integrated sensing and communication","Cramér-Rao bound","position error bound","near-field localization","analog beamforming","semidefinite relaxation","multi-user uplink"],"falsifier":"Run the paper's setup ($N_{\\mathrm{RF}}=8$ microstrips, $N_E=64$ elements, $\\gamma_u=30$ dB, $|\\mathcal{A}|=8$ points) and, after solving P1 or P2, compute the actual per-user SNR $\\Gamma_u$ using the projected weights $\\mathbf{q}_i^{\\mathrm{opt}}=\\exp(j\\angle[\\mathbf{u}_i]_{1:N_E})$ rather than the relaxed matrices. If any $\\Gamma_u$ falls below $\\gamma_u$ or the simulated RMSE departs from the SDR's predicted PEB, the advertised sensing/communication trade-off is not delivered.","tokens_in":8517,"feed_emoji":"📡","tokens_out":7085,"duration_ms":69246,"temperature":0.7,"pith_summary":"The paper tries to show that a Dynamic Metasurface Antenna used as a receiver can be set to do two jobs at once: estimate the positions of radar targets anywhere inside a defined Area of Interest, and keep a set of uplink users connected. Its route is to derive the Cramér–Rao Bound and the Position Error Bound for DMA reception, then minimize the area-wide PEB over a grid of points subject to per-user signal-to-noise-ratio constraints. Because the DMA's tunable elements are arranged in microstrips (a partially-connected architecture), the otherwise nonconvex weight-design problem is recast as convex semidefinite programs. The simulations claim all three proposed designs—direct PEB minimization, a lower-bound approximation, and a closed-form solution—outperform a DMA-based CRB-minimizing receiver that is even given the exact target positions.","feed_headline":"One metasurface receiver senses a whole area while serving uplink users","feed_subtitle":"By optimizing for the whole region, not known target spots, the receiver beats localization-first designs that know every target's location.","key_machinery":"The load-bearing object is the partially-connected DMA reception model: $N_{\\mathrm{RF}}$ microstrips, each with $N_E$ Lorentzian-constrained elements whose analog weights belong to $\\mathcal{W}=\\{0.5(j+e^{j\\varphi}):\\varphi\\in[-\\pi/2,\\pi/2]\\}$, plus the near-field channel and steering-vector model that turns element responses into a Fisher Information Matrix. The PEB, defined as the trace of the inverse FIM, is the objective; semidefinite relaxation with rank-one matrices $\\mathbf{Q}_i$ and the projection $\\mathbf{q}_i^{\\mathrm{opt}}=\\exp(j\\angle[\\mathbf{u}_i]_{1:N_E})$ converts the nonconvex codebook design into convex problems P1 and P2. The lower-bound trick $\\operatorname{Tr}\\{\\mathbf{I}^{-1}\\}\\ge(\\operatorname{Tr}\\{\\mathbf{I}\\})^{-1}$, drawn from the harmonic-geometric mean inequality, is what produces the cheaper P2 and the closed-form solution.","core_discovery":"In the paper's own terms, the central claim is that minimizing the Position Error Bound across a discretized Area of Interest, rather than minimizing the Cramér–Rao Bound for known target coordinates, is the right objective for a dual-function DMA receiver. The paper derives the Fisher Information Matrix for the DMA-received signal, expresses the PEB as $\\mathrm{PEB}(\\mathbf{W}_{\\mathrm{RX}};\\xi)=\\sqrt{\\operatorname{Tr}\\{\\mathbf{I}_\\xi^{-1}\\}}$ for both active users and passive targets in the AoI, and optimizes the Lorentzian phase-constrained analog weights under per-user SNR constraints $\\Gamma_u\\ge\\gamma_u$. The reported simulations show the localization RMSE converging to the corresponding PEB and all proposed solutions outperforming the benchmark DMA-based CRB-minimizing design, even when that benchmark knows all target positions.","pith_inferences":["An ablation that runs the same area-wide objective without the SNR constraints would separate the gain due to area-wide sensing from the gain due to the QoS-aware formulation, since the benchmark comparison is the paper's strongest test.","Because the rank-one tightness of the semidefinite relaxation is imported from a general theory result rather than verified for this specific multi-constraint problem, a direct numerical check of each user's SNR after the Lorentzian projection is a natural prerequisite for deployment.","The same PEB-over-AoI formulation could be extended to moving targets by replacing the static grid $\\mathcal{A}$ with a trajectory-weighted set, or to wideband operation by treating each subcarrier's FIM independently."],"forward_implications":["If the area-wide PEB result holds, a DMA receiver can monitor a region without any prior target locations, whereas localization-first designs appear to need that prior knowledge to stay competitive.","The convex formulations P1 and P2 mean the beamforming weights can be computed by standard convex programming solvers, making the approach a practical alternative to existing DMA designs.","The lower-bound solution P2 and the closed-form solution offer sensing accuracy comparable to direct PEB minimization at lower computational cost, so a system designer can trade complexity against accuracy.","As the required communication SNR grows, the optimized sensing accuracy degrades at a similar rate for P1 and P2, while the closed-form solution holds its PEB roughly constant, showing that the QoS constraint is what consumes sensing performance."],"supporting_citations":[{"why":"The DMA-based CRB-minimizing RX design used as the benchmark that all proposed solutions must beat.","marker":"[12]"},{"why":"Supplies the near-field channel and steering-vector model, the DMA propagation matrix, and the modified MUSIC estimator used in the simulations.","marker":"[10]"},{"why":"Provides the Fisher Information Matrix and CRB estimation machinery on which the PEB objective is built.","marker":"[18]"},{"why":"Supplies the Schur complement reformulation used to make the P1 design problem convex.","marker":"[19]"},{"why":"Shows the Lorentzian weight transformation that lets the codebook constraint be folded into the semidefinite relaxation as the matrices $\\mathbf{Q}_i$.","marker":"[20]"},{"why":"The semidefinite relaxation theory invoked to justify the claim that the relaxed problem yields solutions satisfying the necessary rank-one constraints.","marker":"[22]"},{"why":"Cited for the harmonic-geometric mean inequality that produces the lower-bound objective behind the low-complexity P2 formulation.","marker":"[23]"}],"fun_headline_variants":["One DMA receiver senses the whole region while serving uplink users","Area-wide sensing goal beats target-specific designs in DMA receivers","DMA antenna optimizes for whole area, not known targets, for dual function","Sensing the entire region: DMA receiver outperforms known-target designs","Metasurface receiver does area-wide sensing and uplink in one aperture"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The design's promised guarantee depends on the assumption that the relaxed optimization, after projecting its solution back onto the allowed antenna phases, still gives every user the required signal strength.","fun_headline_variants_meta":{"raw":{"variants":["One DMA receiver senses the whole region while serving uplink users","Area-wide sensing goal beats target-specific designs in DMA receivers","DMA antenna optimizes for whole area, not known targets, for dual function","Sensing the entire region: DMA receiver outperforms known-target designs","Metasurface receiver does area-wide sensing and uplink in one aperture"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000608,"raw_usage":{"total_tokens":2817,"prompt_tokens":917,"completion_tokens":1900,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":533,"completion_tokens_details":{"reasoning_tokens":1808}},"tokens_in":533,"tokens_out":1900,"duration_ms":13790,"temperature":1.0,"reasoning_tokens":1808,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:08:41.359387+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the paper's setup ($N_{\\mathrm{RF}}=8$ microstrips, $N_E=64$ elements, $\\gamma_u=30$ dB, $|\\mathcal{A}|=8$ points) and, after solving P1 or P2, compute the actual per-user SNR $\\Gamma_u$ using the projected weights $\\mathbf{q}_i^{\\mathrm{opt}}=\\exp(j\\angle[\\mathbf{u}_i]_{1:N_E})$ rather than the relaxed matrices. If any $\\Gamma_u$ falls below $\\gamma_u$ or the simulated RMSE departs from the SDR's predicted PEB, the advertised sensing/communication trade-off is not delivered.","supporting_citations":[{"cited_title":"Near-field localization with dynamic metasurface antennas at THz: A CRB minimizing approach,","cited_arxiv_id":null,"evidence_quote":"The DMA-based CRB-minimizing RX design used as the benchmark that all proposed solutions must beat."},{"cited_title":"Optimal spatial signal design for mmWave positioning under imperfect synchronization,","cited_arxiv_id":null,"evidence_quote":"Supplies the Schur complement reformulation used to make the P1 design problem convex."},{"cited_title":"Metasurface-based receivers with 1-bit ADCs for multi-user uplink communications,","cited_arxiv_id":null,"evidence_quote":"Shows the Lorentzian weight transformation that lets the codebook constraint be folded into the semidefinite relaxation as the matrices $\\mathbf{Q}_i$."},{"cited_title":"Semidefinite relaxation of quadratic optimization problems,","cited_arxiv_id":null,"evidence_quote":"The semidefinite relaxation theory invoked to justify the claim that the relaxed problem yields solutions satisfying the necessary rank-one constraints."},{"cited_title":"The physics of the near-field,","cited_arxiv_id":null,"evidence_quote":"Cited for the harmonic-geometric mean inequality that produces the lower-bound objective behind the low-complexity P2 formulation."}],"review_version":1}