{"id":"5cc830e4-8cd9-4882-936a-1ca49f22f1de","arxiv_id":"2507.02731","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"A theoretical framework shows that RIS-assisted ISAC cellular networks can, in simulation, detect millimeter-level structural deformations by subtracting time-separated measurements.","lead":"This paper proposes using cellular communication signals, reflected off a tunable surface mounted on a building, to detect tiny structural deformations. It derives theoretical accuracy bounds and simulation-based detection rates for millimeter-level monitoring without dedicated sensors.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The differential-measurement premise in Eq. (6), that background scatterers are identical across adjacent RIS phase configurations, is the most load-bearing assumption; without a bound on the residual clutter from time-varying scatterers, the millimeter-level PEB claim is not robust.","rationale":"I reviewed the abstract's central claim against the derivation. The clean differential model (10) is the foundation: it is obtained solely by subtracting adjacent measurements under Eq. (6). All later FIM formulas, the Wald test, and the detection-probability curves inherit this assumption. Among the issues raised by the reader, the static-scatterer assumption is the most load-bearing because it is a modeling premise about the physical environment rather than a fixable algebraic slip. The beta = 4M^4 versus 4M^2 discrepancy is real, but because the numerics set SNR directly, it mainly affects the physical interpretation of the SNR scaling rather than the fixed-SNR PEB curves. The diagonalization of Q_R in Eq. (61) is a genuine mathematical gap in Theorem 5, but it concerns the secondary claim about receivers with self-positioning errors, not the core millimeter-level detection claim. The paper deserves credit for the NOMP validation and the closed-form EFIM expressions, which are internally consistent under the stated assumptions. My concern does not change the verdict: the reader's CONDITIONAL status is appropriate, with the condition being a demonstration that Eq. (6) holds or that residual clutter is negligible at the millimeter level.","tokens_in":22560,"tokens_out":6083,"duration_ms":64159,"concrete_test":"Simulate the Section V-C geometry with the paper's Omega-/Omega+ phase alternation but add one time-varying point scatterer, e.g., moving at 1 m/s with RCS 20 dB below the RIS reflection, so that H_s[i] differs from H_s[i-1]. Run the NOMP estimator on si,i-1 and compare the RIS position estimate to the truth. If the resulting bias exceeds the 1 mm deformation threshold, Eq. (6) is not a safe premise for the central claim. As an analytical companion, repeat the EFIM derivation with an additive residual term delta H_s of covariance sigma_c^2 I and report the sigma_c value at which the PEB crosses 1 mm.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that RIS-aided ISAC achieves millimeter-level deformation detection rests on Eq. (6), H_s[i]=H_s[i-1], which allows adjacent measurements to cancel all background multipath. This premise turns the received signal into the cleaned form (10), and every FIM expression and PEB in Sections III-V follows from (10). In a real SHM deployment, the two observation instants are separated by at least one RIS phase switch; within that interval, vehicles, pedestrians, foliage, or thermal/turbulence-induced phase changes can alter the background channel. The paper neither quantifies the magnitude of the residual delta H_s = H_s[i] - H_s[i-1] nor models it in the FIM. At the claimed millimeter accuracy, even a weak residual scatterer introduces a bias comparable to the deformation signal. The abstract's unconditional statement that 'theoretical and numerical analyses confirm ISAC's capability for millimeter-level deformation detection' is therefore only valid under an unverified stationarity assumption. The paper does state that the assumption is 'reasonable' for closely spaced instants, but the numerical validation in Section V-C uses synthetic data generated with exactly the same static-background model, so it cannot detect a violation of (6). A robustness analysis or an explicit bound on permissible background dynamics is required before the central claim can be accepted.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes an RIS-aided cooperative ISAC framework for structural health monitoring. The key idea is to mount an RIS on a possibly deformable structure, alternate its phase configuration between two closely spaced observation instants, and subtract the received signals so that the static background multipath channel cancels. The remaining differential signal is then used to estimate the RIS position via a Fisher information analysis. The authors derive closed-form FIM and EFIM expressions, extend them to temporal accumulation, multiple cooperative receivers, and receivers with self-positioning errors, and build a Bayesian/Wald detection model for structural state identification. Numerical results validate the CRLB through the NOMP algorithm and study the PEB and detection probability as functions of bandwidth, array size, RIS size, observation time, and cooperative node geometry.","tokens_in":22762,"tokens_out":9858,"duration_ms":117161,"significance":"If the result holds, the paper is a useful contribution: it translates SHM accuracy requirements into concrete network design parameters and provides analytic expressions for the PEB of an RIS reference point as a function of bandwidth, antenna count, RIS size, observation time, node placement, and receiver position uncertainty. The ideal-case EFIM derivation follows standard CRLB machinery, and the NOMP simulation in Section V-C supports the single-configuration CRLB expressions. The paper also makes falsifiable predictions (e.g., PEB decreases with T, K, M, and B) and compares concrete deployment strategies. However, the central claim of millimeter-level deformation detection is not yet robust: the RIS gain expression appears to contain a factor-M^2 algebraic error, the static-background assumption that enables the whole differential measurement is not bounded or stress-tested, and the detection model has dimensional inconsistencies. These issues are load-bearing for the central claim, but they appear correctable within the manuscript's scope.","major_comments":[{"comment":"The text states that with the optimized RIS phases, beta_{**} = 4M^4. Direct evaluation of Eq. (19a) with unit-modulus steering vectors and |Omega[i]-Omega[i-1]|_mm = 2 gives beta_{**} = 4M^2, because each scalar factor in (19a) is a sum of M terms of magnitude at most 2, so the product has magnitude at most (2M)^2 = 4M^2. The asserted value 4M^4 overestimates the RIS contribution by a factor of M^2. Since the per-antenna SNR in Eq. (20) and all subsequent FIM entries, including (24), (26), (28), (30), and (32), inherit this factor, the reported PEB values in Figs. 5-9 and the detection probabilities in Figs. 11-12 are optimistically biased. This algebraic error must be corrected and the simulations rerun before the millimeter-level claim can be accepted.","section":"Section III-A, Eqs. (19a)-(21)"},{"comment":"The central cancellation H_s[i] = H_s[i-1] is assumed but never bounded. In a real deployment, the two observation instants are separated by at least one RIS phase switch, and vehicles, pedestrians, foliage, wind, or thermal drift can alter the background scatterer channel. The residual Delta H_s = H_s[i] - H_s[i-1] enters the differential signal in Eq. (10) as unmodeled clutter, and at the claimed millimeter accuracy even a weak residual scatterer can produce a bias comparable to the deformation signal. The numerical validation in Section V-C uses synthetic data generated under exactly the same static-background model, so it cannot detect a violation of (6). The authors should either model Delta H_s in the FIM or provide an explicit condition (e.g., a bound on residual scatterer power or a coherence-time requirement) under which the millimeter-level PEB claim remains valid.","section":"Section II-A, Eq. (6)"},{"comment":"The step Q_R^{-1} = U tilde-Q_R^{-1} U^H assumes that the receiver position-error covariance is diagonal in the target-centric U basis. For the diagonal Cartesian covariance stated in Eq. (59), this is generally false: the columns of U are geometry-dependent direction vectors (u_tau, u_phi_az, u_phi_el) and are not eigenvectors of an arbitrary diagonal Q_R. Consequently, Theorem 5 and Corollary 6 do not hold for arbitrary diagonal Q_R. The authors should either derive the EFIM for a general Q_R, or state and justify an additional assumption such as isotropic error (Q_R = qI), or show that the diagonalization in the U basis is a conservative bound.","section":"Section IV-C, Eq. (61)"},{"comment":"The detection model is dimensionally inconsistent. The statistic d(hat-r) = (hat-r - r0)^T F(hat-r)(hat-r - r0) is dimensionless because F has units m^{-2}, yet the threshold kappa is stated in millimeters and the assumed variances sigma_p^2 have units m^2. Under H0, the exact distribution of d is a chi-squared distribution with three degrees of freedom (for a known FIM), not a zero-mean Gaussian. This undermines the quantitative detection-probability results in Figs. 11-12. The authors should define a physical displacement statistic with proper units, or use the correct noncentral chi-squared / quadratic-form distribution for the Wald test, and then recompute the detection probabilities.","section":"Section III-C, Eqs. (42)-(44)"},{"comment":"Corollary 4 relies on the EFIM being a sum of positive semidefinite rank-one terms for each receiver. This is valid if each receiver's FIM is expressed in the same global coordinate basis, which the paper does implicitly. However, the proof in Eq. (55) uses an EFIM decomposition in which the U_k matrices differ per receiver; the inequality F[1:K+1] >= F[1:K] follows as stated, but the authors should clarify that the U_k columns are expressed in a common coordinate frame to avoid an apparent basis mismatch.","section":"Section IV-B, Corollary 4"}],"minor_comments":[{"comment":"The beamforming vector is written as f = a_H^T(...) in the text; it should be f = a_T(theta_az, theta_el), a column vector, with the Hermitian transpose used only in the alignment argument.","section":"Section III-A, Eq. (21)"},{"comment":"The notation beta_{**} is undefined; the subscripts should be tied to the parameter pair (e.g., beta_{tau_r tau_r}, beta_{phi_az phi_az}) for readability.","section":"Section III-A, Eq. (19a)"},{"comment":"There is a typo in 'Acoording' in the proof of the non-negativity result; the sentence should read 'According'.","section":"Appendix C, Eq. (86)"},{"comment":"The label 'CRLBApprox.' is missing a space and could be typeset as 'CRLB Approx.'; also, the figure caption should state explicitly which curves are which, since the legend is small.","section":"Section V-C, Fig. 10"},{"comment":"The channel model in Eq. (2) includes the RIS phase matrix Omega[i] as an M x M diagonal matrix premultiplying the RIS steering vectors; the dimensions of the product are not immediately obvious, and a short dimension-check or block diagram would help the reader verify the M-scaling of the final expressions.","section":"Section II-A, Eq. (2)"}],"recommendation":"major_revision","confidential_remarks":"The M^4 versus M^2 discrepancy in the RIS gain is a concrete algebraic error that directly inflates all PEB and detection-probability numbers, and the Q_R diagonalization in Theorem 5 is not valid for arbitrary diagonal covariance matrices. Both are fixable, but the simulations need to be rerun after the corrections. The static-background assumption in Eq. (6) is the other main risk: the paper should not claim unconditional millimeter-level detection without a robustness analysis or an explicit bound on residual background dynamics. I would like the editor to ask for a revised version that addresses these three points explicitly, plus the dimensional inconsistency in the detection model."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Stef, quick take on arXiv:2507.02731. The core idea is fresh: alternate RIS phase configurations and subtract adjacent snapshots to cancel static multipath, turning the RIS into a coherent reference point for structural deformation. I have not seen that specific mechanism applied to cellular ISAC/SHM, and the EFIM treatment of 3D multi-receiver cooperation, including receivers with self-positioning errors, is a reasonable extension of the Shen–Win machinery. The NOMP validation in Fig. 10 is a real positive: the estimators track the closed-form CRLB, so the ideal-case FIM math is doing something right.\n\nNow the soft spots. The reader is right about β. Direct evaluation of (19a) with the stated unnormalized steering vectors gives β = 4M^2, not 4M^4 as printed after (21). That is a factor-M^2 inflation of the SNR, which for a 64x64 RIS is tens of dB. All the numerical headlines that depend on absolute PEB need to be rerun with a consistent expression. This is fixable, but it is not cosmetic.\n\nThe second real problem is Theorem 5. Eq. (61) diagonalizes Q_R^{-1} in the U basis, but a covariance that is diagonal in Cartesian coordinates is not generally diagonal in that target-centric basis. So the clean per-direction compression in (62) only holds under a special assumption that is not stated. The multi-receiver self-positioning results in Fig. 9 inherit that issue.\n\nThe stress-test concern about Eq. (6) is real but I would not call it fatal. The static-background assumption is stated explicitly and is the point of the differential scheme; the issue is that the paper never bounds the residual from background changes and the simulations use the same static model, so the robustness question is simply not addressed. At millimeter accuracy that matters. The abstract should say “under quiescent background conditions,” not unconditionally.\n\nThe Gaussian likelihood for detection probability is a modeling choice without external validation; minor compared with the above.\n\nOverall: the ideal-case framework is coherent, the math is mostly standard, and the new mechanism deserves attention. It just has two concrete technical errors that need correction before the numbers are trustworthy. Send it to review, but expect a major-revision round.\n\nRecommendation: engage with it in review; I would not desk-reject.","headline":"Fresh RIS differential-sensing idea for SHM, but two concrete errors need fixing before the millimeter-level claims are trustworthy.","tokens_in":23347,"tokens_out":5770,"would_cite":true,"duration_ms":63123,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"RIS-aided cellular ISAC can detect millimeter-level building deformation.","keywords":["integrated sensing and communication","structural health monitoring","reconfigurable intelligent surfaces","Fisher information matrix","position error bound","cooperative localization","deformation detection","information ellipsoid"],"falsifier":"Take two adjacent measurements with the RIS alternating phases while a scatterer such as a passing car or a pedestrian changes the background channel between them, and check whether the estimated RIS position still tracks a known millimeter-level displacement applied to the structure; the model predicts no background residual, so any observed bias or PEB increase beyond the predicted bound would falsify the static-scatterer assumption. A second check is to simulate a receiver position-error covariance that is diagonal in Cartesian coordinates but not diagonal in the target basis $\\mathbf{U}$, which would break the assumptions behind Theorem 5.","tokens_in":1875,"feed_emoji":"📡","tokens_out":1758,"duration_ms":63141,"temperature":0.7,"pith_summary":"This paper tries to establish that an ordinary cellular ISAC network, with a reconfigurable intelligent surface mounted on a structure as a reference point, can measure millimeter-level deformation by subtracting measurements taken under two alternating RIS phase configurations. The subtraction cancels static background multipath while the RIS path changes, turning the RIS phase modulation into a radio marker for the structure's position. The authors derive Fisher information bounds for 3D RIS localization and show that observation time, cooperative receivers, bandwidth, and RIS size reduce the position error bound enough to meet structural health monitoring limits. A Bayesian hypothesis test then maps the estimated displacement into an unchanged versus damaged structural state. If correct, this would let dense cellular infrastructure double as quantitative deformation sensors without new dedicated hardware.","feed_headline":"RIS-aided cellular ISAC can detect millimeter-level building deformation","feed_subtitle":"RIS reference points cancel multipath clutter, letting 3D networks measure tiny structural deformations.","key_machinery":"The carrying object is the information ellipsoid, an equivalent Fisher information representation $\\mathbf{F}_e(\\mathbf{r})=\\lambda_{\\tau_r}\\mathbf{u}_{\\tau_r}\\mathbf{u}_{\\tau_r}^H+(\\lambda_{\\phi_r^{az}}-\\chi_{\\phi_r^{az}\\phi_r^{el}})\\mathbf{u}_{\\phi_r^{az}}\\mathbf{u}_{\\phi_r^{az}}^H+(\\lambda_{\\phi_r^{el}}-\\chi_{\\phi_r^{el}\\phi_r^{az}})\\mathbf{u}_{\\phi_r^{el}}\\mathbf{u}_{\\phi_r^{el}}^H$ built from three mutually orthogonal unit vectors for range, azimuth, and elevation. The range intensity grows with bandwidth squared, while the angle intensities grow with effective array aperture and decay with target distance. The companion mechanism is the two-state RIS phase alternation with phases $-\\left(\\bar\\phi_m-\\bar\\theta_m\\right)$ and $\\pi-\\left(\\bar\\phi_m-\\bar\\theta_m\\right)$, which makes the differenced measurement add coherently and gives $\\beta_{**}=4M^4$ and $\\gamma_{**}=N_T^2P_T$, converting RIS size and transmit power directly into Fisher information. The ellipsoid is then fed into a Wald test and a Bayesian state-inference model to produce detection probabilities for structural damage.","core_discovery":"The central claim is that the channel difference between two closely spaced instants with different RIS phase configurations isolates the RIS reflection: $\\mathbf{H}_s[i]=\\mathbf{H}_s[i-1]$, so $\\mathbf{s}_{i,i-1}=(\\mathbf{H}_r[i]-\\mathbf{H}_r[i-1])\\mathbf{f}x+\\mathbf{w}_{i,i-1}$. Fisher information theory then yields a closed-form position error bound in 3D, expressed as an information ellipsoid whose axes are range information intensity, proportional to bandwidth squared, and angle information intensities, proportional to array aperture divided by distance squared. The paper claims that increasing observation time multiplies the equivalent Fisher information by $(T-1)$ under optimized phases, that extra receivers always weakly decrease the Cram\\'er-Rao lower bound, that receivers with self-positioning errors still help unless their error is very large, and that the resulting millimeter-level position error bound supports detection probabilities near 100 percent for millimeter deformations under adequate time, bandwidth, and RIS size. The numerical validation with a Newtonized orthogonal matching pursuit estimator shows delay and angle root-mean-square error close to the derived CRLB.","pith_inferences":["The paper leaves implicit that the same differencing idea could monitor rotation, torsion, and bending by placing several RISs on one structure and combining their estimated displacements, since the single-RIS framework only detects translation.","A testable extension is to apply the method to a real bridge or high-rise with an independent laser or GNSS reference and deliberately create moving background scatterers, because the static-scatterer assumption in Eq. (6) is the first premise likely to break at millimeter accuracy.","An unproven corner is Theorem 5's assumption that the receiver position-error covariance is diagonal in the target-centric basis $\\mathbf{U}$; a numerical test with a Cartesian-diagonal $\\mathbf{Q}_R$ that is not diagonal in $\\mathbf{U}$ would show how much optimism remains."],"forward_implications":["If the central claim holds, a 16x16 RIS with an 8x8 receive array can reach a 100 percent detection probability for millimeter-level deformation by integrating over one 20 ms radio frame with 2240 OFDM symbols.","Cooperative receivers with meter-level self-positioning errors do not degrade the network CRLB, and three receivers with millimeter-level self-positioning errors match a perfectly positioned receiver in the simulated setup.","Deployment Type II, with cooperative receivers spread along the positive y-axis, gives a smaller PEB than Deployment Type I because of spatial diversity, making receiver placement a design lever.","RIS phases optimized at installation remain robust to sub-meter RIS position shifts and to a few degrees of orientation change, so frequent reconfiguration may be unnecessary.","Performance improves monotonically with observation time, receiver count, bandwidth, antenna count, and RIS size under the model."],"supporting_citations":[{"why":"Supplies the diagonal RIS phase-shift model $\\boldsymbol{\\Omega}^{(i)}=\\mathrm{diag}(e^{j\\omega_m^{(i)}})$ and the alternating-phase operation used to create the differenced measurement.","marker":"[20]"},{"why":"Provides the block-diagonal FIM approximation for joint communication and localization that the channel-parameter FIM derivation relies on.","marker":"[29]"},{"why":"Establishes the 3D mmWave localization error-bound geometry, including the parameterization of TOA and AOA with respect to target position.","marker":"[30]"},{"why":"Supplies the fundamental wideband localization framework from which the equivalent Fisher information matrix and information ellipsoid are taken.","marker":"[32]"},{"why":"Provides the theoretical foundation for network localization and navigation, including spatiotemporal cooperation constructs used in the EFIM derivations.","marker":"[34]"},{"why":"The Newtonized orthogonal matching pursuit algorithm used in simulation to estimate delay and angles and to validate the closed-form CRLB expressions.","marker":"[37]"},{"why":"Sets the serviceability limit-state deformation thresholds used to choose the damage-detection threshold in the Wald test and Bayesian inference.","marker":"[18]"}],"fun_headline_variants":["RIS reference points cancel clutter for millimeter SHM","Cellular ISAC with RIS detects millimeter deformations","RIS phases isolate reflections to spot tiny structural shifts","Cooperative RIS-ISAC achieves millimeter-level SHM","ISAC networks use RIS to monitor structural health at mm scale"],"cache_read_input_tokens":25344,"weakest_assumption_plain":"The load-bearing premise is that the background multipath channel is identical at the two closely spaced measurement instants, as stated in Eq. (6), so subtracting adjacent measurements removes all non-RIS reflections; if any scatterer moves or changes between instants, its residual becomes unmodeled clutter that can swamp millimeter-level detection.","fun_headline_variants_meta":{"raw":{"variants":["RIS reference points cancel clutter for millimeter SHM","Cellular ISAC with RIS detects millimeter deformations","RIS phases isolate reflections to spot tiny structural shifts","Cooperative RIS-ISAC achieves millimeter-level SHM","ISAC networks use RIS to monitor structural health at mm scale"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000551,"raw_usage":{"total_tokens":2649,"prompt_tokens":989,"completion_tokens":1660,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":605,"completion_tokens_details":{"reasoning_tokens":1583}},"tokens_in":605,"tokens_out":1660,"duration_ms":12849,"temperature":1.0,"reasoning_tokens":1583,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:23:45.551151+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take two adjacent measurements with the RIS alternating phases while a scatterer such as a passing car or a pedestrian changes the background channel between them, and check whether the estimated RIS position still tracks a known millimeter-level displacement applied to the structure; the model predicts no background residual, so any observed bias or PEB increase beyond the predicted bound would falsify the static-scatterer assumption. A second check is to simulate a receiver position-error covariance that is diagonal in Cartesian coordinates but not diagonal in the target basis $\\mathbf{U}$, which would break the assumptions behind Theorem 5.","supporting_citations":[{"cited_title":"Joint communication and localization in millimeter wave networks,","cited_arxiv_id":null,"evidence_quote":"Provides the block-diagonal FIM approximation for joint communication and localization that the channel-parameter FIM derivation relies on."},{"cited_title":"Error bounds for uplink and downlink 3D localization in 5G millimeter wave systems,","cited_arxiv_id":null,"evidence_quote":"Establishes the 3D mmWave localization error-bound geometry, including the parameterization of TOA and AOA with respect to target position."},{"cited_title":"Fundamental limits of wideband localization — Part I: A general framework,","cited_arxiv_id":null,"evidence_quote":"Supplies the fundamental wideband localization framework from which the equivalent Fisher information matrix and information ellipsoid are taken."},{"cited_title":"A theoretical foundation of network localization and navigation,","cited_arxiv_id":null,"evidence_quote":"Provides the theoretical foundation for network localization and navigation, including spatiotemporal cooperation constructs used in the EFIM derivations."},{"cited_title":"Deformations and displacements of buildings and building el- ements at serviceability limit states,","cited_arxiv_id":null,"evidence_quote":"Sets the serviceability limit-state deformation thresholds used to choose the damage-detection threshold in the Wald test and Bayesian inference."}],"review_version":1}