{"id":"c827e7b9-fcae-4fa9-a2eb-d3fc7fed550f","arxiv_id":"2507.07285","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A RIS-based computational imaging method that uses shifted, randomly aimed directive beams to form focused speckle patterns and reconstructs scenes from fewer measurements than scanning.","lead":"This paper proposes a microwave imaging method in which reconfigurable intelligent surfaces shape overlapping beams into a focused speckle pattern that encodes the scene. Simulations show the approach can reconstruct targets with fewer measurements than raster scanning while suppressing clutter, though no hardware experiment is included.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Inverse crime: HINV is computed from the same ideal simulator that generated the data, so the few-measurement and clutter advantages may not survive model mismatch.","rationale":"The reader's weakest assumption correctly identifies the same-model inversion as the load-bearing issue. The paper's central claim is that focused speckle patterns enable high-quality imaging with few measurements and improved clutter rejection. For that claim to hold, the inversion must accurately model the true fields produced by the RIS. By computing HINV from the same simulator used to generate the measurements, the paper demonstrates only the ideal case where the inverse perfectly matches the forward operator. This is a textbook inverse crime, and it systematically inflates the apparent performance of computational imaging techniques. The self-consistent simulation provides no evidence about how the method behaves when the true sensing matrix differs from the assumed one—exactly the situation in any real implementation. The paper's own statement that it uses 'an ideal model for RISs' and assumes 'no coupling between the elements' is an explicit acknowledgment of this limitation, and the subsequent note that phase quantization would degrade all methods does not address the core issue: model mismatch affects the inversion quality, not just the illumination efficiency. Therefore, the numerical results should be interpreted as an upper bound. The proposed concrete test—perturbing the forward model while keeping the inversion ideal—would directly quantify this sensitivity. If the reconstruction degrades substantially, the central claims would need to be revised or qualified. If it does not, the concern is resolved. Since the reader already recommended CONDITIONAL based on this same concern, the stress-test does not change the verdict.","tokens_in":7452,"tokens_out":4417,"duration_ms":49887,"concrete_test":"Generate simulated measurements using a perturbed forward model that differs from the ideal HINV: e.g., add independent uniformly distributed phase errors in [-10°, +10°] to each RIS element for every mask/frequency, or use a method-of-moments simulator to compute the true fields. Then reconstruct the 3x3 target (Fig. 10) with the unperturbed HINV at 20 dB SNR and 100 measurements. If the target detection quality (e.g., peak-to-sidelobe ratio or localization error) degrades by more than ~20% relative to the ideal-model result, the few-measurement claim is not robust to model mismatch. Repeat for the clutter scenario of Fig. 11.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that focused speckle patterns yield high SNR, low clutter, and few-measurement reconstructions rests on the accuracy of the sensing matrix used for inversion. In Section II, the forward data are generated via Eq. (4) with HFWD computed by evaluating the ideal Tx/Rx fields at each pixel (Eq. 2). The inverse matrix HINV is then computed 'for the same M measurement configurations, in a similar manner as in equation (2),' i.e., from the same ideal simulator. This is a classic inverse crime: the inversion operator is perfectly matched to the data-generation operator, so no model error is present. In practice, RIS element phase errors, mutual coupling, phase quantization, and unmodeled scattering will make the true sensing matrix differ from HINV. The computed reconstructions in Figs. 10 and 11—especially the claimed robustness to clutter and 100-measurement performance—are therefore upper bounds that may not transfer to experiment. The paper's limitation note that phase quantization 'would degrade all methods' does not address this issue, because the proposed method's advantage specifically relies on precise knowledge of the focused speckle pattern; model mismatch corrupts the inversion for all methods, but the comparison in the paper is conducted with zero mismatch. The SNR/clutter/few-measurement claims are thus untested under realistic forward-model uncertainty.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a microwave imaging method that combines radar coincidence imaging (RCI) with computational imaging using reconfigurable intelligent surfaces (RISs). Multiple RISs redirect signals from a transmitter toward a region of interest, and the interference of the resulting directive beams forms speckle-like patterns that encode the reflectivity of the ROI. The authors argue that because the speckle patterns are formed by directive beams rather than fully random patterns, the method enjoys higher SNR and reduced clutter, while still allowing high-quality reconstruction from fewer measurements than raster scanning. The forward model assumes ideal lossless RIS elements with continuous phase control, no mutual coupling, and perfect knowledge of the sensing matrix. Numerical simulations compare the proposed method with raster scanning and random-pattern approaches for a 2D target at 8 m, including a clutter robustness study. The paper concludes that the method combines the benefits of spotlight and computational imaging and suggests applications in security screening and user tracking.","tokens_in":7764,"tokens_out":2132,"duration_ms":25999,"significance":"If validated, the proposed concept would be a useful contribution to RIS-enabled computational imaging, as it directly addresses the SNR and clutter limitations of random-pattern metasurface imagers while retaining the few-measurement advantage of computational imaging. The idea of confining speckle diversity to a predefined ROI via multiple directive beams is conceptually sound and grounded in prior computational imaging and RCI work. The SVD analysis of pattern diversity (Fig. 7) is a sensible design tool, and the qualitative comparisons in Figs. 10 and 11 are suggestive. However, the numerical demonstration currently rests on a self-consistent ideal simulator: the inverse sensing matrix is computed from the same ideal model that generates the simulated measurements, constituting an inverse-crime setup. The absence of quantitative metrics, Monte Carlo statistics, and forward-model mismatch tests means that the central claims of improved SNR, clutter robustness, and few-measurement performance are not yet convincingly established. The paper also provides no experimental or realistic error model.","major_comments":[{"comment":"The numerical demonstration commits an inverse crime: the measured data are generated by evaluating the forward sensing matrix HFWD from Eq. (4), while the inverse matrix HINV is computed 'for the same M measurement configurations, in a similar manner as in equation (2),' i.e., from the same ideal simulator. This means the inversion operator is perfectly matched to the data-generation operator, leaving no modeling error. In practice, RIS element phase errors, mutual coupling, phase quantization, and unmodeled scattering will make the true sensing matrix differ from HINV, and the claimed few-measurement and clutter advantages, as shown in Figs. 10 and 11, are upper bounds that may not transfer to experiment. The authors' note that phase quantization 'would degrade all methods' does not address this issue, because the proposed method's advantage specifically relies on precise knowledge of the focused speckle pattern. Please add experiments with model mismatch, for example by perturbing the RIS phase profiles, introducing element coupling, or using a different forward simulator to generate data than the one used for inversion, and report how reconstruction quality degrades.","section":"§II, Eqs. (4) and (2)"},{"comment":"The paper relies entirely on qualitative visual comparisons and contains no quantitative image-quality metric (e.g., normalized mean-square error, peak signal-to-noise ratio, or resolution) and no Monte Carlo averaging over noise or mask realizations. The abstract and introduction claim 'higher SNR' and 'reduced clutter,' but the manuscript does not define how SNR is computed or quantify the improvement over random patterns. The clutter study in Fig. 11 uses a single realization at each clutter location and a high 50 dB SNR, which is not representative of realistic conditions. Please provide quantitative metrics over multiple noise and clutter realizations, and clearly define the SNR used in the simulations, so that the claimed advantages can be assessed rather than inferred from example images.","section":"Figs. 10 and 11"},{"comment":"The inverse problem is solved with a computational technique identified only as 'CGS' (conjugate gradient squared?) but the manuscript gives no details about the number of iterations, stopping criterion, regularization, or the discretization grid (number of voxels N, pixel size, etc.). Since the central claim is that high-quality images are obtained with M < N, it is essential to report these parameters to assess whether the success is due to the proposed patterns or to the specific inversion implementation. Please specify the reconstruction algorithm, regularization, and grid parameters, and consider showing performance as a function of measurement number M for a range of target configurations.","section":"§II, description of the inverse model"}],"minor_comments":[{"comment":"The equation numbering is inconsistent: Eq. (4) is introduced before Eq. (2), and Eq. (2) is used both for the redirection-angle random component and for the forward sensing matrix element. Please renumber the equations sequentially to avoid confusion.","section":"General / Eq. numbering"},{"comment":"There is a typo in 'a an array of 20 × 20 elements.' Also, the sentence 'To implement random patterns, we assign a random phase to each element of the RIS' in §II could be clarified by stating whether the random-phase patterns are used for both Tx and Rx RISs and whether the same random patterns are used across frequency points.","section":"§II, first paragraph"},{"comment":"The SVD analysis is presented as a design-motivating result, but the figure caption does not explain how the SVD of the 'patterns' is computed (e.g., is it the singular values of the sensing matrix for a fixed ROI?). Please clarify the exact quantity being decomposed and describe how the SVD curves support the choice of 20 masks.","section":"§II, Fig. 7"},{"comment":"In the raster scanning description, the text says 'direct the beam toward a given location inside the RIS,' which should presumably read 'inside the ROI.' Please correct this and specify the scan grid and whether scanning is performed over all six Tx/Rx RIS pairs simultaneously.","section":"§II, raster scanning description"},{"comment":"The manuscript does not state the pixel count, ROI discretization, or the exact frequency sampling scheme beyond 'five frequency points ranging from 5.9 GHz to 6.1 GHz.' Please include these values so that the simulations can be reproduced.","section":"Reproducibility"}],"recommendation":"major_revision","confidential_remarks":"The paper is a conceptual proposal whose numerical support is weakened by an inverse-crime setup and a lack of quantitative validation. The central idea is worth publishing if the authors can demonstrate that the focused-speckle approach remains advantageous under model mismatch and realistic noise. I would ask for a revised manuscript with model-mismatch tests, quantitative metrics, and Monte Carlo statistics. The current version is not ready for acceptance, but I do not see a fundamental flaw in the proposed method itself; the issues are fixable within the scope of a numerical study."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuine idea, not a repackaging. Combining radar coincidence imaging with RIS-generated focused speckle—random phase offsets plus small random beam-angle jitter per panel—is new relative to the random-pattern and raster-scan RIS papers they cite. The SVD analysis in Fig. 7 is an honest, useful check that random offsets alone saturate diversity and that angle jitter extends it. The comparisons in Figs. 10 and 11 are internally consistent, and the proposed method visibly beats the two baselines under their model. The paper is clearly written and does a fair job of keeping the simulation settings identical across methods.\n\nThe soft spot is the one the stress test flags, and it is not minor. The data are generated with HFWD from Eq. (2), and HINV is computed for the same M measurement configurations in a similar manner as in equation (2). That makes the inversion operator perfectly matched to the data generator. No mutual coupling, no phase quantization, no calibration error, and no unmodeled scattering enter the comparison. Their limitation paragraph says phase quantization would degrade all methods, but that misses the point: the proposed method’s advantage depends on knowing the focused speckle pattern precisely, so model mismatch is not equally damaging across methods. The 100-measurement reconstruction and the clutter robustness are upper bounds, not demonstrated properties. There are also no Monte Carlo runs, no error bars, and no quantitative metrics; Figs. 10 and 11 are single realizations at fixed SNR. That is thin for the size of the claims.\n\nI don’t want to overstate the problem. The conceptual mechanism is independently grounded in existing computational imaging and RCI literature, no constants are fitted to data, and the paper is transparent about what was simulated. The inverse-crime issue is fixable: add a mismatched forward model (random phase errors, calibration perturbations), run Monte Carlo over noise and clutter realizations, and show the proposed method still beats the baselines. Hardware validation would be the strongest proof, but even a mismatched-model numerical study would make the claims credible.\n\nWho gets value: researchers working on RIS-based microwave imaging and computational imaging. I would bring this to a reading group as a worthwhile example of a plausible new mechanism with a clear methodological caveat. It deserves a serious referee: the idea is novel enough and the presentation careful enough. I would send it out, but with a clear request to address the inverse-crime concern before acceptance.","headline":"A real and clearly explained new combination of RCI-style coincidence imaging with directive-beam speckle on RISs, but the SNR/clutter/few-measurement claims rest on an inverse-crime simulation that needs to be addressed before the results can be believed.","tokens_in":8202,"tokens_out":2205,"would_cite":false,"duration_ms":26349,"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":"The paper claims that forming speckle patterns from directive RIS beams, rather than from random patterns, yields higher SNR, lower clutter, and high-quality images from fewer measurements than raster scanning.","keywords":["computational imaging","reconfigurable intelligent surface","radar coincidence imaging","speckle patterns","microwave imaging","frequency diversity","compressive sensing","clutter rejection"],"falsifier":"Measure the complex fields radiated by the six transmit and six receive RIS panels in an anechoic chamber, build the inverse sensing matrix from those measured fields, and run the same 100-measurement reconstruction of a 3-by-3 scatterer grid at 20 dB SNR. If the nine scatterers are no longer resolved, or if a clutter object at y = 3 m visibly corrupts the image, the claimed few-measurement and clutter-rejection advantages do not survive contact with a real device.","tokens_in":7278,"feed_emoji":"📡","tokens_out":12367,"duration_ms":122256,"temperature":0.7,"pith_summary":"The paper proposes an imaging method in which several reconfigurable intelligent surfaces (RISs) redirect the transmitter's signal toward the same region of interest, each with a different phase offset and a slight random shift of its beam direction. The interference of these directive beams creates a speckle pattern confined to the ROI, so the method gets the spatial diversity of random-pattern computational imaging together with the high signal level and clutter rejection of spotlight scanning. The authors model the measured signal as the product of transmit and receive fields at each pixel and invert the resulting sensing matrix computationally. In numerical tests at 6 GHz with 100 measurements, their method resolves a 3-by-3 grid of subwavelength scatterers where raster scanning aliases and random-pattern imaging is buried in noise, and it keeps the image clean when a clutter object sits outside the ROI. If the result survives hardware testing, RIS-based security screening and wireless user tracking could image a small region with fewer measurements and no mechanical movement.","feed_headline":"Focused-beam speckle imaging cuts RIS measurements, noise, and clutter","feed_subtitle":"Directive-beam speckle gives the SNR of spotlight imaging with the few-measurement power of computational imaging.","key_machinery":"The central object is the focused speckle pattern: the coherent interference of several RIS-radiated directive beams, all steered into the same ROI but each carrying a random phase offset and a random perturbation of its redirection angles $\\theta$ and $\\phi$ chosen to keep the beam inside the ROI. Each such configuration is called a mask; changing the mask changes the speckle pattern, and sweeping frequency from 5.9 to 6.1 GHz adds diversity and range information. The sensing matrix is computed from simulated fields at every pixel, and the same simulator is reused to build the inverse matrix used for reconstruction; an SVD analysis shows that phase offsets alone exhaust pattern diversity once the number of measurements exceeds the product of the number of Tx and Rx RISs, while randomizing the redirection angles extends the pattern rank.","core_discovery":"The paper's central claim is that focused speckle patterns encode the reflectivity of the whole ROI in a few measurements while retaining the SNR and clutter rejection of directive beams. Under the first-Born (single-scattering) approximation, the measured signal is $g = H_{\\mathrm{FWD}} \\sigma$, with $h(i,j) = E_{\\mathrm{Tx}}^{(i)}(\\mathbf{r}_j) \\cdot E_{\\mathrm{Rx}}^{(i)}(\\mathbf{r}_j)$, and the inverse problem is solved computationally using an inverse sensing matrix built from the same simulator. In simulations with five frequency points from 5.9 to 6.1 GHz and 20 masks (100 measurements) at 20 dB SNR, the method resolves a 3-by-3 grid of subwavelength scatterers at z = 8 m, while raster scanning aliases and random-pattern imaging is noise-dominated. With a clutter object at y = 3, 5, or 10 m, the proposed reconstruction stays clean while the random-pattern reconstruction degrades.","pith_inferences":["The paper does not state this, but a direct hardware test should compare the singular-value spectrum of the measured sensing matrix with the simulated one; if angle randomization no longer extends the rank beyond the phase-offset plateau, the few-measurement advantage will not transfer to practice.","The random perturbation range $C_\\theta, C_\\phi$ sets a resolution-vs-diversity tradeoff: wider perturbations spread energy toward the ROI edges, so an optimal distribution could be tuned per ROI, a knob the paper leaves unexplored.","The same focused-speckle mechanism could serve real-time user tracking by reconstructing a coarse reflectivity map from very few masks, since locating a person requires less spatial detail than resolving fine scatterers.","Because the illumination is confined to the ROI, the approach may also reduce scattered power from moving objects outside the ROI, which could make privacy-preserving through-wall sensing more practical than random-pattern alternatives."],"forward_implications":["An RIS-based imager can reconstruct a 2D region of interest with fewer measurements than raster scanning, because computational inversion fills in spatial frequencies that are not directly sampled.","Concentrating the radiated power into the ROI raises the signal-to-noise ratio relative to random-pattern computational imaging at the same transmit power.","Out-of-region clutter contributes less to the reconstruction, so imaging in crowded areas such as security checkpoints or rooms with moving people should become more reliable.","The same hardware can switch to a different ROI simply by recomputing the redirection angles and phase offsets, with no mechanical movement.","Adding more frequency points extends the approach toward range resolution, a natural path from 2D reflectivity maps to 3D volumetric images."],"supporting_citations":[{"why":"Defines the metasurface computational-imaging framework with random patterns that the proposed method is designed to improve.","marker":"[10]"},{"why":"Provides a representative large-aperture random-pattern imaging demonstration used as a conceptual baseline.","marker":"[9]"},{"why":"Introduces radar coincidence imaging, the idea of encoding a scene in a few spatially diverse measurements solved computationally.","marker":"[29]"},{"why":"Supports directive-beam raster scanning, whose SNR and clutter properties the proposed method aims to retain.","marker":"[30]"},{"why":"Explains how frequency diversity fills k-space, supporting the frequency sweep that reduces aliasing in few-measurement reconstructions.","marker":"[32]"},{"why":"Supplies the simulation platform used to compute the fields and to build both forward and inverse sensing matrices.","marker":"[33]"}],"fun_headline_variants":["RIS-focused beams yield radar images in few measurements","Computational radar with RIS: focused speckle, less noise","Focused beams on RIS cut radar imaging measurements and clutter","Speckle from directive beams boosts radar imaging efficiency","RIS radar imaging: speckle from focused beams, fewer shots"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the inverse sensing matrix is computed from the same ideal simulator that generated the data, so the numerical test contains none of the real-world discrepancies, such as mutual coupling, phase quantization, calibration error, or unmodeled scattering, that would make the simulated and actual RIS fields differ.","fun_headline_variants_meta":{"raw":{"variants":["RIS-focused beams yield radar images in few measurements","Computational radar with RIS: focused speckle, less noise","Focused beams on RIS cut radar imaging measurements and clutter","Speckle from directive beams boosts radar imaging efficiency","RIS radar imaging: speckle from focused beams, fewer shots"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000279,"raw_usage":{"total_tokens":1655,"prompt_tokens":941,"completion_tokens":714,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":557,"completion_tokens_details":{"reasoning_tokens":634}},"tokens_in":557,"tokens_out":714,"duration_ms":6880,"temperature":1.0,"reasoning_tokens":634,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:44:16.517101+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the complex fields radiated by the six transmit and six receive RIS panels in an anechoic chamber, build the inverse sensing matrix from those measured fields, and run the same 100-measurement reconstruction of a 3-by-3 scatterer grid at 20 dB SNR. If the nine scatterers are no longer resolved, or if a clutter object at y = 3 m visibly corrupts the image, the claimed few-measurement and clutter-rejection advantages do not survive contact with a real device.","supporting_citations":[{"cited_title":"Review of metasurface antennas for computational microwave imag- ing,","cited_arxiv_id":null,"evidence_quote":"Defines the metasurface computational-imaging framework with random patterns that the proposed method is designed to improve."},{"cited_title":"Large metasurface aperture for millimeter wave computational imaging at the human-scale,","cited_arxiv_id":null,"evidence_quote":"Provides a representative large-aperture random-pattern imaging demonstration used as a conceptual baseline."},{"cited_title":"Phaseless radar coin- cidence imaging with a mimo sar platform,","cited_arxiv_id":null,"evidence_quote":"Introduces radar coincidence imaging, the idea of encoding a scene in a few spatially diverse measurements solved computationally."},{"cited_title":"Synthetic aperture radar with dynamic metasurface antennas: a conceptual development,","cited_arxiv_id":null,"evidence_quote":"Supports directive-beam raster scanning, whose SNR and clutter properties the proposed method aims to retain."},{"cited_title":"Spatially resolving antenna arrays using frequency diversity,","cited_arxiv_id":null,"evidence_quote":"Explains how frequency diversity fills k-space, supporting the frequency sweep that reduces aliasing in few-measurement reconstructions."},{"cited_title":"Comprehensive simulation platform for a metamaterial imaging system,","cited_arxiv_id":null,"evidence_quote":"Supplies the simulation platform used to compute the fields and to build both forward and inverse sensing matrices."}],"review_version":1}