{"id":"9cf2e95f-4b5b-49f2-a8f6-52f56cfc3966","arxiv_id":"2504.20661","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A bistatic ISAC system with sensing-optimized stacked intelligent metasurfaces achieves large gains in range/velocity estimation and bit error rate over a no-metasurface baseline across OFDM, OTFS, and AFDM waveforms.","lead":"Stacked intelligent metasurfaces, tunable layers of antenna elements, are used to improve a wireless system that both communicates and senses moving targets. Simulations show that optimizing the metasurface phases for sensing improves both radar accuracy and data transmission compared to a system without metasurfaces.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The SIM optimizer (eqs. 8-10, Algorithm 1) requires exact path gains, delays, Dopplers, and angles that the proposed RPE does not estimate—angles are explicitly deferred—so the reported gains may be a genie-aided upper bound.","rationale":"The reader and I converge on the same weakest point: the SIM optimization assumes perfect knowledge of channel parameters that the proposed RPE does not actually provide. I looked for an outright mathematical error, e.g., in the chain-product definitions (1)-(2) or the gradient expressions (9), but the text provides no derivation and no code, so the strongest documented gap is the genie-knowledge dependence rather than a demonstrable algebraic mistake. The paper is a coherent simulation study, and the optimization objective is well-defined under perfect knowledge; coherently combining hundreds of meta-atom phases could plausibly produce large gains. However, Algorithm 1's SIM-optimization stage cannot be executed from the outputs of the PDA stage, which estimates only delays and Dopplers, and the paper explicitly defers angle estimation to future work. If the true angles, gains, delays, and Dopplers are replaced by imperfect estimates, the optimized phases may be misaligned and the headline gain in Fig. 3 may shrink or disappear. This does not refute the theoretical interest of SIM-aided ISAC, but it means the claimed 'large gain' is currently an upper-bound-style result. A conditional acceptance, asking for a sensitivity analysis or a two-stage estimate-then-optimize evaluation, is therefore appropriate and consistent with the original verdict.","tokens_in":11356,"tokens_out":6904,"duration_ms":79075,"concrete_test":"Modify the Fig. 3 simulation so SIM optimization uses corrupted parameters instead of ground truth: at each Monte-Carlo run, perturb each path's delay and Doppler by one grid step with probability p_off, add zero-mean Gaussian errors to hp and to the four angles with standard deviations matching the PDA/RPE output error at the operating SNR, then optimize phases and recompute range/velocity MSE and BER. If the sensing-optimized curves no longer beat the no-SIM baseline by the reported margin—or if MSE degrades by more than roughly 3 dB at mid SNR—the central claim is conditional on genie channel knowledge.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that sensing-optimized SIM phases yield large gains in bistatic RPE accuracy and BER relative to no-SIM operation. To evaluate the objective (8), the gradients (9), and the active weakest path in Algorithm 1, one must know, for every path, the complex gain hp, delay tau_p, Doppler nu_p, and both AoA/AoD pairs. The RPE stage in Section IV estimates only delays and Dopplers via sparse recovery; Section IV-A explicitly states that AoA/AoD estimation is 'left... to be addressed in a follow-up work.' Moreover, Algorithm 1 places the SIM optimization before the PDA-based RPE, so in a real bistatic ISAC deployment the phases would have to be configured from prior or erroneous estimates. No sensitivity analysis is given, so the large gains plotted in Fig. 3 could be an artifact of feeding true channel parameters into the optimizer. This is load-bearing because the paper presents Algorithm 1 as a complete proposed ISAC scheme, not as a genie-aided benchmark, and the claimed advantage rests on phases that cannot be computed from the estimates the scheme itself produces.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a bistatic integrated sensing and communications (ISAC) architecture in which stacked intelligent metasurfaces (SIMs) at the transmitter and receiver are optimized to improve both radar parameter estimation and communication performance over doubly-dispersive channels. The SIM phases are tuned by solving a min-max problem that maximizes the weakest path gain using a steepest-ascent algorithm with claimed closed-form gradients, and the radar parameters (delays and Dopplers) are estimated by a compressed sensing-based probabilistic data association (PDA) algorithm. The scheme is evaluated via simulations for OFDM, OTFS, and AFDM waveforms, reporting large gains in range/velocity MSE and BER relative to a no-SIM baseline. The central claim is that sensing-optimized SIM phases yield these gains across all three waveforms.","tokens_in":11632,"tokens_out":2564,"duration_ms":28840,"significance":"If the reported gains are robust, the paper would be a useful step toward applying SIMs in ISAC: it provides a concrete signal model, an explicit optimization formulation, a sparse-recovery-based RPE algorithm, and a comparison across three waveforms. The reuse of the authors' prior channel model and PDA framework gives the derivation continuity, and the closed-form gradient expressions are a potentially valuable contribution. However, the central simulation claim currently rests on a genie-aided assumption that the SIM optimizer knows the true path parameters, including angles that the proposed RPE does not estimate; without a sensitivity analysis or a repositioning of the results as an upper bound, the practical significance of the claimed gains is not established. The paper deserves major revision rather than rejection because the issue is fixable within the manuscript's scope.","major_comments":[{"comment":"The SIM optimization objective, gradients, and the selection of the active weakest path in Algorithm 1 require exact knowledge of the path parameters {hp, τp, νp} and both AoA/AoD pairs through the terms h~p and Bp in Eq. (8). The RPE stage in Section IV estimates only delays and Dopplers, and Section IV-A explicitly leaves AoA/AoD estimation to future work. Consequently, in the numerical results of Section V, the SIM phases are computed from ground-truth channel parameters, so the large gains in Fig. 3 are a genie-aided upper bound. Since the paper presents Algorithm 1 as a complete bistatic ISAC scheme rather than as a benchmark, this is a load-bearing issue: the authors should either add an analysis of how estimation errors in the path parameters propagate into the optimized phases and the resulting MSE/BER, or clearly frame the current results as an upper-bound study.","section":"Section III-A, Eqs. (8)–(10); Algorithm 1"},{"comment":"The closed-form sub-gradient expressions in Eq. (9) are stated without derivation. Because the entire optimization result depends on these gradients being correct, the authors should provide a derivation or an explicit reference where they are proved. A short verification in an appendix would also help; as written, the reader cannot check whether the gradient computation matches the objective (8) and the layered SIM structure in Eqs. (1)–(2).","section":"Eq. (9)"},{"comment":"The numerical results consist of single curves without error bars, confidence intervals, or multiple Monte Carlo trials. The central claim of a 'large gain' is read from one set of realizations, and the convergence illustration in Fig. 2 is a single example with P=3. The authors should report averaged results over multiple random channel realizations and random SIM initializations, or otherwise quantify the variability; without that, the reader cannot assess whether the reported gains are statistically robust or an artifact of the chosen realization.","section":"Section V-A, Fig. 3"},{"comment":"Algorithm 1 loops over paths and re-selects the index p for which the objective (8) is minimized, but this selection is computed using true path powers. In a practical system, the estimated channel coefficients from the PDA stage would contain errors, potentially changing which path is the weakest and altering the SIM update direction. The manuscript offers no analysis of this sensitivity, which is particularly important because the PDA stage is placed after the SIM optimization in Algorithm 1, meaning the phases would actually be configured from prior or erroneous estimates in a closed-loop deployment.","section":"Algorithm 1, step 1 and Section IV-A"}],"minor_comments":[{"comment":"There are several typos and inconsistent notations: 'metasurfacess' in the title and abstract, 'shits' in footnote 7, and 'A V' in reference [7]. These should be corrected.","section":"Throughout"},{"comment":"The captions use tK=K0=12 without defining these quantities in the text; the grid sizes Kτ and Dν from Eq. (12) should be clearly related to the resolution limit shown in Fig. 3(a).","section":"Figure 3 captions"},{"comment":"The dimensions of the SIM transfer functions v and u are given but the ordering of products in Eqs. (1)–(2) could be clarified with a brief explanation of which layer the index q refers to, as this is essential for understanding the gradient expressions in Eq. (9).","section":"Section II-A"},{"comment":"Footnotes stating that extensions are 'relegated to a journal version' are informal for a journal submission and should either be implemented or rephrased as standard future-work statements.","section":"Footnotes 13 and 14"}],"recommendation":"major_revision","confidential_remarks":"The manuscript relies heavily on the authors' own prior work ([17], [32]), and while reuse is not problematic per se, the novelty claim should be scoped carefully. The most pressing issue is the genie-aided SIM optimization; if the authors can provide a robustness analysis or clearly reposition the results as an achievable upper bound, the paper would be a solid contribution. The lack of error bars is also a concern given the strong conclusions drawn from single realizations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this paper for the combination, not for the performance numbers yet. The authors take their own MPDD channel model, add a min-max optimization of TX/RX SIM phases over the weakest propagation path, and couple it with a PDA-based sparse estimator for delay and Doppler. As a system design, it is coherent and the across-waveform comparison (OFDM, OTFS, AFDM) is useful. The results show consistent gains in both sensing MSE and BER when the SIM is optimized for sensing.\n\nNow the load-bearing problem. The SIM optimization in (8)-(10) and Algorithm 1 needs exact values of the path gains, delays, Dopplers, and both AoA/AoD pairs. The RPE stage only estimates delays and Dopplers; the paper explicitly defers AoA/AoD estimation to follow-up work. Algorithm 1 runs the SIM optimization before any RPE, so in any real deployment the phases would have to come from prior or error-contaminated estimates. There is no sensitivity analysis. That means the large gains in Figure 3 are, as presented, a genie-aided upper bound. This is not a minor omission; it is the central claim of the paper.\n\nTwo lesser issues. The gradients in (9) are asserted as closed-form without derivation; a reader cannot check them without redoing the algebra. The convergence behavior in Figure 2 is one example, not a proof, and the Monte Carlo curves appear to be single realizations without error bars. These are fixable with moderate effort.\n\nWhat holds up: the system model is sound, the optimization problem is plausibly formulated, and the PDA sparse estimation is a reasonable adaptation of prior work. The paper is honest about some limitations, just not the genie-aided one. It is a solid architecture-level contribution to the SIM/ISAC subfield, not a breakthrough.\n\nI would send this to peer review. A good referee will ask for the gradient derivation, a practical phase-configuration strategy, and at least a basic sensitivity study on imperfect parameter knowledge. Without those, the performance claims stay conditional. I wouldn't cite the numbers in the next year, but I would track it.","headline":"Genuine new SIM/ISAC combination worth refereeing, but the main performance claim rests on perfect channel knowledge the proposed receiver never provides.","tokens_in":12153,"tokens_out":4909,"would_cite":false,"duration_ms":40421,"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":"By tuning stacked intelligent metasurfaces at both ends of a bistatic link to strengthen the weakest propagation path, this paper shows that a single phase configuration can simultaneously cut radar range/velocity estimation error and…","keywords":["stacked intelligent metasurfaces","integrated sensing and communications","bistatic radar","doubly-dispersive channels","OTFS","AFDM","probabilistic data association","sparse signal recovery"],"falsifier":"Run the same SIM optimization with path parameters deliberately corrupted by measurement errors (for example, delay off by one grid step, Doppler off by a few hertz, angle off by a few degrees) and compare range/velocity MSE and BER with the perfect-knowledge curves in Fig. 3; if the gains largely disappear, the practical claim fails.","tokens_in":11187,"feed_emoji":"📡","tokens_out":8488,"duration_ms":81987,"temperature":0.7,"pith_summary":"This paper claims that equipping both ends of a bistatic integrated sensing and communications link with a stacked intelligent metasurface—a layered array of reconfigurable metasurfaces whose phase shifts can be tuned—can substantially improve radar range/velocity estimation accuracy and communication bit error rate at the same time. The proposed approach tunes the SIM phases so that the weakest propagation path carries as much power as possible, an objective formulated as a max-min problem and solved by steepest ascent with closed-form gradients. Radar parameter estimation is then carried out by a compressed-sensing-based probabilistic data association receiver operating on a delay-Doppler grid. Simulations with OFDM, OTFS, and AFDM waveforms show large gains over a system without SIMs, and show that the sensing-oriented SIM configuration also preserves most of the communication benefit. If correct, this would make a single tunable metasurface front-end support both radar and communication in high-mobility, doubly dispersive channels.","feed_headline":"Tuned metasurface layers sharpen radar and cut bit errors","feed_subtitle":"A single SIM phase configuration, optimized for the weakest path, lifts bistatic ISAC performance across three waveforms.","key_machinery":"The central object is the parametrized stacked intelligent metasurface (SIM), a layered set of reconfigurable metasurfaces whose per-layer phase shifts can be tuned. The paper models the transmitter SIM as a vector $\\mathbf{v}$ and the receiver SIM as a vector $\\mathbf{u}$, each built from products of per-layer phase-shift diagonal matrices and diffraction matrices. These vectors enter every path gain $\\check{h}_p$ in equation (7), so the channel itself is a function of the phase vectors $\\mathbf{Z}$ and $\\widetilde{\\mathbf{Z}}$. The engine of the argument is the max-min objective (8), which maximizes the weakest path gain, together with the closed-form gradients in (9) that let a greedy steepest-ascent algorithm (Algorithm 1) tune the phases despite the non-convexity. On the sensing side, the received signal is recast as a sparse dictionary recovery problem, and a Bernoulli-Gaussian probabilistic data association (PDA) message-passing receiver estimates the non-zero delay-Doppler taps.","core_discovery":"The central claim, stated on the paper's own terms, is that the phase configuration of the TX and RX SIMs should be chosen to maximize the channel gain of the weakest path, and that doing so yields large improvements in both sensing and communication. Under the metasurface-parametrized doubly dispersive model, each path gain is a scalar function of the SIM phase vectors (equation (7)), so the phases can be optimized before the receiver estimates anything. The optimization itself is the max-min problem in equation (8), whose closed-form gradients (9) drive a greedy steepest-ascent loop (Algorithm 1). The accompanying radar parameter estimator reformulates the received signal as a sparse recovery problem over a delay-Doppler grid and solves it with a Bernoulli-Gaussian PDA message-passing algorithm. The numerical section reports, for OFDM, OTFS, and AFDM, that the sensing-optimized SIM beats both the no-SIM system and the communication-optimized SIM for radar estimation, while still improving communication BER relative to no-SIM.","pith_inferences":["Editorial inference: the reported gains are computed under perfect channel-parameter knowledge; a natural next test is to feed the PDA estimates back into the SIM optimization and measure how much of the gain survives estimation noise.","Editorial inference: the max-min objective can be seen as worst-path diversity enhancement, and the same principle could extend to MIMO SIMs by replacing scalar path gains with matrix gains, likely yielding a joint beamforming-and-phase-tuning problem with the same structure.","Editorial inference: because sensing-only tuning retains most of the communication benefit while communication-only tuning fails for sensing, the sensing-communication tradeoff may be asymmetric; a Pareto or weighted objective could recover the small remaining BER loss without sacrificing radar accuracy."],"forward_implications":["With the SIM phases set by the sensing objective, the same physical layer supports both functionalities: radar range/velocity MSE and communication BER improve together relative to a no-SIM system.","The RPE formulation no longer requires the receiver to know the number of paths in advance; non-zero entries of the estimated sparse channel vector mark the delay-Doppler grid points, from which delays and Doppler shifts are read off.","Across the three waveforms tested, the sensing-optimized SIMs improve both figures of merit, with OTFS and AFDM retaining larger BER gains than OFDM under the same SIM configuration.","The communication-only SIM optimization from the authors' prior channel-model work fails for sensing, while the sensing-only optimization preserves much of the communication gain, suggesting the sensing objective is the safer single choice for an ISAC system."],"supporting_citations":[{"why":"Defines the metasurface-parametrized doubly-dispersive MIMO model, including the SIM transfer functions and the communication-only SIM optimization used as a baseline in the comparisons.","marker":"[32]"},{"why":"Provides the AFDM input-output relation (5) and the per-waveform effective path matrices (6) that define the received signal for OFDM, OTFS, and AFDM.","marker":"[25]"},{"why":"Gives the Rayleigh-Sommerfeld diffraction model and correlation matrix structure through which the SIM phase shifts shape the channel path gains.","marker":"[29]"},{"why":"Supplies the PDA-based message-passing machinery (soft interference cancellation, belief generation, replica generation) that the proposed RPE algorithm adapts.","marker":"[17]"},{"why":"Establishes the near-equivalence of OFDM, OTFS, and AFDM for communication-centric ISAC, which the numerical study uses as its no-SIM reference point.","marker":"[16]"}],"fun_headline_variants":["Weakest-path SIM tuning improves bistatic ISAC radar and link","Metasurface phases set to weakest path lift ISAC sensing and comm","One SIM phase configuration, selected for weakest path, aids radar and BER","Bistatic ISAC with SIMs: weakest-path optimization sharpens sensing","Stacked metasurfaces tuned to weakest path improve ISAC radar and link"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole optimization stage assumes the true delays, Doppler shifts, complex gains, and arrival/departure angles of every path are known when computing the objective and gradients; in a real bistatic ISAC setting these must be estimated and will carry error.","fun_headline_variants_meta":{"raw":{"variants":["Weakest-path SIM tuning improves bistatic ISAC radar and link","Metasurface phases set to weakest path lift ISAC sensing and comm","One SIM phase configuration, selected for weakest path, aids radar and BER","Bistatic ISAC with SIMs: weakest-path optimization sharpens sensing","Stacked metasurfaces tuned to weakest path improve ISAC radar and link"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00023,"raw_usage":{"total_tokens":1449,"prompt_tokens":878,"completion_tokens":571,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":494,"completion_tokens_details":{"reasoning_tokens":474}},"tokens_in":494,"tokens_out":571,"duration_ms":6120,"temperature":1.0,"reasoning_tokens":474,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T05:23:04.471551+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same SIM optimization with path parameters deliberately corrupted by measurement errors (for example, delay off by one grid step, Doppler off by a few hertz, angle off by a few degrees) and compare range/velocity MSE and BER with the perfect-knowledge curves in Fig. 3; if the gains largely disappear, the practical claim fails.","supporting_citations":[{"cited_title":"Stacked intelligent metasurfaces for efficient holographic MIMO communications in 6G,","cited_arxiv_id":null,"evidence_quote":"Gives the Rayleigh-Sommerfeld diffraction model and correlation matrix structure through which the SIM phase shifts shape the channel path gains."},{"cited_title":"On the effective- ness of OTFS for joint radar parameter estimation and communication,","cited_arxiv_id":null,"evidence_quote":"Establishes the near-equivalence of OFDM, OTFS, and AFDM for communication-centric ISAC, which the numerical study uses as its no-SIM reference point."}],"review_version":1}