{"id":"01b18ffd-8c6e-4ce6-892c-71ab83ef1c9b","arxiv_id":"2506.14985","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A unified doubly-dispersive MIMO channel model with SIM and RIS is derived, and SIM phase optimization is shown to improve BER and radar parameter estimation for OFDM, OTFS, and AFDM.","lead":"This paper builds a channel model for wireless links that use stacked intelligent metasurfaces (SIM) and reconfigurable intelligent surfaces (RIS) in fast-moving environments. It derives signal equations for OFDM, OTFS, and AFDM, then shows that tuning the SIM phases improves communication and radar sensing performance in simulations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 5.3 normalizes all channels to equal Frobenius norm, which may cancel the very SIM power gain that Eq. (43) optimizes; the headline BER/RPE gains are therefore not yet interpretable.","rationale":"The reader's verdict rationale already flags the normalization as ambiguous, so we agree on that point. However, the reader's formal weakest_assumption concerns the physical SIM model (inter-layer coupling), which we consider secondary for the central quantitative claim: the BER/RPE demonstrations in Figs. 3-6 do not include RISs and rely on the SIM model from prior work. The normalization issue directly affects whether the reported gains can be interpreted as evidence for the headline claim. If the normalization is applied after optimization, the optimizer's intended power gain is discarded, and the source of the remaining BER differences is unexplained. The proposed test would settle whether the gains survive at equal transmit SNR without normalization, and at equal received power with the no-SIM baseline scaled up. Because this is an addressable methodological ambiguity rather than a proven error, the CONDITIONAL verdict should be retained.","tokens_in":32656,"tokens_out":18657,"duration_ms":188327,"concrete_test":"Re-run the simulations of Figs. 3-5 under two conditions: (a) omit the Frobenius-norm equalization entirely, preserving the actual channel gains produced by the optimized SIM, with all curves plotted against the same transmit E_s/N_0; (b) keep the Section 5.3 equalization but give the 'no SIM' baselines the same Frobenius norm as the optimized SIM channels, so all systems have identical received power. If the optimized-SIM advantage disappears or reverses in (a), the published gains are artifacts of the rescaling. If it persists in both (a) and (b), the central claim is robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 5.3 states that 'the complete channels are normalized such that ||H_OFDM||_F^2 = ||H_OTFS||_F^2 = ||H_AFDM||_F^2 = ||H_MIMO||_F^2 for all the cases.' But the SIM optimization objective in Eq. (43) is to maximize the sum of squared Frobenius norms of the per-path matrices, i.e., the total received signal power. If the optimized channel matrices are subsequently rescaled to have the same Frobenius norm as the no-SIM conventional MIMO channel, the power/beamforming gain produced by the optimizer is removed before BER and RPE are measured. The remaining differences are then caused only by incidental structural changes (which paths are emphasized, delay-Doppler reshaping), effects that the objective function did not target and that the paper does not analyze. The text does not state whether normalization is applied before or after SIM optimization, nor how the SNR axis (transmit vs. received SNR) is defined. Consequently, the large BER gains in Figs. 3-4 and RPE gains in Fig. 5 cannot be attributed to the proposed SIM optimization as claimed; they could be artifacts of the rescaling. This is the most load-bearing uncertainty for the central claim because it bears directly on the quantitative evidence for 'significant performance gains.'","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a metasurfaces-parametrized doubly-dispersive (MPDD) MIMO channel model that integrates stacked intelligent metasurfaces (SIM) at both the transmitter and receiver together with an arbitrary number of reconfigurable intelligent surfaces (RISs) in the environment. It derives discrete-time input-output relations and effective channel matrices for OFDM, OTFS, and AFDM, then formulates SIM phase-optimization problems for communication and sensing, and evaluates the resulting BER and radar-parameter-estimation performance using GaBP and PDA detectors. The central claim is that optimized SIM phases yield significant performance gains for all three waveforms in high-mobility and ISAC scenarios.","tokens_in":32931,"tokens_out":17427,"duration_ms":173365,"significance":"If the performance claims hold, this would provide a useful unified framework for modeling, optimizing, and detecting doubly-dispersive MIMO links enhanced by stacked intelligent metasurfaces. The structural derivation of the effective channel matrices for OFDM, OTFS, and AFDM within a common model, and the closed-form gradient expressions for SIM optimization, are valuable and go beyond a purely conceptual treatment. The paper also explicitly builds on prior published work on DD waveform modeling and SIM-MIMO, so the novelty is incremental but concrete. However, the quantitative evidence for the headline gains is currently weakened by the normalization procedure in Section 5.3 and by a scaling inconsistency in the RIS path of Eq. (22), both of which need to be resolved before the numerical claims can be fully interpreted.","major_comments":[{"comment":"The normalization statement \"the complete channels are normalized such that ||H_OFDM||_F^2 = ||H_OTFS||_F^2 = ||H_AFDM||_F^2 = ||H_MIMO||_F^2 for all the cases\" equalizes the total received power across the SIM and no-SIM cases. Since the SIM optimization in Eq. (43) maximizes exactly the total received power (the sum of per-path Frobenius norms), this normalization removes the power gain that the optimization is designed to produce. The BER gains in Figs. 3-4 would then reflect only secondary structural changes (which paths are emphasized, delay-Doppler reshaping) that the objective in Eq. (43) does not directly target and that the paper does not analyze. The manuscript must state whether the normalization is applied before or after the SIM optimization, define the SNR axis (transmit SNR vs. post-normalization receive SNR), and report results both with and without the cross-case normalization so that the power gain and any structural gain are separated. This is load-bearing because the central claim of significant performance gains rests on these figures. The RPE results in Fig. 5 should also state whether the same normalization is applied, since without this information the absolute gains cannot be interpreted.","section":"Section 5.3, Figs. 3-4"},{"comment":"The scaling factor for the RIS-reflected path is written as J / sqrt(Mtilde M Pbar Ptilde). From the definitions in Eqs. (19)-(20), the product Htilde_RX,k Phi_k Htilde_k,TX carries a factor J sqrt(Mtilde M / (Pbar Ptilde)) when the unit-norm UPA steering vectors are taken into account. As written, the RIS path is suppressed by a factor of Mtilde M in power relative to the direct path, which is inconsistent with the component models and would make the RIS contributions negligible for the typical M=100, J=100 parameters used in Section 4.5. Please correct the coefficient and verify its propagation into Eqs. (28), (34), (40), and (41). This error does not directly alter the numerical results in Sections 5-6 because those simulations set K=0, but it affects the correctness of the claimed general MPDD model.","section":"Eq. (22)"}],"minor_comments":[{"comment":"The statement that \"unoptimized SIM have no effect onto the DD channels\" is not clearly justified: with Z = I the transfer matrices Upsilon_T and Upsilon_R in Eqs. (11)-(12) still contain the non-identity diffraction matrices Gamma and Xi, so the channel is not identical to the no-SIM case used in [8]. Please clarify whether \"unoptimized\" means only that all phase shifts are unity, and discuss whether the comparison in Fig. 2 accounts for the diffraction-induced spatial transformation.","section":"Section 4.5"},{"comment":"The equation numbering in Section 4.4 reuses numbers (39), (40), and (41) that were already assigned to the OTFS equations in Section 4.3; please renumber the AFDM equations sequentially.","section":"Section 4.4"},{"comment":"The phrase \"no power advantage other than the effect of the parametrized SIM is provided\" is contradictory, because Eq. (43) is specifically a power-maximizing objective. Please rephrase to describe what is actually being normalized and how the SNR axis is defined.","section":"Section 5.3"},{"comment":"The loop \"for p=1 to P+1\" appears to be an off-by-one error; if the intent is to greedily optimize each of the P paths in turn, the loop should likely run to P. Please check and correct.","section":"Algorithm 2"},{"comment":"Fig. 5 does not include a no-SIM baseline, which makes it difficult to quantify the absolute gain in radar parameter estimation due to the SIM; adding such a curve (or explaining why it is omitted) would improve interpretability.","section":"Figure 5"},{"comment":"There are several typos and minor notation issues: \"chaper\" in the Introduction, \"exhbits\" in Section 6.4, \"wlg\" in Section 2.1, and the use of \"N_bar = M_bar = N d_s x N d_s\" in Section 5.2 where scalar dimensions are intended. These should be corrected in a final revision.","section":"General presentation"}],"recommendation":"major_revision","confidential_remarks":"The main technical issue to be resolved is the normalization in Section 5.3; if the authors can rerun the BER/RPE experiments without cross-case normalization and clarify the SNR definition, the paper's core modeling contribution is likely sound. The scaling error in Eq. (22) is also important for the unified model but does not affect the no-RIS simulations. The manuscript appears to be written as a book chapter and is somewhat terse in places, but the scope is appropriate for the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis is a paper to know about if you work on metasurface-assisted high-mobility links. It builds a metasurface-parametrized doubly-dispersive MIMO model with stacked intelligent metasurfaces at both ends plus an arbitrary number of RISs, derives effective OFDM, OTFS, and AFDM channel matrices in one framework, and adds SIM phase optimization for both communication and sensing. That combination is not in the cited literature, including the authors' own MIMO-SIM companion paper. The framework is plausible and well-structured; the derivations track the established SISO DD framework and the SIM model from An et al.\n\nWhat is genuinely new: the MPDD model itself, the RIS-plus-dual-SIM generalization, the unified effective channels for the three waveforms, and the two optimization flavors (total-power maximization for communication, weakest-path maximization for sensing). The GaBP and PDA detectors are standard but sensibly integrated. The paper is also honest about scope: it presents an illustrative application, not a hardware-validated model.\n\nThe soft spots, in proportion. First and most important: Section 5.3's normalization statement. The text says the complete channels are normalized so that the Frobenius norms of OFDM, OTFS, AFDM, and conventional MIMO are equal for all cases. But the SIM optimization objective in Eq. (43) maximizes exactly that total received power. If the optimized channels are rescaled to the no-SIM norm before BER and RPE are measured, the main claimed SIM power gain is removed. The remaining gains would have to come from incidental structural reshaping, which the paper does not analyze. The text does not say whether normalization happens before or after optimization, nor whether the SNR axis is transmit or receive SNR. This is the load-bearing uncertainty in the quantitative evidence. It is addressable, but right now the headline gains in Figs. 3-5 are not interpretable as SIM power gains.\n\nSecond, the gradient derivation in Section 5.1 has small consistency issues: Eq. (53) is labeled the TX-SIM gradient but is actually the RX-SIM gradient, and some intermediate definitions are easy to misread. These look like typo-level problems, not fatal flaws, but they should be cleaned up.\n\nThird, the SIM model assumes independently tunable phase layers and no inter-layer multiple reflections; the paper inherits that from prior work. Fine as a modeling assumption, but the simulations cannot validate it.\n\nWho benefits: researchers working on SIM, RIS, OTFS/AFDM, or high-mobility ISAC. The model and unified derivations are worth engaging even if the simulation evidence is currently compromised. I would send this to a serious referee, with the normalization question as the top issue to resolve. If the authors clarify that and rerun the comparisons fairly, this becomes a citeable reference.\n\nRecommendation: accept for peer review, conditional on fixing the normalization ambiguity.","headline":"Useful unified model for SIM-assisted doubly-dispersive MIMO, but the headline BER/RPE gains are compromised by an ambiguous normalization that may cancel the very power gain the SIM optimizer targets.","tokens_in":33491,"tokens_out":3019,"would_cite":true,"duration_ms":33975,"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":"The paper claims that a metasurface-parametrized doubly dispersive MIMO channel model, with stacked intelligent metasurfaces at both ends and arbitrary RISs in the environment, lets optimized SIM phases improve BER and radar estimation…","keywords":["doubly-dispersive channel","MIMO","stacked intelligent metasurface","reconfigurable intelligent surface","OTFS","AFDM","integrated sensing and communications","radar parameter estimation"],"falsifier":"Construct a single multi-layer SIM, set all layers to known phases, measure the through response, then change only one layer's phase vector and re-measure: the product model in Eq. (11) predicts the change is exactly the action of that diagonal phase matrix against fixed diffraction matrices, so any observed inter-layer coupling or phase-dependent diffraction beyond the sinc spatial correlation would show that the optimized phases of Algorithm 1 cannot produce the claimed BER and MSE gains. A companion simulation check is to run a full-wave solver with inter-layer reflections and compare the two-hop effective path, with summed delay and Doppler, against the exact cascaded response for large delay spreads.","tokens_in":32498,"feed_emoji":"📡","tokens_out":11094,"duration_ms":99756,"temperature":0.7,"pith_summary":"This chapter sets out to bring doubly-dispersive (delay-Doppler) channel modeling, which is standard for high-mobility and integrated-sensing scenarios, into the MIMO regime with artificial metasurfaces at both ends. It introduces a metasurface-parametrized doubly-dispersive MIMO channel model, named MPDD, that includes a stacked intelligent metasurface (SIM) at the transmitter, a SIM at the receiver, and an arbitrary number of reconfigurable intelligent surfaces (RISs) in the environment. The authors derive end-to-end input-output relations for OFDM, OTFS, and AFDM in this model, optimize the SIM phase layers with gradient ascent, and show that the optimized SIMs markedly lower bit-error rate and radar range/velocity estimation error relative to SIM-free or unoptimized systems. If correct, this gives one modeling, optimization, and detection framework for metasurface-enhanced high-mobility MIMO links and communication-centric integrated sensing and communications (ISAC).","feed_headline":"Metasurface phases shrink OFDM's gap to OTFS and sharpen radar","feed_subtitle":"One optimized SIM stack improves BER and radar estimation across OFDM, OTFS, and AFDM in high-mobility links.","key_machinery":"The load-bearing object is the metasurface-parametrized channel matrix $\\mathbf{H}(\\mathcal{Z},\\tilde{\\mathcal{Z}},\\mathcal{F},t,\\tau)$ of Eq. (15), assembled from the SIM transfer matrices $\\mathbf{U}_T(\\mathcal{Z}) = \\prod_{q=1}^{Q} \\boldsymbol{\\Psi}_{Q-q+1} \\boldsymbol{\\Gamma}_{Q-q+1}$ and the analogous $\\mathbf{U}_R(\\tilde{\\mathcal{Z}})$, where each $\\boldsymbol{\\Psi}_q$ is a diagonal layer of tunable phase shifts and each $\\boldsymbol{\\Gamma}_q$ is a fixed diffraction matrix from Rayleigh-Sommerfeld theory; sub-wavelength coupling of the outer SIM layers enters through sinc-correlation matrices $\\mathbf{R}_{TX}$ and $\\mathbf{R}_{RX}$. For RIS paths, Eq. (22) reduces every TX-SIM to RIS to RX-SIM route to a single effective path whose normalized delay and Doppler are the sums of the two hops, and the effective channels for OFDM, OTFS, and AFDM all take the shared form $\\bar{\\mathbf{H}} = \\sum_p \\check{\\mathbf{H}}_p \\otimes \\mathbf{G}_p$. This shared structure is what lets a single gradient-ascent optimizer, one Gaussian belief propagation (GaBP) detector, and one probabilistic data association (PDA) estimator serve all three waveforms.","core_discovery":"The central claim is that the doubly-dispersive MIMO channel can be written as a function of the phase configurations of the SIMs and RISs, $\\mathbf{H}(\\mathcal{Z},\\tilde{\\mathcal{Z}},\\mathcal{F},t,\\tau) = \\mathbf{U}_R(\\tilde{\\mathcal{Z}}) \\mathbf{R}_{RX}^{1/2} \\tilde{\\mathbf{H}}(\\mathcal{F},t,\\tau) \\mathbf{R}_{TX}^{1/2} \\mathbf{U}_T(\\mathcal{Z})$, where each SIM transfer matrix is a cascade of diagonal phase layers separated by fixed Rayleigh-Sommerfeld diffraction matrices, and the RIS-parametrized middle term is a direct path plus one-hop reflected paths. On top of this model, the paper derives effective channel matrices for OFDM, OTFS, and AFDM that share the same Kronecker structure, so the same optimization and detection machinery applies to all three. The optimization objective is the total received power across paths for communication, or the weakest path's power for sensing, both tuned by closed-form gradients. Simulation results claim that optimized SIMs substantially reduce BER compared with SIM-free operation and bring OFDM close to OTFS and AFDM, while sensing-optimized SIMs push range and velocity MSE toward the delay-Doppler grid resolution limit with only a modest communication penalty.","pith_inferences":["Beyond the paper, the layer-by-layer phase cascade could be trained end-to-end as a wave-domain linear transform, so part of delay-Doppler estimation happens inside the SIM before digital processing; the paper stops at power-based objectives and does not test this.","Beyond the paper, one could sweep the trade-off between the communication objective (total received power) and the sensing objective (weakest-path power) to trace the ISAC Pareto frontier explicitly; the paper reports two endpoints but not the curve.","Beyond the paper, because the model treats an arbitrary number of RISs, a natural extension is joint optimization of SIM phases and RIS reflection coefficients, which the gradients in Section 5.1 do not currently include.","Beyond the paper, the two-hop single-effective-path approximation in Eq. (22) should be stress-tested against a full wave simulation with multiple reflections; if it fails at large delay spreads, a rank-augmented version preserving the Kronecker form would still fit the framework."],"forward_implications":["With optimized SIM phases, OFDM's BER in doubly dispersive channels approaches that of OTFS and AFDM, so a conventional waveform plus wave-domain processing can substitute for more complex delay-Doppler schemes.","One effective-channel expression describes OFDM, OTFS, and AFDM, so channel estimation and detection can be designed once and instantiated per waveform.","Because the same model covers TX-SIM, RX-SIM, and any number of RISs, performance optimization can treat the whole metasurface-enhanced link as a single programmable system rather than separate components.","Switching the SIM objective from total received power to weakest-path power yields large radar range and velocity MSE gains toward the grid resolution limit, while the communication BER penalty remains relatively small, showing one hardware configuration can serve both ISAC functions.","All special cases, including SIM-only, RIS-only, conventional MIMO, and SISO doubly dispersive channels, collapse out of the same equations, giving a unified reference model for subsequent metasurface-aided work."],"supporting_citations":[{"why":"Supplies the unified doubly-dispersive channel framework that expresses OFDM, OTFS, and AFDM with a shared Kronecker structure, which the MPDD model generalizes to MIMO with SIMs and RISs.","marker":"[8]"},{"why":"Provides the stacked-intelligent-metasurface transfer model with Rayleigh-Sommerfeld diffraction matrices and sinc spatial correlation used in Eqs. (10)-(12).","marker":"[61]"},{"why":"Supplies the joint channel, data, and radar parameter estimation procedure for AFDM that the sensing section adapts to SIM-optimized bistatic radar.","marker":"[25]"},{"why":"Establishes the preceding doubly-dispersive MIMO channel model with stacked intelligent metasurfaces that this chapter extends with environment RISs.","marker":"[65]"},{"why":"Provides the gradient-ascent phase update routine that the SIM optimization in Algorithm 1 builds on.","marker":"[85]"},{"why":"Defines the AFDM waveform and its chirp parameter design, which Section 4.4 relies on for the AFDM effective channel derivation.","marker":"[34]"},{"why":"Supplies the cyclic-prefix and vectorized OTFS signaling that Section 4.3 uses to obtain the OTFS effective channel in the same form as OFDM and AFDM.","marker":"[74]"},{"why":"Supplies convergence conditions for Gaussian belief propagation that justify the damped GaBP detector used in the BER simulations.","marker":"[87]"}],"fun_headline_variants":["SIM-tuned MIMO closes OFDM-OTFS gap in doubly-dispersive links","Metasurface phases narrow BER gap, boost radar in fast-fading MIMO","Optimized SIM stacks lift OFDM to OTFS class and sharpen sensing","Phase-tuned SIMs improve OFDM, OTFS, and AFDM in high-mobility channels"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that each SIM layer acts as an independent diagonal phase screen with fixed diffraction between layers and no multiple reflections or mutual coupling, and that each TX-SIM-to-RIS-to-RX-SIM route collapses to a single path with summed delay and Doppler; if a real SIM couples layers or its diffraction changes with phase, the optimized phases will not deliver the predicted gains.","fun_headline_variants_meta":{"raw":{"variants":["SIM-tuned MIMO closes OFDM-OTFS gap in doubly-dispersive links","Metasurface phases narrow BER gap, boost radar in fast-fading MIMO","Optimized SIM stacks lift OFDM to OTFS class and sharpen sensing","Phase-tuned SIMs improve OFDM, OTFS, and AFDM in high-mobility channels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000972,"raw_usage":{"total_tokens":4169,"prompt_tokens":1021,"completion_tokens":3148,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":637,"completion_tokens_details":{"reasoning_tokens":3057}},"tokens_in":637,"tokens_out":3148,"duration_ms":19952,"temperature":1.0,"reasoning_tokens":3057,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:47:35.021393+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a single multi-layer SIM, set all layers to known phases, measure the through response, then change only one layer's phase vector and re-measure: the product model in Eq. (11) predicts the change is exactly the action of that diagonal phase matrix against fixed diffraction matrices, so any observed inter-layer coupling or phase-dependent diffraction beyond the sinc spatial correlation would show that the optimized phases of Algorithm 1 cannot produce the claimed BER and MSE gains. A companion simulation check is to run a full-wave solver with inter-layer reflections and compare the two-hop effective path, with summed delay and Doppler, against the exact cascaded response for large delay spreads.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the unified doubly-dispersive channel framework that expresses OFDM, OTFS, and AFDM with a shared Kronecker structure, which the MPDD model generalizes to MIMO with SIMs and RISs."},{"cited_title":"Stacked Intelligent Metasurfaces for Efficient Holographic MIMO Communications in 6G,","cited_arxiv_id":null,"evidence_quote":"Provides the stacked-intelligent-metasurface transfer model with Rayleigh-Sommerfeld diffraction matrices and sinc spatial correlation used in Eqs. (10)-(12)."},{"cited_title":"Stacked Intelligent Metasurface performs a 2D DFT in the Wave Domain for DOA Estimation,","cited_arxiv_id":null,"evidence_quote":"Provides the gradient-ascent phase update routine that the SIM optimization in Algorithm 1 builds on."},{"cited_title":"Affine Frequency Division Multiplexing for Next Generation Wireless Communications,","cited_arxiv_id":null,"evidence_quote":"Defines the AFDM waveform and its chirp parameter design, which Section 4.4 relies on for the AFDM effective channel derivation."},{"cited_title":"Interference Cancellation and Iterative De- tection for Orthogonal Time Frequency Space Modulation,","cited_arxiv_id":null,"evidence_quote":"Supplies the cyclic-prefix and vectorized OTFS signaling that Section 4.3 uses to obtain the OTFS effective channel in the same form as OFDM and AFDM."},{"cited_title":"On Convergence Conditions of Gaussian Belief Propagation,","cited_arxiv_id":null,"evidence_quote":"Supplies convergence conditions for Gaussian belief propagation that justify the damped GaBP detector used in the BER simulations."}],"review_version":2}