{"id":"abfe725a-d6f2-4b6f-86d4-1542fe2c8ba4","arxiv_id":"2506.07266","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Hardware impairments modeled as random multiplicative errors in the BD-RIS scattering matrix degrade channel estimation NMSE, with off-diagonal (mutual impedance) errors being the most harmful and able to make conventional RIS outperform BD-RIS at high SNR.","lead":"This paper models three kinds of hardware errors in beyond-diagonal reconfigurable intelligent surfaces and simulates how they hurt channel estimation. It shows that when errors strike the inter-element connections, a simpler conventional RIS can beat the more advanced BD-RIS at high signal-to-noise ratios.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Impairment model in Eq. (6) violates reciprocity and passivity of scattering matrices; the claimed conventional-RIS advantage may be an artifact of non-physical surfaces.","rationale":"The reader's weakest assumption was that the impairment model never checks unitarity or passivity, so the impaired surfaces may not be physically realizable. I agree that this is the most load-bearing issue, but I would sharpen it: the model also violates reciprocity because E is enforced Hermitian (E = E^H) while a reciprocal scattering matrix requires complex symmetry (S = S^T). The product S_t ⊙ E is then generally not symmetric, so even the paper's own 'physical reciprocity of energy' claim is internally inconsistent. This matters because the entire Type-1 comparison is constructed so that a diagonal RIS has no off-diagonal elements and is therefore immune, while the BD-RIS is impaired. Under a physical impedance-perturbation model, mutual-impedance errors propagate through the impedance-to-scattering transformation to all entries of S̄, so the diagonal entries of a BD-RIS would also be affected, and passivity would bound the singular values. The paper does not provide a physical derivation of E or any constraint-satisfaction check, so the central claim is not established for real hardware. This reinforces the reader's CONDITIONAL verdict: the paper is internally consistent as a stylized simulation, but its physical applicability must be demonstrated before the conclusion can be accepted. I recommend no change to the verdict because the concern, while serious, does not invalidate the mathematical trace; it adds a specific validation condition that the reader already hinted at.","tokens_in":7902,"tokens_out":10867,"duration_ms":136718,"concrete_test":"Reproduce the Fig. 3 Type-1 setup (N=32, N̄=32, 20% affected, K=100) and compute for each trial (i) the normalized reciprocity residual ||S̄_t - S̄_t^T||_F/||S̄_t||_F and (ii) the maximum singular value of each block S̄_t^{(q)}. If many trials show residual > 0.1 or σ_max > 1+10^-6, re-run the NMSE comparison using a physical impairment model: perturb only off-diagonal entries of the impedance matrix Z (keeping Z symmetric and diagonal fixed), compute S̄ = (Z - Z0 I)(Z + Z0 I)^{-1}, and test whether the high-SNR NMSE ordering of N̄=1 vs N̄=32 in Fig. 3 is preserved. If the ordering reverses, the paper's headline conclusion is an artifact of the unphysical S_t⊙E model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that a conventional diagonal RIS can outperform a group-connected BD-RIS under Type-I hardware impairments—rests entirely on the elementwise multiplicative model in Eq. (6), S̄_t = S_t ⊙ E. For a physical BD-RIS, the true scattering matrix must be reciprocal (complex symmetric, S = S^T) and passive (all singular values ≤ 1, and unitary if lossless). The impairment model in Section IV does not satisfy either constraint. The paper states that 'to maintain a physical reciprocity of energy' it enforces Hermitian symmetry E = E^H; however, reciprocity for scattering matrices is transposition, not conjugate transposition. With the ideal S symmetric and E Hermitian, S̄_ij = S_ij E_ij while S̄_ji = S_ij E_ij^*, so S̄_t is generally non-symmetric for complex phase distortions. Hence the simulated impaired BD-RIS is not reciprocal, contradicting the paper's own physical motivation. Furthermore, no passivity or unitarity check is performed: e.g., a 2×2 block of Type-I impairment with diagonal 1 and |E_12| = 0.9 has maximum singular value 1.9, exceeding the passive limit. Consequently, the NMSE saturation and the superiority of the diagonal RIS may be artifacts of an active, non-reciprocal surrogate of the hardware. The paper also never derives E from an impedance model, so the relative ranking of architectures under physically admissible faults is unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes three hardware impairment models for beyond diagonal reconfigurable intelligent surfaces (BD-RIS), all based on an elementwise multiplicative distortion of the scattering matrix, S̄_t = S_t ⊙ E (Eq. 6). The considered impairments affect off-diagonal entries (Type 1), diagonal entries (Type 2), or all entries (Type 3) of the scattering matrix. Using the least-squares channel estimator of [9] for a group-connected BD-RIS, the paper evaluates the normalized mean square error (NMSE) via Monte Carlo simulations. The results show that impairments affecting off-diagonal elements (Types 1 and 3) cause NMSE saturation at high SNR, while Type 2 is less harmful, and a conventional diagonal RIS (N̄=1) is unaffected by Type 1. The paper concludes that there exist scenarios where the traditional RIS can outperform the BD-RIS.","tokens_in":8195,"tokens_out":8020,"duration_ms":79868,"significance":"If the elementwise multiplicative impairment model were a physically valid abstraction of hardware faults, the paper would offer a useful, reproducible simulation-based comparison of BD-RIS and conventional RIS under imperfections, and it would be among the first to study hardware impairments specifically for BD-RIS. The Monte Carlo methodology is sound and the internal mathematics is consistent with the stated model. However, the physical relevance of the model is not established: the paper incorrectly equates Hermitian symmetry with reciprocity for scattering matrices, and it never checks whether the impaired scattering matrix satisfies passivity/unitarity constraints. Moreover, the headline 'traditional RIS can outperform BD-RIS' is largely a definitional consequence of confining Type 1 impairments to off-diagonal entries. The significance is therefore conditional on a physically defensible impairment model.","major_comments":[{"comment":"The paper states that 'to maintain a physical reciprocity of energy' it enforces the Hermitian symmetry E^(q) = (E^(q))^H. For scattering matrices, reciprocity requires complex symmetry, S = S^T (equivalently E_ij = E_ji), not Hermitian symmetry. As defined in Type 1, off-diagonal entries have E_ij = α_ij e^{jφ_ij} and E_ji = α_ij e^{-jφ_ij}; for a symmetric ideal S_t, the impaired matrix S̄_t then has S̄_ij ≠ S̄_ji in general. Thus the simulated BD-RIS is not a reciprocal network, contradicting the stated physical motivation. The authors should either enforce E = E^T (with identical phases on symmetric positions) or explicitly state that the model describes non-reciprocal impairments and justify that scenario physically.","section":"Section IV, paragraphs before Eq. (6) and in subsections A-C"},{"comment":"The elementwise multiplicative model S̄_t = S_t ⊙ E is assumed without derivation from an impedance or circuit model. For a lossless passive BD-RIS, the scattering matrix must be unitary (and reciprocal if passive). The paper never checks whether S̄_t remains passive or unitary under the generated E. Even with α_ij ∈ (0,1], the phase perturbations can break the unitarity of S_t, and singular values of S̄_t can exceed 1, implying an active surface. This could contaminate the NMSE saturation seen in Figures 3, 5, and 6. The authors should either prove that their impairment model preserves passivity (e.g., by projecting S̄_t onto the set of unitary matrices or by deriving E from physically constrained impedance perturbations) or report the distribution of the largest singular value of S̄_t to quantify the fraction of non-passive realizations.","section":"Section IV, Eq. (6)"},{"comment":"The result that the conventional diagonal RIS (N̄=1) is immune to Type 1 impairment is a tautology of the model: a diagonal scattering matrix has no off-diagonal elements, so the construction of Type 1 cannot affect it. The paper acknowledges this in the text, but the abstract and conclusion present it as a scenario where 'the traditional RIS can outperform the BD-RIS.' This overstates the significance. To make a non-trivial comparison, the authors should apply an equivalent impairment model to the conventional RIS as well (e.g., diagonal amplitude/phase errors arising from varactor nonidealities) and compare the two architectures under matched physical fault conditions.","section":"Section V, Figure 3 and Abstract/Conclusion"}],"minor_comments":[{"comment":"The expression y_t = ∑_q G^(q) (S_t^(q) ⊙ E) H^(q)T x_t is imprecise because E is defined as block diagonal with blocks E^(q); it should read S_t^(q) ⊙ E^(q). Also, Eq. (8) reuses S̄_t without the group superscript, which is confusing for a block-diagonal structure.","section":"Section III, Eq. (7)"},{"comment":"The NMSE definition uses c^(k) and ĉ^(k), but it is not explicitly stated that these are the true and estimated combined channel vectors for the k-th Monte Carlo trial; please clarify the notation.","section":"Section V, NMSE definition"},{"comment":"The statement 'the maximum number of affected impedances is N(N̄−1)/2 = (N̄^2 Q − N)/2' is correct, yet it may be clearer to note explicitly that for N̄=1 this number is zero, which is exactly why the conventional RIS is unaffected by Type 1.","section":"Section IV, Type 1"},{"comment":"The caption does not specify the simulation parameters (e.g., 20% affected impedances, N=32, M_T=2, M_R=4) that are described only in the text; adding these details to the caption would improve readability.","section":"Figure 3 caption"},{"comment":"The paper does not cite prior work on BD-RIS with hardware impairments, mutual coupling, or impedance mismatch models. If such work exists, it should be referenced to properly position the contribution.","section":"Introduction and Related Work"}],"recommendation":"major_revision","confidential_remarks":"The paper is a compact conference contribution with a clear internal logic, but the physical validity of the impairment model is the central concern. The reciprocity error (Hermitian vs. transpose symmetry) is a concrete technical mistake that must be corrected. If the correction leads to a complex-symmetric E, the simulation results may change because the phase statistics differ. Even if the qualitative trends survive, the paper currently lacks a passivity analysis, and the 'conventional RIS can outperform BD-RIS' claim needs to be reframed as a definitional consequence. These issues are fixable, so I recommend major revision rather than rejection. The authors should be encouraged to compare against physically constrained impairment models (e.g., unitary projections or impedance-derived perturbations) to strengthen the relevance of their conclusions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a tidy conference paper that asks the right question—how do hardware imperfections affect BD-RIS channel estimation—but its central cautionary result is built on an impairment model that doesn't respect the physics of scattering matrices. The claim that a conventional diagonal RIS can beat a group-connected BD-RIS when mutual impedances are faulty is currently an artifact of the model, not a property of real hardware.\n\nWhat the paper does well: the writing is clear, the extension of the decoupled estimation scheme from [12] is straightforward, and the three impairment masks (off-diagonal, diagonal, all entries) are easy to follow. The Monte Carlo results are internally consistent: Type 2 affects fewer entries and does less damage, while Type 1 and Type 3 saturate NMSE at high SNR. As a pure simulation exercise under a stated elementwise-error model, the math checks out.\n\nThe soft spot is load-bearing. In Eq. (6) the observed scattering matrix is S̄_t = S_t ⊙ E, and the authors justify Hermitian E as preserving 'physical reciprocity of energy.' But reciprocity for a scattering matrix is transpose symmetry, S = S^T, not conjugate-transpose. If the ideal S is symmetric and E is Hermitian, then S̄_ij = S_ij E_ij while S̄_ji = S_ij E_ij^*, so S̄ is non-reciprocal for any complex phase distortion. The paper never checks unitarity or passivity either; a 2x2 block with diagonal ones and |E_12|=0.9 gives a singular value of 1.9, which would require an active surface. The simulated 'impaired BD-RIS' may therefore be an active, non-reciprocal surrogate rather than a passive lossy one. If real faults are constrained to reciprocal passive surfaces, the NMSE ordering could easily change. The paper also leaves impairment distributions unspecified (only α_ij ∈ (0,1], ϕ_ij ∈ [0,2π)) and shows no error bars for K=100 runs.\n\nSo: the idea is worth exploring, and the paper is honest in scope, but the physical validity of the model needs to be addressed before the headline comparison can be taken seriously. I'd recommend major revision: either rework E so that S̄ obeys reciprocity and passivity (e.g., perturb the impedance matrix Z and re-derive S), or at minimum add a discussion of why the unconstrained elementwise model is a valid proxy. As it stands, I wouldn't cite the main result.\n\nFor referees: it deserves a serious review—the topic is timely and the framework is reusable—but I'd want the model fixed before acceptance.","headline":"Useful first look at BD-RIS under impairments, but the impairment model violates reciprocity and passivity, so the headline ordering result is not yet trustworthy.","tokens_in":8722,"tokens_out":2424,"would_cite":false,"duration_ms":27099,"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":"Hardware errors in the inter-element couplings of a BD-RIS create an error floor in channel estimation that a conventional diagonal RIS does not share.","keywords":["beyond diagonal RIS","hardware impairments","channel estimation","scattering matrix","normalized mean square error","group-connected architecture","mutual impedance","impedance mismatch"],"falsifier":"Build or simulate a physically constrained impairment: perturb only mutual impedances while enforcing passivity and, for a lossless surface, unitarity of the scattering matrix, then run the same matched-filter channel estimation. If the NMSE error floor at high SNR disappears or the crossover with the conventional RIS shifts, the elementwise-Hermitian impairment model overstates the Type 1 degradation.","tokens_in":7687,"feed_emoji":"📶","tokens_out":5999,"duration_ms":62014,"temperature":0.7,"pith_summary":"Beyond diagonal reconfigurable intelligent surfaces (BD-RIS) connect reflecting elements to one another, giving a non-diagonal scattering matrix instead of the diagonal one of a conventional RIS. This paper asks whether those extra interconnections survive real hardware imperfections such as impedance mismatches and varactor defects. It models three kinds of scattering-matrix errors by multiplying the ideal BD-RIS response elementwise by a random error matrix: errors in the off-diagonal mutual impedances, errors in the diagonal self-impedances, and both together. Simulating channel estimation with a matched-filter estimator, the paper finds that mutual-impedance errors make the estimation NMSE saturate at high signal-to-noise ratio, and that a conventional diagonal RIS, having no off-diagonal elements, can outperform the BD-RIS in that regime. If true, the practical benefit of BD-RIS over traditional RIS is conditional on keeping inter-element coupling hardware accurate.","feed_headline":"Mutual-impedance errors make BD-RIS lose to a plain RIS","feed_subtitle":"With impaired inter-element couplings, channel-estimation error floors and the diagonal RIS wins.","key_machinery":"The central object is the impaired scattering matrix $\\bar{S}_t = S_t \\odot E$, built from a block-diagonal, group-connected BD-RIS scattering matrix $S_t$ and a Hermitian impairment matrix $E$ whose entries are random amplitude-and-phase factors $\\alpha_{ij}e^{j\\phi_{ij}}$. The paper inserts $\\bar{S}_t$ into the matched-filter channel estimator $\\hat{c} = \\frac{\\bar{N}}{T}(\\Omega \\otimes I_{M_R})^H y$, where $\\Omega = (S \\diamond X)^T$ is the ideal training design, so the impairment appears as a mismatch between the assumed and actual scattering matrices. That mismatch, not the noise, drives the NMSE error floor observed at high SNR.","core_discovery":"The paper's central claim is that the performance advantage of BD-RIS over a conventional diagonal RIS shrinks and can reverse when hardware impairments disturb the off-diagonal entries of the scattering matrix. Under the elementwise-multiplicative model $ar{S}_t = S_t \\odot E$, where $S_t$ is the ideal group-connected BD-RIS response and $E$ is a Hermitian random error matrix, the paper shows that Type 1 impairment (only mutual impedances, i.e., only off-diagonal entries) and Type 3 impairment (all entries) cause the channel-estimation NMSE to saturate as the SNR grows, because the number of disturbed scattering-matrix elements increases with the BD-RIS group size. Type 2 impairment (only self-impedances, i.e., only diagonal entries) is the mildest, and its effect lessens as the group size grows. The conventional diagonal RIS, corresponding to group size $\\bar{N}=1$, is immune to Type 1 because its scattering matrix has no off-diagonal elements to be impaired, and the simulations show it can therefore beat the BD-RIS at high SNR.","pith_inferences":["If real mutual-impedance errors are accompanied by even small diagonal shifts of the self-impedances, the conventional RIS's immunity to Type 1 would be diluted; that combined case is close to the paper's Type 3 and is worth testing.","Because the impairment matrix $E$ is assumed constant over all $T$ pilot slots, a natural extension is to estimate $E$ jointly with the channel by reserving a few pilot slots, which the paper lists as future work.","Physically constrained impairments that preserve passivity (unitarity for lossless surfaces) would couple diagonal and off-diagonal corrections, so the three impairment types cannot occur independently in a real lossless BD-RIS; this may change the predicted crossover.","The interaction between impairment and the pilot-overhead scaling $T = M_T N \\bar{N}$ suggests that choosing the group size could trade estimation accuracy against robustness, a design knob the paper does not explicitly turn."],"forward_implications":["Under the off-diagonal (Type 1) impairment, the channel-estimation NMSE saturates as SNR increases, while the conventional diagonal RIS keeps improving because it has no off-diagonal elements to impair.","The self-impedance (Type 2) impairment is the mildest, and its effect shrinks as the BD-RIS group size grows, because only the $N$ diagonal entries of the scattering matrix are disturbed.","Affecting both self- and mutual impedances (Type 3) is slightly worse than Type 1, since it disturbs the largest number of scattering-matrix entries.","At low SNR the impairment is hidden by noise and BD-RIS retains its advantage; the crossover to the conventional RIS appears as the SNR increases.","Even a modest impairment level of 20% of the impedances creates a noticeable gap from the ideal case, so practical BD-RIS designs should include impairment estimation or element selection."],"supporting_citations":[{"why":"Introduces the beyond-diagonal RIS concept of a non-diagonal scattering matrix, the object the paper then impairs.","marker":"[5]"},{"why":"Provides the scattering-parameter network analysis that grounds the BD-RIS scattering matrix in physical impedances, motivating the self- and mutual-impedance impairment models.","marker":"[6]"},{"why":"Supplies the orthogonal BD-RIS training matrix design that the paper takes as the ideal scattering matrix and then distorts with impairments.","marker":"[9]"},{"why":"Supplies the system model and matched-filter channel estimator whose NMSE the paper evaluates under the three impairment models.","marker":"[12]"}],"fun_headline_variants":["Mutual-impedance errors flip BD-RIS vs RIS advantage","BD-RIS loses its edge when off-diagonal terms are flawed","Hardware impairments make BD-RIS worse than diagonal RIS","Coupling errors cause BD-RIS to trail plain RIS at high SNR"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The simulations assume that hardware errors can be any random amplitude-and-phase distortion multiplied onto the ideal surface response, with only a mirror-symmetry constraint, and never check that the resulting surface is physically realizable (a real passive surface obeys stricter limits).","fun_headline_variants_meta":{"raw":{"variants":["Mutual-impedance errors flip BD-RIS vs RIS advantage","BD-RIS loses its edge when off-diagonal terms are flawed","Hardware impairments make BD-RIS worse than diagonal RIS","Coupling errors cause BD-RIS to trail plain RIS at high SNR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000968,"raw_usage":{"total_tokens":4094,"prompt_tokens":896,"completion_tokens":3198,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":512,"completion_tokens_details":{"reasoning_tokens":3124}},"tokens_in":512,"tokens_out":3198,"duration_ms":23038,"temperature":1.0,"reasoning_tokens":3124,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:37:43.594933+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build or simulate a physically constrained impairment: perturb only mutual impedances while enforcing passivity and, for a lossless surface, unitarity of the scattering matrix, then run the same matched-filter channel estimation. If the NMSE error floor at high SNR disappears or the crossover with the conventional RIS shifts, the elementwise-Hermitian impairment model overstates the Type 1 degradation.","supporting_citations":[{"cited_title":"Reconfigurable intelligent surfaces 2.0: Beyond diagonal phase shift matrices,","cited_arxiv_id":null,"evidence_quote":"Introduces the beyond-diagonal RIS concept of a non-diagonal scattering matrix, the object the paper then impairs."},{"cited_title":"Modeling and architecture design of reconfigurable intelligent surfaces using scattering parameter network analysis,","cited_arxiv_id":null,"evidence_quote":"Provides the scattering-parameter network analysis that grounds the BD-RIS scattering matrix in physical impedances, motivating the self- and mutual-impedance impairment models."},{"cited_title":"Channel estimation for beyond diagonal reconfigurable intelligent surfaces with group-connected archi- tectures,","cited_arxiv_id":null,"evidence_quote":"Supplies the orthogonal BD-RIS training matrix design that the paper takes as the ideal scattering matrix and then distorts with impairments."},{"cited_title":"A decoupled channel estimation method for beyond diagonal ris,","cited_arxiv_id":null,"evidence_quote":"Supplies the system model and matched-filter channel estimator whose NMSE the paper evaluates under the three impairment models."}],"review_version":1}