Pith. sign in

REVIEW 4 major objections 5 minor 15 references

Beyond Diagonal RIS Design for Parameter Estimation With and Without Eavesdropping

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A beyond-diagonal RIS whose response is an arbitrary unitary matrix can substantially improve secure parameter estimation, with the no-eavesdropper optimum given in closed form by an eigenvector alignment.

desk verdict A correct and clean extension of secure parameter estimation to BD-RIS, whose numerical claims of significant gain rest on a single realization and a heuristic diagonal baseline. read the letter →

arxiv 2505.05971 v1 pith:F3GNWSR3 submitted 2025-05-09 eess.SP cs.ITmath.IT

Özlem Tuğfe Demir , Sinan Gezici This is my paper · ORCID
classification eess.SPcs.ITmath.IT
keywords beyond-diagonalRISreconfigurableintelligentsurfaceparameterestimationFisherinformationCramér–Raolowerboundphysicallayersecurityeavesdroppingunitaryresponsematrix
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether a beyond-diagonal reconfigurable intelligent surface (BD-RIS), whose response is a full unitary matrix rather than diagonal phase shifts, can improve parameter estimation when the transmitter-to-receiver and transmitter-to-eavesdropper direct links are blocked. The paper's answer is yes: it derives the optimal unitary response in closed form for the no-eavesdropper case, gives algorithms for the symmetric (reciprocal) and eavesdropper-constrained cases, and shows numerically that BD-RIS substantially outperforms conventional diagonal RIS, especially when the eavesdropper's allowed Fisher information is small. If the result holds, BD-RIS gives wireless systems a concrete way to shape the propagation environment for both estimation accuracy and information-leakage control.

What carries the argument

The central object is the $r\times r$ BD-RIS response matrix $\boldsymbol{\Omega}$, constrained to be unitary ($\boldsymbol{\Omega}^H\boldsymbol{\Omega}=\mathbf{I}_r$) in the general non-reciprocal case and additionally symmetric ($\boldsymbol{\Omega}=\boldsymbol{\Omega}^T$) in the reciprocal case. The objective is the trace of the Fisher information matrix, written as $\operatorname{tr}(\boldsymbol{\Omega}^H\mathbf{E}_b\boldsymbol{\Omega}\mathbf{M})$ with $\mathbf{E}_b=\mathbf{H}_{rb}^H\boldsymbol{\Sigma}_b^{-1}\mathbf{H}_{rb}$ and $\mathbf{M}=\mathbf{H}_{ar}\mathbf{P}\mathbf{P}^H\mathbf{H}_{ar}^H$. The argument is carried by von Neumann's trace inequality, which upper-bounds this trace by the sum of descending-ordered eigenvalue products and is achieved by aligning the eigenbases of $\mathbf{E}_b$ and $\mathbf{M}$. For the eavesdropping case, the constraint $\operatorname{tr}(\boldsymbol{\Omega}^H\mathbf{E}_e\boldsymbol{\Omega}\mathbf{M})\le\epsilon$ is handled by vectorization, which turns the inner update into a quadratically constrained least-squares problem whose KKT solution is a bisection search over the Lagrange multiplier $\mu$.

What would settle it

Repeat the numerical experiments over many independent random channel realizations, for example 1000 draws at $r=36,\,k=10$, and compare the average and median trace Fisher information of the BD-RIS designs against the paper's diagonal-RIS baseline at the same eavesdropping threshold; if BD-RIS does not consistently achieve higher average Fisher information than the diagonal baseline, the central numerical claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that replacing the diagonal phase-shift matrix of a conventional RIS with a fully connected beyond-diagonal RIS, whose response matrix $\boldsymbol{\Omega}\in\mathbb{C}^{r\times r}$ is an arbitrary unitary matrix (or a symmetric unitary matrix in the reciprocal case), can substantially improve estimation quality at the intended receiver while limiting the information leaked to an eavesdropper. In the no-eavesdropper non-reciprocal case, the paper shows that the trace of the Fisher information matrix, $\operatorname{tr}(\boldsymbol{\Omega}^H \mathbf{E}_b \boldsymbol{\Omega} \mathbf{M})$, is maximized by $\boldsymbol{\Omega}=\mathbf{V}_E\mathbf{V}_M^H$, where $\mathbf{V}_E$ and $\mathbf{V}_M$ hold the eigenvectors of $\mathbf{E}_b$ and $\mathbf{M}$ sorted by descending eigenvalue, and the resulting maximum is the sum of the ordered eigenvalue products. For the reciprocal case the paper supplies a manifold alternating-optimization algorithm initialized at the closest symmetric unitary matrix to that unconstrained optimum, and for the secure case it supplies a penalty-dual-decomposition algorithm with a semi-closed-form update obtained by diagonalizing $\mathbf{M}^T\otimes\mathbf{E}_e$. The numerical section reports that BD-RIS achieves substantially higher trace Fisher information, equivalently a lower Cramér–Rao bound and lower MSE of the maximum-likelihood estimator, than the diagonal-RIS baseline, with the reciprocal design nearly matching the non-reciprocal one.

Load-bearing premise

The load-bearing premise is that a fully connected BD-RIS can realize any $r\times r$ unitary (orthonormal-column) response matrix, symmetric in the reciprocal case; if practical hardware cannot realize such matrices, or if the single simulated channel realization is unrepresentative, the derived limits and the large reported gains over diagonal RIS may not hold.

Editorial extensions

If this is right

  • Without an eavesdropper and with a non-reciprocal BD-RIS, the maximum average Fisher information is exactly $\sum_{i=1}^r \delta_{E,i}\delta_{M,i}$, so any suboptimal design can be measured against a closed-form ceiling.
  • Because the model is linear and Gaussian, the Cramér–Rao bound is achievable by the maximum-likelihood estimator, so the reported trace-Fisher-information gains translate directly into mean-squared-error reductions.
  • The negligible gap between reciprocal and non-reciprocal designs means reciprocal hardware, which is symmetric and easier to build, can capture most of the BD-RIS benefit.
  • Under tight eavesdropping limits, conventional diagonal RIS may become infeasible, whereas BD-RIS keeps the problem feasible and maintains high estimation quality at the intended receiver.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the numerical evidence uses a single channel realization, the size of the gain over diagonal RIS should be checked across many random realizations; the closed-form result suggests the gain depends on how mismatched the eigenbases of $\mathbf{E}_b$ and $\mathbf{M}$ are.
  • The eigen-alignment view implies that when $\mathbf{E}_b$ and $\mathbf{M}$ are nearly diagonal or share eigenvectors, the BD-RIS advantage over a diagonal RIS should shrink, which is a testable prediction the paper does not make.
  • The eavesdropper-constrained problem could be extended to compute the full trade-off frontier between Bob's trace Fisher information and the eavesdropper threshold $\epsilon$; the semi-closed-form structure suggests the frontier is piecewise-parametric in $\mu$.
  • A natural follow-up is to jointly optimize the power allocation $\mathbf{P}$ with the BD-RIS response, since the objective is linear in $\mathbf{M}$ and the closed form for $\boldsymbol{\Omega}$ may simplify that joint optimization.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies the design of a beyond-diagonal reconfigurable intelligent surface (BD-RIS) for parameter estimation in a system where direct links between the transmitter (Alice) and both the intended receiver (Bob) and an eavesdropper (Eve) are blocked. The performance metric is the trace of the Fisher information matrix (FIM) at Bob, optionally subject to a limit on Eve's average Fisher information. For the non-reciprocal BD-RIS without an eavesdropper, the authors prove Lemma 1, providing the closed-form optimal response Ω = V_E V_M^H and the maximum trace equal to the sum of sorted eigenvalue products. For the reciprocal (symmetric-unitary) case, they propose a manifold alternating-optimization algorithm (Algorithm 1). For secure estimation, they formulate a constrained problem and solve it via penalty dual decomposition (PDD). Numerical results compare the proposed BD-RIS designs with a conventional diagonal RIS and report substantial improvements in Fisher information and Cramér–Rao bound.

Significance. If the results are solid, the paper provides a valuable extension of RIS-assisted parameter estimation to the increasingly important BD-RIS hardware model. The main theoretical contribution is Lemma 1, which is cleanly stated and correctly proved via von Neumann's trace inequality; it gives an ultimate bound for the non-reciprocal no-eavesdropper case. The algorithms for the reciprocal and secure cases follow established techniques (manifold optimization and PDD) and appear plausible, though no convergence proofs are given. The central weakness is the numerical validation: the claimed significant gain over diagonal RIS rests on a single channel realization and a heuristic diagonal baseline, so the magnitude of the advantage is not established robustly. The modeling assumption that arbitrary unitary response matrices are realizable is also load-bearing and not explicitly discussed.

major comments (4)
  1. [Section V, first paragraph] The numerical comparison is based on a single channel realization in MATLAB, which is insufficient to support the paper's central claim that BD-RIS 'notably enhances' estimation performance. Since the channel matrices are random, the reported gaps may be specific to that draw. Please provide Monte Carlo averages over many channel realizations, with confidence intervals or box plots, and ensure the conclusions hold statistically.
  2. [Section V, conventional RIS baseline] The diagonal-RIS baseline is obtained by SDP followed by Gaussian randomization, and in the eavesdropping case by relaxing the unit-modulus constraint and then clipping entries. Because the diagonal RIS is the reference point for the claimed improvement, the baseline's suboptimality could artificially inflate the BD-RIS gain. Please validate the baseline, e.g., for small r compare against a global optimization method, report the number of randomizations used, and verify that the clipped solutions still satisfy the Eve information constraint.
  3. [Section III, constraints (7b) and (11b)] The assumption that a fully connected BD-RIS can realize any unitary (or symmetric-unitary) response matrix is load-bearing for the interpretation of the results as 'ultimate limits'. If practical devices restrict the achievable response matrices, the derived optima are upper bounds rather than achievable gains. Please add a discussion of the hardware feasibility of arbitrary unitary responses, citing the relevant BD-RIS literature, or explicitly state that the results are valid under the ideal unitary model.
  4. [Section III, Algorithm 1] The alternating-optimization algorithm for the reciprocal case lacks a convergence proof or even a monotonicity argument. The paper cites similar manifold methods in [13] and [10], but the specific update with adaptive step size is not shown to converge to a stationary point. This affects the credibility of the reciprocal-BD-RIS curves in Figs. 1–4. Please state the convergence properties or characterize the algorithm's behavior (e.g., monotone increase of the cost function).
minor comments (5)
  1. [Eq. (10)] The phrase 'in ∆ and Σ' should read 'in Δ_E and Δ_M' to refer to the eigenvalue matrices of E_b and M, respectively.
  2. [Eq. (23)] In the term '2αiℜ(e^{φ i}bH ui)', the exponential should be e^{jφ_i} (with the imaginary unit j) to be dimensionally consistent and to match the subsequent phase optimization.
  3. [Algorithm 1] The convergence threshold ǫ and the eavesdropping information limit ε share the same symbol; using different symbols would prevent confusion for the reader.
  4. [Section IV, Eq. (18)] The augmented Lagrangian in (18) is not fully standard in the sign of the penalty term; adding a brief explanation or a pointer to [9, Alg. 1] would help readers verify the subsequent update equations.
  5. [Figure 1 and 2 captions] The axis label '10 104' in Fig. 1 appears to be a formatting artifact; please check that the tick labels render correctly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the BD-RIS optimizations and the FIM gains are derived from stated channel models, and no fitted parameter is relabeled as a prediction.

full rationale

The paper's central derivation is self-contained. In the no-eavesdropper non-reciprocal case, Lemma 1 solves problem (8) by applying von Neumann's trace inequality to the eigenvalue decompositions of E_b and M; the optimal Omega = V_E V_M^H and the bound sum of ordered eigenvalue products follow directly from the stated objective (8) rather than from any assumed answer, and the proof exhibits the equality case. The reciprocal and eavesdropping cases are solved as constrained optimizations (AO and PDD), not by fitting any parameter to the outcome being claimed. The numerical 'significant enhancement' over diagonal RIS is the solution of these optimization problems over the larger unitary feasible set versus the diagonal-unit-modulus baseline; the set-containment gives a one-sided inequality by construction, but the magnitude of the gain is an empirical finding, not an input. The SDP/Gaussian-randomization diagonal baseline and the single channel realization are potential sources of overstatement, but they are evidence-quality limitations, not circularity. Self-citations ([2], [10], [12]) are used for the system model, for a symmetric-unitary projection lemma, and for an analogy to capacity limits; none of these is an unverified premise that already contains the paper's conclusion, and the cited mathematical lemma is parameter-free and independently checkable. Accordingly, no step reduces by definition to its own inputs.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The derivation introduces no new physical entities and no fitted constants. The central analysis rests on the standard linear Gaussian measurement model, the blocked-direct-link assumption, perfect channel knowledge, and the hardware assumption that a fully connected BD-RIS can realize arbitrary or symmetric unitary response matrices. The remaining chosen quantities are simulation settings, not parameters fitted to data.

free parameters (4)
  • Channel coefficient range = Uniform [-0.1, 0.1]
    All channel matrix coefficients in Section V are drawn from this interval for a single realization; the reported gain magnitude may depend on this scale.
  • Noise covariance scale = 10^-5 times identity
    Set in Section V; sets the absolute CRB scale but cancels in relative comparisons.
  • Power allocation matrix = sqrt(30/k) times identity
    Fixed equal power allocation chosen in Section V; not optimized.
  • Eve information threshold epsilon = Varied from 10^3 to 10^6
    An input constraint swept in the figures; not fitted, but the shape of the improvement curves depends on its range.
assumptions (5)
  • domain assumption The direct Alice-to-Bob and Alice-to-Eve links are blocked, so communication occurs only through the cascaded RIS channels.
    System model (1)-(2) inherits this from [2]; if direct links exist, the problem changes qualitatively.
  • domain assumption The fully connected BD-RIS response can be any r by r complex unitary matrix, and reciprocal designs correspond exactly to symmetric unitary matrices.
    Constraint (7b) and Section III rely on this hardware modeling premise from [8] and [10]; feasibility of arbitrary unitary response is not questioned in the paper.
  • domain assumption The channels and noise covariances are perfectly known to the designer.
    All optimizations assume perfect channel state information; no estimation error or robustness analysis is included.
  • standard math Von Neumann's trace inequality and the eigenvalue decompositions of positive semidefinite matrices are valid.
    Used in the proof of Lemma 1, Eq. (10), to bound the objective and construct the optimal unitary matrix.
  • domain assumption The alternating optimization and PDD algorithms inherit the convergence guarantees of [9], [13], and [14].
    Algorithm 1 and Section IV-C refer to prior analyses; the paper offers no new convergence proof for its specific nonconvex problems.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Diagonal RIS Design for Parameter Estimation With and Without Eavesdropping." pith.science (2026). https://pith.science/paper/F3GNWSR3

@misc{pith2026250505971,
  author       = {Pith},
  title        = {Pith review of: Beyond Diagonal RIS Design for Parameter Estimation With and Without Eavesdropping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F3GNWSR3}},
  note         = {Machine review of arXiv:2505.05971}
}
read the original abstract

In this letter, we investigate the transmission of a complex-valued parameter vector from a transmitter to an intended receiver, considering both the presence and absence of an eavesdropper. The direct links from the transmitter to both the intended receiver and the eavesdropper are assumed to be blocked, and communications occur solely through cascaded channels facilitated by a beyond-diagonal reconfigurable intelligent surface (BD-RIS). While previous research has considered this system under conventional (diagonal) RIS assistance, we extend the setup to incorporate BD-RIS and quantify the resulting improvement in estimation performance at the intended receiver. This performance is measured by the trace of the Fisher information matrix (FIM), or equivalently, the average Fisher information, while simultaneously limiting the estimation capability of the eavesdropper. We propose solutions and algorithms for optimizing the BD-RIS response matrix and demonstrate their effectiveness. Numerical results reveal that the BD-RIS provides a significant enhancement in estimation quality compared to conventional diagonal RIS architectures.

Figures

Figures reproduced from arXiv: 2505.05971 by the authors.

Figure 1
Figure 1. Trace of FIM at Bob achieved by non-reciprocal BD-RIS [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. CRB (equivalently, MSE of the MLE) at Bob achieved by n [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Trace of FIM at Bob achieved by non-reciprocal BD-RIS [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: CRB (equivalently, MSE of the MLE) at Bob achieved by n [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 14 canonical work pages

  1. [13]

    MIMO capacity maximization with beyond-diagonal RIS,

    I. Santamaria, M. Soleymani, E. Jorswieck, and J. Guti´ errez, “MIMO capacity maximization with beyond-diagonal RIS,” in 2024 IEEE 25th International W orkshop on Signal Processing Advances in Wi reless Communications (SPAWC), 2024, pp. 936–940

    work page 2024

  2. [10]

    Wideband channel capacity maximization with beyond diagonal RIS reflection matrices,

    ¨O. T. Demir and E. Bj¨ ornson, “Wideband channel capacity maximization with beyond diagonal RIS reflection matrices,” IEEE Wireless Commu- nications Letters , 2024

    work page 2024

  3. [1]

    Proacti ve anti-eavesdropping with trap deployment in wireless netwo rks,

    Q. He, S. Fang, T. Wang, Y . Liu, S. Zhao, and Z. Lu, “Proacti ve anti-eavesdropping with trap deployment in wireless netwo rks,” IEEE Transactions on Dependable and Secure Computing , vol. 20, no. 1, pp. 637–649, 2022

    work page 2022

  4. [2]

    Joint RIS pha se profile design and power allocation for parameter estimation in pre sence of eavesdropping,

    E. M. Abadi, A. N. Hokmabadi, and S. Gezici, “Joint RIS pha se profile design and power allocation for parameter estimation in pre sence of eavesdropping,” IEEE Transactions on V ehicular Technology , vol. 73, no. 12, pp. 19 186–19 202, 2024

    work page 2024

  5. [3]

    Safeguarding 5G-and- beyond networks with physical layer security,

    Y . Wu, T. Q. Duong, and A. L. Swindlehurst, “Safeguarding 5G-and- beyond networks with physical layer security,” IEEE Wireless Commu- nications, vol. 26, no. 5, pp. 4–5, 2019

    work page 2019

  6. [4]

    Wireless physical-layer securi ty: Lessons learned from information theory,

    A. Y ener and S. Ulukus, “Wireless physical-layer securi ty: Lessons learned from information theory,” Proceedings of the IEEE , vol. 103, no. 10, pp. 1814–1825, 2015

    work page 2015

  7. [5]

    Reconfigurable intelligent surfaces: Principles and oppo rtunities,

    Y . Liu, X. Liu, X. Mu, T. Hou, J. Xu, M. Di Renzo, and N. Al-Dh ahir, “Reconfigurable intelligent surfaces: Principles and oppo rtunities,” IEEE Communications Surveys & Tutorials , vol. 23, no. 3, pp. 1546–1577, 2021

    work page 2021

  8. [6]

    RIS-ass isted physical layer security in emerging RF and optical wireless communica- tion systems: A comprehensive survey,

    M. H. Khoshafa, O. Maraqa, J. M. Moualeu, S. Aboagye, T. M. Ngatched, M. H. Ahmed, Y . Gadallah, and M. Di Renzo, “RIS-ass isted physical layer security in emerging RF and optical wireless communica- tion systems: A comprehensive survey,” IEEE Communications Surveys & Tutorials, 2024

    work page 2024

Show all 15 references
  1. [7]

    Reconfigurable intelligent surfaces 2.0: Beyond diagonal phase shift matrices,

    H. Li, S. Shen, M. Nerini, and B. Clerckx, “Reconfigurable intelligent surfaces 2.0: Beyond diagonal phase shift matrices,” IEEE Communica- tions Magazine, vol. 62, no. 3, pp. 102–108, 2024

    2024

  2. [8]

    Modeling and architec ture design of reconfigurable intelligent surfaces using scattering pa rameter network analysis,

    S. Shen, B. Clerckx, and R. Murch, “Modeling and architec ture design of reconfigurable intelligent surfaces using scattering pa rameter network analysis,” IEEE Transactions on Wireless Communications , vol. 21, no. 2, pp. 1229–1243, 2021

    2021

  3. [9]

    Opti mizing power consumption, energy efficiency, and sum-rate using be yond diagonal RIS—a unified approach,

    Y . Zhou, Y . Liu, H. Li, Q. Wu, S. Shen, and B. Clerckx, “Opti mizing power consumption, energy efficiency, and sum-rate using be yond diagonal RIS—a unified approach,” IEEE Transactions on Wireless Communications, vol. 23, no. 7, pp. 7423–7438, 2023

    2023

  4. [11]

    Power adapt ation for vector parameter estimation according to Fisher information base d optimality criteria,

    D. G¨ urg¨ uno˘ glu, B. Dulek, and S. Gezici, “Power adapt ation for vector parameter estimation according to Fisher information base d optimality criteria,” Signal Processing, vol. 192, p. 108390, 2022

    2022

  5. [12]

    Capacity maximization for MIMO channels assisted by beyond-diagonal RIS,

    E. Bj¨ ornson and ¨O. T. Demir, “Capacity maximization for MIMO channels assisted by beyond-diagonal RIS,” arXiv preprint arXiv:2411.18298, 2024

    2024 arXiv

  6. [14]

    Penalty dual decomposition method f or non- smooth nonconvex optimization-Part I: Algorithms and conv ergence analysis,

    Q. Shi and M. Hong, “Penalty dual decomposition method f or non- smooth nonconvex optimization-Part I: Algorithms and conv ergence analysis,” IEEE Transactions on Signal Processing , vol. 68, pp. 4108– 4122, 2020

    2020

  7. [15]

    J. R. Magnus and H. Neudecker, Matrix Differential Calculus with Applications in Statistics and Econometrics . John Wiley & Sons, 2019

    2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.