Capacity Maximization for MIMO Channels Assisted by Beyond-Diagonal RIS

Emil Bj\"ornson; \"Ozlem Tu\u{g}fe Demir

REVIEW 1 major objections 5 minor 17 references

Capacity Maximization for MIMO Channels Assisted by Beyond-Diagonal RIS

T0 review · 1 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper derives the first closed-form capacity-maximizing reflection matrix for a MIMO channel assisted by a beyond-diagonal RIS, showing that the surface should pair the singular-value directions of the two channel matrices in order of…

desk verdict Rigorous closed-form optimal BD-RIS solution for no-direct-path MIMO, with a clear geometrical story; the direct-path caveat is real but explicitly scoped. read the letter →

arxiv 2411.18298 v2 pith:ZPY3L4PY submitted 2024-11-27 eess.SP cs.ITmath.IT

Emil Björnson , Özlem Tuğfe Demir This is my paper · ORCID

classification eess.SPcs.ITmath.IT

keywords beyond-diagonalRISMIMOcapacityclosed-formoptimizationsingularvaluedecompositionreflectionmatrixwaterfillingpowerallocationsemi-unitarychannelsreconfigurableintelligentsurfaces

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 establishes a closed-form solution for the capacity-maximizing reflection matrix of a beyond-diagonal reconfigurable intelligent surface (BD-RIS) in a narrowband MIMO channel where the transmitter and receiver are connected only through the surface. The optimal surface is $\Theta = V_F V_G^H$, formed from the right singular vectors of the two channel matrices, and the optimal transmit covariance is waterfilling over the squared singular values $\sigma_i^2(F)\sigma_i^2(G^H)$. The result matters because it replaces iterative algorithms with an analytic formula and reveals the geometry: the surface pairs the strongest transmitter-to-surface path with the strongest surface-to-receiver path, then continues in descending order of strength. It also identifies special cases, such as semi-unitary channels with few surface elements, where a conventional diagonal RIS already achieves the same capacity.

What carries the argument

The machinery is the singular value decomposition combined with a majorization-type inequality on singular values of matrix products (Lemma 1). The feasible set for the BD-RIS is the unitary group $\{\Theta:\Theta^H\Theta = I_M\}$, which lets $\Theta$ act as a passive rotation between the right singular-vector bases of $G$ and $F$. The proof bounds the capacity for any $\Theta$ and $Q$ by $\sum_i \log_2(1 + q_i\sigma_i^2(F)\sigma_i^2(G^H)/N_0)$ using the inequality, then shows that $\Theta = V_F V_G^H$ and the waterfilling covariance meet the bound term by term. $U_G$ diagonalizes the transmit covariance, while $V_F$ and $V_G$ realize the direction pairing.

What would settle it

Search for a counterexample: for randomly generated $F$ and $G$, compute the capacity from Theorem 1 and compare it with the largest value of $\log_2\det(I_{N_r} + F\Theta G^H Q G\Theta^H F^H/N_0)$ found by numerical optimization over unitary $\Theta$ and trace-constrained $Q$. The theorem claims equality for every $F,G$, so one instance where the numerical search exceeds the closed form would refute it; under the model, the search should never win.

Watch

Extended reading notes

Core claim

The central claim, stated as Theorem 1, is that for the channel $H = F\Theta G^H$ with no direct path, the maximum of $\log_2\det(I_{N_r} + HQH^H/N_0)$ over unitary $\Theta$ and positive semidefinite $Q$ with trace bound is attained by $\Theta = V_F V_G^H$ and $Q = U_G\operatorname{diag}(q_1,\dots,q_{N_t})U_G^H$, where $V_F$, $V_G$, and $U_G$ come from the singular value decompositions of $F$ and $G$, and $q_i$ are waterfilling powers. The resulting capacity is $\sum_{i=1}^K \log_2(1 + q_i\sigma_i^2(F)\sigma_i^2(G^H)/N_0)$ with $K=\min(N_t,N_r,M)$. Geometrically, the optimal BD-RIS maps the $i$th strongest singular direction of $G$ to the $i$th strongest singular direction of $F$, and the waterfilling level decides how many of those paired paths carry power. Corollary 1 shows that at high SNR the pairing order among the $K$ strongest directions can be any permutation, and Corollary 2 shows that for semi-unitary $F$ and $G$ with $M\le\max(N_r,N_t)$, every unitary $\Theta$ is optimal, so a conventional RIS matches the BD-RIS performance.

Load-bearing premise

The load-bearing premise is the assumption that there are no transmitter-receiver paths except through the BD-RIS; if a direct path exists, the channel becomes $F\Theta G^H + H_d$ and the closed-form $\Theta$ from Theorem 1 is generally no longer optimal.

Editorial extensions

If this is right

A BD-RIS-assisted MIMO link with no direct path can be configured for maximum capacity in closed form, without iterative search.
The optimal reflection matrix does not depend on the transmit power or noise level; waterfilling only selects how many of the paired singular directions are active.
At high SNR, the $K$ strongest directions of $F$ and $G$ can be paired in any order, so the surface only needs to match the two subspaces, not the exact ordering.
When $F$ and $G$ are semi-unitary with equal singular values and $M\le\max(N_t,N_r)$, any unitary reflection matrix achieves the same capacity, and a conventional diagonal RIS is sufficient.
The spatial multiplexing gain is $\min(N_t,N_r,M)$, so the number of BD-RIS elements caps the number of parallel data streams when it is smaller than both antenna counts.

Reading between the lines

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

Beyond the paper, if a direct transmitter–receiver path is added, the effective channel becomes $F\Theta G^H + H_d$, and the closed form is no longer claimed optimal; a natural test is to quantify how far $\Theta = V_F V_G^H$ falls from the numerical optimum as the direct path grows.
A design heuristic the paper does not explore is to project the SVD-pairing solution $\Theta = V_F P V_G^H$ onto the symmetric-unitary set required by reciprocal BD-RIS circuits, then measure the capacity loss; this would connect the closed form to the reciprocal architectures used in earlier work.
A testable consequence of Corollary 2 is a design rule for when BD-RIS hardware is worthwhile: use BD-RIS only when $M > \max(N_t,N_r)$ or when the singular values of $F$ and $G$ are markedly unequal, since otherwise a conventional RIS already reaches the same capacity.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

You should know: this paper gives the first closed-form globally optimal BD-RIS reflection matrix for a MIMO channel, and the proof holds up. The main caveat is the no-direct-path assumption, which is explicit in Section II but never revisited for robustness.

The result itself is clean. Theorem 1 says the optimal Θ is V_F V_G^H, pairing the ordered singular-vector directions of the two channel matrices, with Q a waterfilling covariance in the U_G basis. The proof via Lemma 1's majorization bound is correct and self-contained. The corollaries add real value: at high SNR any permutation among the K strongest singular directions works, and for semi-unitary channels with M ≤ max(Nt,Nr) any unitary Θ is optimal, which tells you when BD-RIS buys nothing over a diagonal RIS. The numerical results are consistent with the theorems, including the semi-unitary behavior in Fig. 4.

What's genuinely new is the explicit construction and the geometric interpretation; the capacity expression itself aligns with the already-published results [14] and [15], and the paper says so. That honesty is to its credit. The SVD-pairing technique is standard, but applying it to close the BD-RIS MIMO problem in one shot is a solid contribution to the subfield.

The soft spots are real but not damaging. The no-direct-path assumption is load-bearing: if H_d ≠ 0, the channel is FΘG^H + H_d, and the Theorem 1 Θ is not generally optimal. The paper states the assumption clearly, so there is no deception, but the title's "MIMO channels assisted by BD-RIS" overstates the demonstrated scope. A short paragraph on how a direct path breaks the solution, or a bound on suboptimality, would fix this. Minor issues: a typo in Appendix C where I_Nr appears instead of I_Nt, and the simulation curves have no error bars, though the gaps between curves are large enough that the qualitative conclusions are not in doubt.

The paper is for RIS/MIMO researchers, especially those working on BD-RIS architectures. It deserves serious refereeing. I would accept it with minor revisions rather than desk-reject.

Referee Report

1 major / 5 minor

Summary. This paper studies a narrowband MIMO channel assisted by a fully connected beyond-diagonal RIS, under the assumption that there are no transmitter-receiver paths except through the RIS. The end-to-end channel is H=FΘG^H with Θ constrained to be unitary. The main result, Theorem 1, provides a closed-form global solution: Θ=V_F V_G^H and Q=U_G diag(q_1,...,q_Nt)U_G^H, where the q_i are waterfilling powers, and the capacity is the sum of K=min(Nt,Nr,M) terms involving the products of the ordered singular values of F and G^H. The proof uses a majorization lemma for singular values of matrix products and an explicit achievability calculation. Corollaries characterize the high-SNR permutation invariance and the semi-unitary channel case. Numerical results compare the closed-form solution with the iterative reciprocal BD-RIS algorithm of [11] and with a conventional RIS baseline.

Significance. If the result is taken within its stated no-direct-path model, it is a significant advance: it gives the first closed-form globally optimal BD-RIS reflection matrix for a MIMO channel, replacing iterative algorithms with a parameter-free analytical solution. The proof via majorization is elegant and self-contained, the achievability step is explicit, and the path-pairing interpretation provides genuine geometric insight. The numerical experiments corroborate the analytical claims and also quantify the gap to reciprocal BD-RIS and conventional RIS. These strengths make the paper a useful contribution to the BD-RIS literature, provided the scope is communicated precisely.

major comments (1)

[Section II, Abstract, and Title] The optimality of Theorem 1 is entirely conditional on the Section II assumption that there are no transmitter-receiver paths except via the BD-RIS. If a direct path H_d is present, the channel becomes H=FΘG^H+H_d, the factorization used in equation (14) no longer holds, the singular-value alignment argument in equation (16) fails, and Θ=V_F V_G^H is not generally optimal. Since the title and abstract claim a capacity-maximizing BD-RIS reflection matrix 'for a MIMO channel' without this qualification, the demonstrated scope is narrower than the stated claim. Please qualify the title, abstract, and conclusions, and either add a discussion of the direct-link case or provide a bound on the suboptimality of the proposed Θ when H_d≠0.

minor comments (5)

[Appendix C, equation (20)] The second determinant in equation (20) should be with respect to I_{N_t}, not I_{N_r}; as written, it would scale with N_r and contradict equation (12) when N_r≠N_t.
[Appendix B, Corollary 1] The proof drops the '+1' when passing from the exact capacity expression to equation (17), so Corollary 1 should be stated as an asymptotic high-SNR result; for finite SNR, a permuted pairing can be strictly suboptimal.
[Appendix C, Corollary 2] The statement that F and G each have K equally large singular values is ambiguous in cases such as N_t≤M≤N_r, where a semi-unitary F with orthonormal columns has M nonzero singular values while K=N_t; please clarify the convention used in the proof, where F^H F=σ_F^2 I_M and G G^H=σ_G^2 I_{N_t}.
[Section IV] The plotted curves are averages over many random channel realizations, but no error bars, confidence intervals, or number of realizations are reported; please add this information so the variability of the comparisons can be assessed.
[Footnote 3] The initialization expression 'FHH*G' is not typeset clearly and the matrix H* is defined only in prose; please rewrite the expression with explicit definitions and dimensions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the closed-form capacity result follows from an independent majorization inequality and explicit achievability; self-citations are contextual only.

full rationale

The paper's derivation chain is self-contained. Under the explicitly stated assumption that no transmitter–receiver path exists except via the BD-RIS, the channel is H = FΘG^H, and the optimization problem (3)–(5) is a well-defined mathematical problem. Theorem 1 is proved in Appendix A using Lemma 1, which is based on external matrix-inequality results [16], [17], to establish an upper bound on the capacity for any feasible Θ and Q. The upper bound is then shown to be achievable by the explicit construction Θ = V_F V_G^H and Q = U_G diag(q_1,...,q_Nt) U_G^H via the singular-value computation in (16). No parameter is fitted to data, and no empirical quantity is renamed as a prediction. The waterfilling power allocation is standard and is cited to the authors' textbook [13], but the theorem does not rest on that citation; the same waterfilling step is a direct consequence of the singular-value decomposition of the effective channel. Self-citations [10] and [13] are contextual or standard, not load-bearing. The numerical section validates the analytical expressions on random channel realizations and compares with the algorithm of [11]; it does not set any constant used in the derivation. The only significant limitation is the no-direct-path assumption, which is stated in Section II rather than smuggled in; the proof would indeed not extend automatically to H = FΘG^H + H_d, but this is a scoping condition, not circular reasoning. Thus no circular step is present.

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

The central claim rests only on the standard BD-RIS channel model and on classical majorization inequalities. No parameters are fitted to data; qmax and N0 are problem inputs. The channel matrices F and G are assumed known via perfect CSI.

assumptions (5)

domain assumption The end-to-end channel is H = F Θ G^H with no direct transmitter-receiver path
Section II states that the only paths are via the BD-RIS; this defines the optimization problem.
domain assumption The BD-RIS reflection matrix Θ is any unitary matrix (fully connected, lossless, non-reciprocal)
Section II defines the feasible set U = {Θ: Θ^H Θ = I_M}; non-reciprocity is required for full unitarity.
standard math Gaussian signalling with waterfilling is capacity-achieving
Section II invokes standard MIMO capacity results [13].
standard math Majorization inequalities [16], [17] bound products of singular values
Appendix A, Lemma 1 relies on [16, Th. H.1.b] and [17, Th. 3.3.10(b)].
domain assumption Perfect channel state information is available
Section II assumes perfect CSI for both F and G.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Capacity Maximization for MIMO Channels Assisted by Beyond-Diagonal RIS." pith.science (2026). https://pith.science/paper/ZPY3L4PY

@misc{pith2026241118298,
  author       = {Pith},
  title        = {Pith review of: Capacity Maximization for MIMO Channels Assisted by Beyond-Diagonal RIS},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZPY3L4PY}},
  note         = {Machine review of arXiv:2411.18298}
}

read the original abstract

Reconfigurable intelligent surfaces (RISs) can improve the capacity of wireless communication links by passively beamforming the impinging signals in desired directions. This feature has been demonstrated both analytically and experimentally for conventional RISs, consisting of independently reflecting elements. To further enhance reconfigurability, a new architecture called beyond-diagonal RIS (BD-RIS) has been proposed. It allows for controllable signal flows between RIS elements, resulting in a non-diagonal reflection matrix, unlike the conventional RIS architecture. Previous studies on BD-RIS-assisted communications have predominantly considered single-antenna transmitters/receivers. One recent work provides an iterative capacity-improving algorithm for multiple-input multiple-output (MIMO) setups but without providing geometrical insights. In this paper, we derive the first closed-form capacity-maximizing BD-RIS reflection matrix for a MIMO channel. We describe how this solution pairs together propagation paths, how it behaves when the signal-to-noise ratio is high, and what capacity is achievable with ideal semi-unitary channel matrices. The analytical results are corroborated numerically.

Figures

Figures reproduced from arXiv: 2411.18298 by the authors.

**Figure 2.** The capacity achieved with different RIS architectures [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** The capacity achieved versus the SNR for [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: The capacity achieved versus the number of elements [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 13 canonical work pages

[14]

Line-of-sight MIMO via intelligent reflecting surface,

H. Do, N. Lee, and A. Lozano, “Line-of-sight MIMO via intelligent reflecting surface,” IEEE Transactions on Wireless Communications , vol. 22, no. 6, pp. 4215–4231, 2022

work page 2022
[15]

Spatial multiplexing in near field MIMO channels with reconfigurable intelligent surfaces,

G. Bartoli, A. Abrardo, N. Decarli, D. Dardari, and M. Di Renzo, “Spatial multiplexing in near field MIMO channels with reconfigurable intelligent surfaces,” IET Signal Processing , vol. 17, no. 3, p. e12195, 2023

work page 2023
[11]

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 IEEE Int. W. Signal Process. Adv. Wireless Commun. (SPAWC), 2024

work page 2024
[1]

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]

Toward intelligent metasurfaces: The progress from globally tunable metasurfaces to software-defined metasurfaces with an embedded network of controllers,

O. Tsilipakos et al., “Toward intelligent metasurfaces: The progress from globally tunable metasurfaces to software-defined metasurfaces with an embedded network of controllers,” Advanced Optical Materials , no. 2000783, 2020

work page 2020
[3]

Smart radio environments empowered by reconfig- urable intelligent surfaces: How it works, state of research, and the road ahead,

M. Di Renzo et al., “Smart radio environments empowered by reconfig- urable intelligent surfaces: How it works, state of research, and the road ahead,” IEEE Journal of Selected Areas in Communications , vol. 38, no. 11, pp. 2450–2525, 2020

work page 2020
[4]

Reconfigurable intelligent surfaces: A signal pro- cessing perspective with wireless applications,

E. Bj ¨ornson, H. Wymeersch, B. Matthiesen, P. Popovski, L. Sanguinetti, and E. de Carvalho, “Reconfigurable intelligent surfaces: A signal pro- cessing perspective with wireless applications,” IEEE Signal Processing Magazine, vol. 39, no. 2, pp. 135–158, 2022

work page 2022
[5]

Reconfigurable intelligent surface (RIS) in the sub- 6 GHz band: Design, implementation, and real-world demonstration,

A. Araghi et al. , “Reconfigurable intelligent surface (RIS) in the sub- 6 GHz band: Design, implementation, and real-world demonstration,” IEEE Access, vol. 10, pp. 2646–2655, 2022

work page 2022

Show all 17 references

[6]

Modeling and architecture design of reconfigurable intelligent surfaces using scattering parameter network analysis,

S. Shen, B. Clerckx, and R. Murch, “Modeling and architecture design of reconfigurable intelligent surfaces using scattering parameter network analysis,” IEEE Trans. Wireless Commun., vol. 21, no. 2, pp. 1229–1243, 2021

2021
[7]

Beyond diagonal reconfigurable intelligent surfaces: From transmitting and reflecting modes to single- , group-, and fully-connected architectures,

H. Li, S. Shen, and B. Clerckx, “Beyond diagonal reconfigurable intelligent surfaces: From transmitting and reflecting modes to single- , group-, and fully-connected architectures,” IEEE Transactions on Wireless Communications, vol. 22, no. 4, pp. 2311–2324, 2022

2022
[8]

Closed-form global optimization of beyond diagonal reconfigurable intelligent surfaces,

M. Nerini, S. Shen, and B. Clerckx, “Closed-form global optimization of beyond diagonal reconfigurable intelligent surfaces,” IEEE Transactions on Wireless Communications, vol. 23, no. 2, pp. 1037–1051, 2024

2024
[9]

Optimiz- ing power consumption, energy efficiency and sum-rate using beyond diagonal RIS—a unified approach,

Y . Zhou, Y . Liu, H. Li, Q. Wu, S. Shen, and B. Clerckx, “Optimiz- ing power consumption, energy efficiency and sum-rate using beyond diagonal RIS—a unified approach,” IEEE Trans. Wirel. Commun., 2023

2023
[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, vol. 13, no. 10, pp. 2687–2691, 2024

2024
[12]

Channel estimation and beamforming for beyond diagonal reconfigurable intelligent surfaces,

H. Li, S. Shen, Y . Zhang, and B. Clerckx, “Channel estimation and beamforming for beyond diagonal reconfigurable intelligent surfaces,” arXiv preprint arXiv:2403.18087 , 2024

2024 arXiv
[13]

Introduction to multiple antenna com- munications and reconfigurable surfaces,

E. Bj ¨ornson and ¨O. T. Demir, “Introduction to multiple antenna com- munications and reconfigurable surfaces,” Now Publishers, Inc. , 2024

2024
[16]

A. W. Marshall, I. Olkin, and B. C. Arnold, Inequalities: Theory of majorization and its applications . Springer, 2011

2011
[17]

R. A. Horn and C. R. Johnson, Topics in Matrix Analysis . Cambridge University Press, 1991

1991

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [14]

Line-of-sight MIMO via intelligent reflecting surface,

H. Do, N. Lee, and A. Lozano, “Line-of-sight MIMO via intelligent reflecting surface,” IEEE Transactions on Wireless Communications , vol. 22, no. 6, pp. 4215–4231, 2022

work page 2022

[2] [15]

Spatial multiplexing in near field MIMO channels with reconfigurable intelligent surfaces,

G. Bartoli, A. Abrardo, N. Decarli, D. Dardari, and M. Di Renzo, “Spatial multiplexing in near field MIMO channels with reconfigurable intelligent surfaces,” IET Signal Processing , vol. 17, no. 3, p. e12195, 2023

work page 2023

[3] [11]

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 IEEE Int. W. Signal Process. Adv. Wireless Commun. (SPAWC), 2024

work page 2024

[4] [1]

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

[5] [2]

Toward intelligent metasurfaces: The progress from globally tunable metasurfaces to software-defined metasurfaces with an embedded network of controllers,

O. Tsilipakos et al., “Toward intelligent metasurfaces: The progress from globally tunable metasurfaces to software-defined metasurfaces with an embedded network of controllers,” Advanced Optical Materials , no. 2000783, 2020

work page 2020

[6] [3]

Smart radio environments empowered by reconfig- urable intelligent surfaces: How it works, state of research, and the road ahead,

M. Di Renzo et al., “Smart radio environments empowered by reconfig- urable intelligent surfaces: How it works, state of research, and the road ahead,” IEEE Journal of Selected Areas in Communications , vol. 38, no. 11, pp. 2450–2525, 2020

work page 2020

[7] [4]

Reconfigurable intelligent surfaces: A signal pro- cessing perspective with wireless applications,

E. Bj ¨ornson, H. Wymeersch, B. Matthiesen, P. Popovski, L. Sanguinetti, and E. de Carvalho, “Reconfigurable intelligent surfaces: A signal pro- cessing perspective with wireless applications,” IEEE Signal Processing Magazine, vol. 39, no. 2, pp. 135–158, 2022

work page 2022

[8] [5]

Reconfigurable intelligent surface (RIS) in the sub- 6 GHz band: Design, implementation, and real-world demonstration,

A. Araghi et al. , “Reconfigurable intelligent surface (RIS) in the sub- 6 GHz band: Design, implementation, and real-world demonstration,” IEEE Access, vol. 10, pp. 2646–2655, 2022

work page 2022