Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

A Novel Q-stem Connected Architecture for Beyond-Diagonal Reconfigurable Intelligent Surfaces

T0 review · 4 major / 6 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 For a multi-user MISO downlink, a Q-stem connected BD-RIS reaches the sum channel gain of a fully connected RIS when Q = 2M − 1 (M the number of independent streams), using far fewer tunable admittances.

desk verdict Q-stem is a useful new BD-RIS topology, but the LS derivation drops the SVD phase and the headline parity claim is empirical. read the letter →

arxiv 2411.18480 v1 pith:CGJTLLKM submitted 2024-11-27 cs.IT eess.SPmath.IT

Xiaohua Zhou , Tianyu Fang , Yijie Mao This is my paper · ORCID
classification cs.ITeess.SPmath.IT
keywords beyond-diagonalreconfigurableintelligentsurfaceQ-stemconnectedRISsumchannelgainmaximizationscatteringmatrixdesignleastsquaresalgorithmquasi-Newtonmethodmulti-userMISOgraph-basedarchitecture
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 proposes a new beyond-diagonal reconfigurable intelligent surface (BD-RIS) topology, the Q-stem connected RIS, in which the first Q ports connect to every other port while the remaining ports connect only to those Q ports. The authors argue this family interpolates between single-connected, tree-connected, and fully connected RIS, and that in an L-antenna base station serving K single-antenna users it can match the sum channel gain of a fully connected surface once Q = 2M − 1, where M is the number of independent data streams. To design the surface, they give a closed-form least-squares algorithm that approximately satisfies the channel-alignment condition V_M^H Θ P_M = Φ, plus a quasi-Newton refinement initialized from that least-squares solution. The practical claim is that near-full performance can be bought with far fewer tunable impedance components and much lower computational cost than full connectivity.

What carries the argument

The load-bearing object is the Cayley-type scattering matrix Θ = (I_N + jZ_0 B)^{-1}(I_N − jZ_0 B), where the symmetric real susceptance matrix B encodes which ports are connected by tunable admittances. For Q-stem connected RIS the constraint is [B]_{n,m} = 0 whenever both indices exceed Q, so B is dense only in the first Q rows and columns. The design machinery is the independent vectorization veci(B) = b with a 0/1 assembly matrix R satisfying vec(B) = R b, which turns the channel-alignment condition B C = D into the linear system A b = z; the least-squares inverse (A^T A)^{-1} A^T z gives a closed-form scattering design, and the same b initializes the quasi-Newton method on the original non-convex problem.

What would settle it

Compute, for generic random channel realizations with M > 1, whether there exists any diagonal unitary Φ such that the Q-stem constraint set B_Q contains a symmetric B satisfying (I + jZ_0 B)^{-1}(I − jZ_0 B) = V_M Φ P_M^H + V_{N−M} X P_{N−M}^H; if the answer is no while the simulation still shows parity, then the observed gain parity is not explained by the stated derivation. A simpler direct test is to simulate N=128, L=K=8 with Q=2M−1 and compare the least-squares design against fully connected RIS over many channel realizations; any systematic gap refutes the parity claim.

Watch

Extended reading notes

Core claim

Working in a multiuser MISO downlink where the BS-RIS channel E and the RIS-user channels H are given, the paper shows that the sum channel gain ∥H^H Θ E∥$_F^{2}$ is upper-bounded by ∥S_M Σ_M∥$_F^{2}$ via the SVD of H^H and E, and that this bound would require the channel-alignment identity V_M^H Θ P_M = Φ with Φ a diagonal unitary matrix. It then proves that when the multiplexing gain M > 1, no reciprocal BD-RIS, including the fully connected one, can satisfy that identity exactly, so full connectivity is not the bound but the practical gold standard. The proposed Q-stem connected RIS—where the first Q ports connect to all ports and the last N − Q ports connect only to the hub—is shown in simulation to match the fully connected sum channel gain when Q = 2M − 1, using only QN + N − Q(Q+1)/2 tunable admittances instead of N(N+1)/2. The design is carried by a least-squares solution of the linearized identity B C = D and, when higher accuracy is wanted, a quasi-Newton refinement initialized from that least-squares point.

Load-bearing premise

The load-bearing premise is that the phase matrix Φ in the sufficient condition V_M^H Θ P_M = Φ can be set to identity when the condition is rewritten as B C = D, so that the least-squares solution targets exactly the condition that guarantees the upper bound rather than a stricter one.

Editorial extensions

If this is right

  • When Q is fixed, the circuit complexity QN + N − Q(Q+1)/2 grows linearly with N, so the same surface can be scaled to many elements without quadratic hardware growth.
  • At the threshold Q = 2M − 1, the simulated sum channel gain of Q-stem connected RIS equals that of fully connected RIS, while the number of tunable admittances drops from N(N+1)/2 to roughly (2M−1)N.
  • The least-squares design is closed-form with O(Q^3 N^3) complexity, and using it to initialize quasi-Newton beats random initialization; for Q ≥ 7 the least-squares method alone matches the quasi-Newton result in the simulated N=64, L=K=4 setup.
  • Because Q=0, Q=1, and Q=N−1 reproduce single-connected, tree-connected, and fully connected RIS, any performance gain found for intermediate Q is a smooth interpolation among those existing architecture classes.

Reading between the lines

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

  • The Q = 2M − 1 rule, if it survives other settings, gives hardware designers a direct recipe: interconnect roughly twice as many hub ports as the number of spatial streams, independent of the total element count.
  • The same vectorization-and-least-squares machinery should transfer to any BD-RIS topology whose susceptance support is fixed, since only the 0/1 matrix R changes; multi-sector BD-RIS and STAR-RIS are natural next targets.
  • A clean proof that the phase matrix Φ can be absorbed into the SVD factors would upgrade the parity observation from simulation to theorem; without it, the reported Q=2M−1 law should be read as empirical.
  • The analysis assumes blocked direct links and Rayleigh fading; with strong direct paths or correlated channels the optimal Q may shift, so the threshold is worth testing beyond the current model.
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.

Referee Report

4 major / 6 minor

Summary. The paper proposes a new beyond-diagonal reconfigurable intelligent surface (BD-RIS) architecture, termed Q-stem connected RIS, which interpolates between single-connected, tree-connected, and fully-connected RIS by letting a set of Q ports connect to all other ports while remaining ports connect only to the first Q ports. The authors formulate a sum channel gain maximization problem for a multi-user MISO downlink, derive an SVD-based upper bound, and propose two algorithms: a closed-form least-squares (LS) design and an LS-initialized quasi-Newton method. Simulation results claim that with a specific Q (e.g., Q=2M-1 in one scenario) the Q-stem architecture attains the sum channel gain of fully connected RIS at reduced circuit complexity.

Significance. If the claims hold, the proposed architecture is a useful addition to the BD-RIS design space, offering a tunable complexity-performance trade-off. The circuit complexity expression QN + N - Q(Q+1)/2 is derived correctly and reduces to known values for Q=0, Q=1, and Q=N-1. The SVD-based upper-bound analysis is a standard and useful tool. The paper also shows the value of good initialization for quasi-Newton methods. However, the theoretical derivation of the LS algorithm contains a load-bearing algebraic error, so the significance of the algorithmic contribution is currently not established.

major comments (4)
  1. [III-B, Eq. (16)] Equation (16) is not equivalent to Equation (12). Substituting the Cayley form (8b) into (12) with Phi = I gives V_M^H (I - jZ0B)(I + jZ0B)^{-1} P_M = I, which is not linear in B. In contrast, the equation B C = D with C = jZ0(V_M + P_M) and D = P_M - V_M is equivalent to (I - jZ0B) P_M = (I + jZ0B) V_M, i.e., to the stronger condition Theta P_M = V_M. The latter implies V_M^H Theta P_M = I but is not implied by it. The LS algorithm therefore solves a stricter alignment problem than the sufficient condition for the upper bound, so the stated derivation does not support Algorithm 1.
  2. [III-B, Eq. (12)-(16)] The diagonal phase matrix Phi in Eq. (12) is silently set to identity. For a general SVD output, the condition should be B(P_M + V_M Phi) = (P_M - V_M Phi)/(jZ0) (when expressed as the stronger linear form). Standard SVD routines return arbitrary column phases, so without an explicit gauge normalization of V_M and P_M, the LS solution depends on the SVD routine's choice of phases. The paper provides no such normalization, making the algorithm and the subsequent simulation results irreproducible with respect to the SVD gauge.
  3. [Appendix, Proposition 1] Proposition 1 states that fully connected RIS 'can not achieve' the upper bound when M > 1, but the proof only establishes that the symmetry condition Lambda = Lambda^T fails with probability 1 for generic channel realizations. There exist special channel matrices for which Lambda is symmetric, so the statement should be qualified as 'almost surely' or 'with probability 1'. This matters because the impossibility claim is used to motivate the LS approximation for all BD-RIS structures.
  4. [Algorithm 1, Step 6] The least-squares solution b = (A^T A)^{-1} A^T z requires A^T A to be invertible. The paper does not prove that A has full column rank. The matrix A has size 2MN x (QN + N - Q(Q+1)/2), so for small Q the number of unknowns can exceed the number of equations, and for specific channel realizations rank deficiency can occur even when the dimensions are compatible. The authors should provide a rank analysis or replace the inverse with a regularized pseudo-inverse.
minor comments (6)
  1. [Algorithm 1 heading] The heading reads 'least spare' and should be 'least squares'.
  2. [II-D] The word 'connnected' is misspelled; it should be 'connected'.
  3. [II-B] The phrase 'a feasible susceptance matrice' should be 'a feasible susceptance matrix'.
  4. [Figures 3-5] The figures appear without axis labels or legends in the text; please add clear labels (e.g., 'Q' and 'sum channel gain') and a legend identifying each curve.
  5. [III-A, Eq. (10)] The DoF definition M = min(K,L,N) is presented as a limit; it would be clearer to state that M is the maximum rank of the effective channel F, which is at most min(K,L,N).
  6. [Appendix] The phrase 'Without loss of generality, we set phi_m = 0' needs a justification, e.g., by the gauge freedom in the SVD of H^H and E. Without such a note, the reduction to phi_m = 0 appears unjustified.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the Q-stem architecture and LS design are constructed from the SVD upper bound and benchmarked empirically; self-citations are background only.

full rationale

The paper's derivation is self-contained. The Q-stem susceptance constraint (Def. 1, Eq. (5)) and the circuit-complexity count are direct definitions and combinatorics, not imported from the claimed performance. Algorithm 1 is an explicit construction: it decomposes H^H and E by SVD, forms C=jZ0(V_M+P_M) and D=P_M-V_M, and solves a linear least-squares problem (20)-(21) for the independent entries of B; the simulated parity with fully connected RIS is an empirical outcome benchmarked against the upper bound ||S_M Sigma_M||_F, not a fitted parameter. The quasi-Newton method optimizes the actual objective (22) and is only initialized by the LS solution, so the LS output is not renamed as a prediction. Self-citations [11] and [15] are background references to existing fully-connected algorithms and to a simulation path-loss convention; they do not carry the central claim. The one notable weakness is a derivation gap, not circularity: the reduction of (12) to (16) sets the diagonal unitary Phi to identity, and the Appendix's 'Without loss of generality, we set phi_m = 0' (Appendix, after Eq. (25)) is asserted rather than proved for fixed SVD outputs. This affects the rigor of the sufficient-condition argument, but the algorithm's output is not defined in terms of the simulated performance, and no result reduces to its own inputs by construction.

Assumptions & free parameters 0 free parameters · 6 assumptions · 1 invented entities

The central derivation rests on standard SVD and microwave-network axioms. The main added assumptions are the unstated phase normalization in Eq. (16) and the full-rank assumption for A^T A. No data-fitted parameters are introduced; Q is a design variable swept in simulation, and the path-loss constants are inherited from prior literature.

assumptions (6)
  • domain assumption Scattering matrix model Θ=(I+jZ0B)^{-1}(I-jZ0B) for a reciprocal lossless network
    Eq. (1); standard microwave network theory adopted from refs [2], [8], and [9].
  • domain assumption Effective channels are characterized by the sum of squared Frobenius norms of H^H Θ E, with direct BS-user links blocked
    Section II-A; Rayleigh fading path-loss model; objective is sum channel gain, not sum rate.
  • standard math The relaxed upper bound ∥S_M Σ_M∥_F^2 is the relevant performance ceiling and Eq. (12) is its sufficient and necessary condition under unitarity
    Section III-A; consequence of SVD and unitary invariance.
  • domain assumption Λ=P_M^H V_M^* is non-symmetric with probability 1 for random Rayleigh channels, so fully connected BD-RIS cannot hit the bound when M>1
    Appendix; true for generic continuous channel distributions, but Proposition 1 states it without the almost-sure qualifier.
  • ad hoc to paper The phase matrix Φ in Eq. (12) can be removed or absorbed into V_M and P_M, making Eq. (16) equivalent
    Section III-B, Eq. (16); this normalization is not stated or proved.
  • ad hoc to paper The matrix A^T A in Eq. (21) is invertible for the LS solution
    Section III-B, Algorithm 1; no rank or conditioning analysis is provided.
invented entities (1)
  • Q-stem connected RIS topology
    purpose: A graph architecture where Q hub ports connect to all N ports and the remaining ports connect only to the hubs, providing a tunable complexity-performance trade-off.
    Validated only through Monte Carlo simulation; no prototype, circuit measurement, or independent benchmark is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Novel Q-stem Connected Architecture for Beyond-Diagonal Reconfigurable Intelligent Surfaces." pith.science (2026). https://pith.science/paper/CGJTLLKM

@misc{pith2026241118480,
  author       = {Pith},
  title        = {Pith review of: A Novel Q-stem Connected Architecture for Beyond-Diagonal Reconfigurable Intelligent Surfaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CGJTLLKM}},
  note         = {Machine review of arXiv:2411.18480}
}
read the original abstract

Beyond-diagonal reconfigurable intelligent surface (BD-RIS) has garnered significant research interest recently due to its ability to generalize existing reconfigurable intelligent surface (RIS) architectures and provide enhanced performance through flexible inter-connection among RIS elements. However, current BD-RIS designs often face challenges related to high circuit complexity and computational complexity, and there is limited study on the trade-off between system performance and circuit complexity. To address these issues, in this work, we propose a novel BD-RIS architecture named Q-stem connected RIS that integrates the characteristics of existing single connected, tree connected, and fully connected BD-RIS, facilitating an effective trade-off between system performance and circuit complexity. Additionally, we propose two algorithms to design the RIS scattering matrix for a Q-stem connected RIS aided multi-user broadcast channels, namely, a low-complexity least squares (LS) algorithm and a suboptimal LS-based quasi-Newton algorithm. Simulations show that the proposed architecture is capable of attaining the sum channel gain achieved by fully connected RIS while reducing the circuit complexity. Moreover, the proposed LS-based quasi-Newton algorithm significantly outperforms the baselines, while the LS algorithm provides comparable performance with a substantial reduction in computational complexity.

Figures

Figures reproduced from arXiv: 2411.18480 by the authors.

Figure 1
Figure 1. The circuit architecture of Q-stem connected RIS and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The shape of a feasible susceptance matrix [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 5
Figure 5. Channel gain versus the number of streams when L = 5 and N = 64. APPENDIX When considering a fully connected RIS, i.e., Q = N − 1, problem (8) reduces to max Θ ∥SMVH MΘPMΣM∥ 2 F (23a) s.t. ΘΘH = IN , Θ = ΘT . (23b) Similar to the analysis of problem (9), the theoretical upper bound of problem (23) can be achieved if and only if the following equation set has solution VH MΘPM = Φ, (24a) ΘΘH = IN , (24b) Θ = ΘT . (24c… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Dual-Polarized Beyond Diagonal RIS

    cs.IT 2024-12 conditional novelty 7.0 of 10

    Grouping two opposite-polarization elements in a beyond diagonal RIS achieves the fully connected performance bound in dual-polarized line-of-sight links, and dual-polarized BD-RIS offers gains over conventional RIS e...

Reference graph

Works this paper leans on

15 extracted references · 10 canonical work pages · cited by 1 Pith paper

  1. [1]

    Towards smart and reconfigurable environment: Intelligent reflecting surface aided wireless network,

    Q. Wu and R. Zhang, “Towards smart and reconfigurable environment: Intelligent reflecting surface aided wireless network,” IEEE Commun. Mag., vol. 58, no. 1, pp. 106–112, Jan 2020

    work page 2020

  2. [2]

    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, Feb. 2022

    work page 2022

  3. [3]

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

    H. Li, S. Shen, and B. Clerckx, “Beyond diagonal reconfigurable intelli- gent surfaces: From transmitting and reflecting modes to single-, group- and fully-connected architectures,” IEEE Trans. Wireless Commun. , vol. 22, no. 4, pp. 2311–2324, Apr. 2023

    work page 2023

  4. [4]

    Intelligent reflecting surface enhanced wireless network via joint active and passive beamforming,

    Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless network via joint active and passive beamforming,”IEEE Trans. Wireless Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019

    2019

  5. [5]

    Beyond diagonal reconfig- urable intelligent surfaces utilizing graph theory: Modeling, architecture design, and optimization,

    M. Nerini, S. Shen, H. Li, and B. Clerckx, “Beyond diagonal reconfig- urable intelligent surfaces utilizing graph theory: Modeling, architecture design, and optimization,” IEEE Trans. Wireless Commun., vol. 23, no. 8, pp. 9972–9985, Aug. 2024

    2024

  6. [6]

    STAR-RISs: Simultaneous transmitting and reflecting reconfigurable intelligent surfaces,

    J. Xu, Y . Liu, X. Mu, and O. A. Dobre, “STAR-RISs: Simultaneous transmitting and reflecting reconfigurable intelligent surfaces,” IEEE Commun. Lett., vol. 25, no. 9, pp. 3134–3138, Sep. 2021

    work page 2021

  7. [7]

    Beyond diagonal reconfigurable intelligent surfaces: A multi-sector mode enabling highly directional full-space wireless coverage,

    H. Li, S. Shen, and B. Clerckx, “Beyond diagonal reconfigurable intelligent surfaces: A multi-sector mode enabling highly directional full-space wireless coverage,”IEEE J. Sel. Areas Commun, vol. 41, no. 8, pp. 2446–2460, Aug. 2023

    work page 2023

  8. [8]

    D. M. Pozar, Microwave engineering: theory and techniques . John wiley & sons, 2021

    2021

Show all 15 references
  1. [9]

    A universal framework for multiport network analysis of reconfigurable intelligent surfaces,

    M. Nerini, S. Shen, H. Li, M. D. Renzo, and B. Clerckx, “A universal framework for multiport network analysis of reconfigurable intelligent surfaces,” IEEE Trans. Wireless Commun. , pp. 1–1, Jun. 2024

    2024

  2. [10]

    SNR maximization in beyond diagonal RIS-assisted single and multiple antenna links,

    I. Santamaria, M. Soleymani, E. Jorswieck, and J. Guti ´errez, “SNR maximization in beyond diagonal RIS-assisted single and multiple antenna links,” IEEE Signal Process. Lett. , vol. 30, pp. 923–926, Jul. 2023

    2023

  3. [11]

    A low-complexity beamforming design for beyond-diagonal RIS aided multi-user networks,

    T. Fang and Y . Mao, “A low-complexity beamforming design for beyond-diagonal RIS aided multi-user networks,” IEEE Commun. Lett. , Jan. 2024

    2024

  4. [12]

    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 Trans. Wireless Commun., Feb. 2024

    2024

  5. [13]

    Channel shaping using beyond diagonal reconfigurable intelligent surface: Analysis, op- timization, and enhanced flexibility,

    Y . Zhao, H. Li, M. Franceschetti, and B. Clerckx, “Channel shaping using beyond diagonal reconfigurable intelligent surface: Analysis, op- timization, and enhanced flexibility,” arXiv preprint arXiv:2407.15196 , 2024

    2024

  6. [14]

    Least squares methods,

    ˚A. Bj ¨orck, “Least squares methods,” Handbook of numerical analysis , vol. 1, pp. 465–652, 1990

    1990

  7. [15]

    Fully connected reconfigurable intelligent surface aided rate-splitting multiple access for multi-user multi-antenna transmission,

    T. Fang, Y . Mao, S. Shen, Z. Zhu, and B. Clerckx, “Fully connected reconfigurable intelligent surface aided rate-splitting multiple access for multi-user multi-antenna transmission,” in IEEE Int. Conf. Commun. Workshops (ICC Workshops), May. 2022, pp. 675–680

    2022

Pith tools

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