Pith. sign in

REVIEW 4 major objections 5 minor 19 references

Dynamical ON-OFF Control with Trajectory Prediction for Multi-RIS Wireless Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that predicting user trajectories and switching RISs on or off in advance maximizes SINR in dense multi-RIS uplinks.

desk verdict LSTM-plus-on/off RIS control is a fresh combination, but the paper oversells it: the phase-selection proxy and greedy ON-OFF rule do not maximize SINR, and the simulations test a channel model that differs from the analysis. read the letter →

arxiv 2505.20887 v1 pith:MHKOY65N submitted 2025-05-27 cs.NI eess.SP

Kaining Wang , Bo Yang , Yusheng Lei , Zhiwen Yu , Xuelin Cao , George C. Alexandropoulos , Marco Di Renzo , Chau Yuen This is my paper
classification cs.NIeess.SP
keywords reconfigurableintelligentsurfaceRISON-OFFcontroltrajectorypredictionLSTMSINRmaximizationinterferencemitigationcodebookphaseconfigurationmulti-RISnetworks
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

Reconfigurable intelligent surfaces reflect whatever impinges on them, so in dense networks they can amplify interference and degrade the signal-to-interference-plus-noise ratio (SINR). The paper proposes a trajectory-prediction-based dynamical control (TPC) scheme: a long short-term memory (LSTM) model at the base station forecasts user positions, and a codebook-based algorithm decides which RISs should be ON or OFF and how to set their phase shifts ahead of time. The aim is to pre-emptively silence RISs that would reflect interfering signals. Simulation results compare TPC with always-on RISs and with an ISL-based spectrum-learning baseline, and claim that TPC attains the highest SINR across transmission powers and numbers of RIS elements.

What carries the argument

The central mechanism is the binary activation vector $V$ and the phase-shift codebook $\Phi$. After an LSTM predicts user coordinates, the algorithm computes the phase configuration $\Phi_i^*$ that maximizes $\left|\sqrt{\eta_l} g_l + \sqrt{\eta_{li}} h_i \Phi_i h_{li}\right|^2$ for each RIS, evaluates the SINR $\gamma_l$ from Eq. (5) with that RIS ON, and compares it with the all-OFF SINR $\gamma'_l$ from Eq. (10); if $\gamma_l < \gamma'_l$, the RIS is turned OFF. This decomposition turns a mixed-integer nonlinear program into a sequence of per-RIS codebook searches and threshold comparisons.

What would settle it

For a small instance with two RISs, two interferers, a two-bit phase codebook, and two elements per RIS, exhaustively search all phase and ON-OFF combinations to find the true maximum SINR from Eq. (5), then compare it with the SINR produced by TPC's per-RIS phase rule from Eq. (9). If any joint configuration beats TPC, the paper's claim of optimal SINR is false.

Watch

Extended reading notes

Core claim

The paper's central claim is that the SINR degradation caused by 'blind reflection' of RISs can be overcome by predicting mobility and controlling RIS binary states before interference builds up. For each RIS, phases are chosen from a discrete codebook by maximizing the magnitude of the desired reflected signal (Eq. (9)); then the RIS is kept ON only if the resulting SINR in Eq. (5) is no worse than the all-OFF direct-link SINR in Eq. (10). The paper asserts that this per-RIS greedy procedure avoids unnecessary interference amplification and achieves an optimal SINR, with simulations showing gains over always-on and ISL-based control for varying transmission power and RIS element counts.

Load-bearing premise

The load-bearing premise is that selecting each RIS's phases to maximize the desired signal magnitude, per Eq. (9), is an adequate proxy for maximizing the full SINR in Eq. (5), even though that selection ignores every interference term in the denominator.

Editorial extensions

If this is right

  • In dense multi-RIS uplinks, keeping all RISs always on can lower SINR; the TPC result implies that switching RISs OFF in time is a viable interference-mitigation lever.
  • The LSTM predictor's roughly $\pm 60$ m trajectory accuracy is enough, in the simulated settings, to precompute useful ON-OFF sequences, connecting mobility prediction to physical-layer control.
  • The codebook phase selection plus per-RIS threshold comparison gives higher SINR than the ISL spectrum-learning baseline across the tested transmission powers and RIS element counts.
  • The three-stage frame (pilot, configuration, transmission) provides a concrete way to implement predictive RIS control in near-real-time.

Reading between the lines

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

  • The paper does not analyze how trajectory-prediction error propagates into ON-OFF decisions; a sensitivity study varying the LSTM error would show at what accuracy TPC degrades to the always-on baseline.
  • The per-RIS phase rule in Eq. (9) ignores all interference terms, so a joint phase-selection variant across RISs is a natural next step that could exceed the reported SINR.
  • The simulations assume reflected power falls with incident angle while the analytical model in Section II-B does not; adding that dependence to Eq. (5) would make the optimality claim analytically checkable.
  • The same predictive ON-OFF logic could apply to any nearly passive reflector deployment where user positions are tracked, such as vehicular or aerial networks.
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 / 5 minor

Summary. The manuscript considers a multi-RIS uplink network in which RISs may reflect both desired and interfering signals, and proposes a trajectory-prediction-based ON-OFF control (TPC) scheme. An LSTM model predicts user trajectories from real GPS data, and a codebook-based phase configuration followed by a greedy binary ON-OFF control algorithm is used to maximize the SINR. Simulation results compare TPC with an always-on RIS baseline and an ISL-based control method.

Significance. If the proposed algorithm truly achieved the claimed optimal SINR, the work would be a useful contribution to RIS interference management under user mobility. The use of a real trajectory dataset for training the LSTM is a positive aspect, and the paper clearly identifies a relevant practical problem (blind reflection of interference). However, the central optimality claim is not supported by the derivations, and the simulation model is inconsistent with the analytical model. As a result, the significance of the paper is not currently established.

major comments (4)
  1. [III-B1, Eq. (9)] The phase selection rule in Eq. (9) maximizes |√η_l g_l + √η_li h_i Φ_i h_li|^2 for each RIS i independently, which is the magnitude of the desired signal component, rather than the SINR γ_l in Eq. (5). The denominator of γ_l contains interference terms such as √η_mi h_i Φ_i h_mi for interfering users m, which depend on the same phase choices, so a phase selection that maximizes the numerator can also strengthen interference. The paper provides no argument that the interference terms are negligible in the regime it targets; the motivating scenario of 'blind reflection' assumes strong interference. Consequently, the claim that Eq. (9) solves problem P2 is not established, and the later statement in Section IV-B2 that the algorithm 'achieves an optimal SINR' is unsupported.
  2. [III-B2, Algorithm 1] Algorithm 1 initializes all RISs to ON and, for each i, compares the SINR with all RISs ON (γ_l) against the SINR with all RISs OFF (γ'_l). This greedy per-RIS rule does not solve the joint binary optimization over V in P1 because the contribution of each RIS depends on the states of the other RISs; the algorithm never evaluates the SINR of the partially updated configuration during the loop. No proof is given that the output V* is optimal, and the caption's reference to 'Optimal ON-OFF vector' is therefore not justified.
  3. [IV-B1] The simulation settings state that 'the reflected power from the RIS is inversely proportional to the incident angle of the signal,' but this angle-dependent law is absent from the analytical channel model in Section II-B, where path loss is given by the product-distance formula in Eq. (2). The incident angle is not defined in the model, and its effect on Eqs. (5), (9), and (10) is not specified. This discrepancy means the simulations do not validate the model analyzed in the paper, so the reported SINR gains cannot be attributed to the proposed algorithm as derived.
  4. [IV-B2, Figs. 4 and 5] The SINR results in Figs. 4 and 5 are reported as single curves without error bars, confidence intervals, or a statement of the number of channel realizations. Since the SINR in Eq. (5) depends on random small-scale fading and user positions, the absence of statistical characterization does not support the comparative claims of superiority.
minor comments (5)
  1. [II-B, Eqs. (3) and (5)] The reflected path in Eq. (3) uses Φ_l while Eq. (5) uses Φ_i; the notation should be made consistent, and the role of the binary variable v_i in Eq. (3) should be clarified.
  2. [III-B1] The statement that 'the switching states of the N reflection units generate 2^N discrete configuration patterns' is unclear; with N elements and 2^b phase quantization levels, the number of patterns is (2^b)^N.
  3. [Abstract and IV-B2] The phrases 'maximize the received SINR' and 'achieves an optimal SINR' are used without qualification; given the heuristic nature of Algorithm 1, these claims should be softened (e.g., 'improved SINR') unless an optimality proof is provided.
  4. [IV-A and Conclusion] The paper does not report how the predicted trajectory errors propagate into the ON-OFF decisions; the conclusion acknowledges this as future work, but a sensitivity analysis or at least a quantitative statement of prediction error (e.g., the ±60 m mentioned in Section IV-A) would strengthen the link between the LSTM block and the SINR results.
  5. [IV-A] The mapping from the Geolife GPS coordinates to the distances d_u, d_i, and d_ui used in Eqs. (1) and (2) is not described; it is unclear how the user trajectories relate to the RIS and BS geometry.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ON-OFF decisions and SINR evaluation are self-consistent uses of the same model, not fitted predictions; the LSTM is trained on external trajectory data.

full rationale

The claimed derivation chain is not circular. The LSTM trajectory predictor is trained offline on the external Geolife dataset (Section IV-A, IV-B1) and its outputs are predicted coordinates, not SINR values; no parameter is fitted to the SINR it later reports. The codebook phase selection in Eq. (9) and the greedy per-RIS ON-OFF rule in Algorithm 1 use the same SINR expression (5) that defines the performance metric, but this is a design heuristic, not a tautology: the comparison gamma_l vs gamma'_l is an explicit optimization step, and the reported SINR is independently computed from (5) for TPC, always-on, and the ISL baseline. Self-citations among the authors (e.g., [3], [6], [7], [9], [10], [11], [17]) are background or baseline material and do not carry the derivation. The paper's claim of 'optimal SINR' (Section IV-B, Fig. 5) is unsupported because Eq. (9) maximizes only the desired-signal magnitude rather than the SINR of Eq. (5), and Algorithm 1 is a greedy comparison, but that is a correctness/optimality gap, not a circular reduction. No fitted input is renamed as a prediction; the only 'prediction' is the LSTM forecast of user coordinates from historical positions, which is externally trained and evaluated by Haversine error.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim depends on the accuracy of the LSTM predictor and on two unproven shortcuts: a per-RIS signal-power proxy for SINR maximization, and an angle-dependent reflection model used only in simulation. Neither is derived from first principles or benchmarked externally.

free parameters (3)
  • LSTM model weights and hyperparameters = trained on Geolife data, not disclosed
    The trajectory predictor is central to the control; its accuracy and settings are not reported.
  • Phase quantization bits b = not specified in simulations
    The codebook size determines the phase resolution in Eq. (9) and the complexity of the search, but b is never given.
  • Trajectory sliding window (64 segments of 8 points) = not further specified
    The input structure is described in Section III-A2, but stride, overlap, and normalization are not given.
assumptions (4)
  • domain assumption Each RIS serves exactly one user (U_L desired users and U_I interfering users)
    Stated in Section II-A; the optimization only considers one desired user l, ignoring that other desired users would also be affected by the on/off decisions.
  • domain assumption The BS obtains user location history via 3GPP positioning capabilities
    Section III-A1 invokes Release 18/19 standards to justify availability of trajectory data.
  • ad hoc to paper Reflected power from each RIS is inversely proportional to incident angle
    Introduced only in Section IV-B1 for simulation; not present in the product-distance path loss model of Eqs. (1)-(2), making the simulation inconsistent with the analytical model.
  • ad hoc to paper The phase selection via Eq. (9) solves Problem P2
    P2 maximizes SINR γ_l, but Eq. (9) maximizes only the desired signal power, ignoring interference; no equivalence is shown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamical ON-OFF Control with Trajectory Prediction for Multi-RIS Wireless Networks." pith.science (2026). https://pith.science/paper/MHKOY65N

@misc{pith2026250520887,
  author       = {Pith},
  title        = {Pith review of: Dynamical ON-OFF Control with Trajectory Prediction for Multi-RIS Wireless Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MHKOY65N}},
  note         = {Machine review of arXiv:2505.20887}
}
read the original abstract

Reconfigurable intelligent surfaces (RISs) have demonstrated an unparalleled ability to reconfigure wireless environments by dynamically controlling the phase, amplitude, and polarization of impinging waves. However, as nearly passive reflective metasurfaces, RISs may not distinguish between desired and interference signals, which can lead to severe spectrum pollution and even affect performance negatively. In particular, in large-scale networks, the signal-to-interference-plus-noise ratio (SINR) at the receiving node can be degraded due to excessive interference reflected from the RIS. To overcome this fundamental limitation, we propose in this paper a trajectory prediction-based dynamical control algorithm (TPC) for anticipating RIS ON-OFF states sequence, integrating a long-short-term-memory (LSTM) scheme to predict user trajectories. In particular, through a codebook-based algorithm, the RIS controller adaptively coordinates the configuration of the RIS elements to maximize the received SINR. Our simulation results demonstrate the superiority of the proposed TPC method over various system settings.

Figures

Figures reproduced from arXiv: 2505.20887 by the authors.

Figure 1
Figure 1. In the left figure, we show a conventional multi-user uplink RIS-assisted wireless communication system without TPC. Some interfering signals and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The proposed TPC algorithm structure. The input includes the history points of the pedestrian trajectories, and the output is the predicted coordinates. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison of real trajectory and predicted trajectory [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: SINR comparison across transmission power [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: SINR comparison across numbers of RIS elements [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

19 extracted references · 17 canonical work pages

  1. [1]

    Reconfigurable intelligent surfaces for 6G: Emerging hardware architectures, applications, and open challenges,

    E. Basar, G. C. Alexandropoulos, Y . Liu, Q. Wu, S. Jin, C. Yuen, O. A. Dobre, and R. Schober, “Reconfigurable intelligent surfaces for 6G: Emerging hardware architectures, applications, and open challenges,” IEEE Veh. Technol. Mag., vol. 19, no. 3, pp. 27–47, 2024

    2024

  2. [2]

    Intelligent surfaces empowered wireless network: Recent advances and the road to 6G,

    Q. Wu, B. Zheng, C. You, L. Zhu, K. Shen, X. Shao, W. Mei, B. Di, H. Zhang, E. Basar, L. Song, M. Di Renzo, Z.-Q. Luo, and R. Zhang, “Intelligent surfaces empowered wireless network: Recent advances and the road to 6G,”Proc. IEEE, vol. 112, no. 7, pp. 724–763, 2024

    work page 2024

  3. [3]

    Reconfigurable intelligent surface-assisted aerial-terrestrial communications via multi-task learning,

    X. Cao, B. Yang, C. Huang, C. Yuen, M. D. Renzo, D. Niyato, and Z. Han, “Reconfigurable intelligent surface-assisted aerial-terrestrial communications via multi-task learning,”IEEE Journal on Selected Areas in Communications, vol. 39, no. 10, pp. 3035–3050, 2021

    work page 2021

  4. [4]

    Reconfigurable intelligent computational surfaces for mec-assisted autonomous driving networks: Design opti- mization and analysis,

    X. Zhang, B. Yang, Z. Yu, X. Cao, G. C. Alexandropoulos, Y . Zhang, M. Debbah, and C. Yuen, “Reconfigurable intelligent computational surfaces for mec-assisted autonomous driving networks: Design opti- mization and analysis,”IEEE Transactions on Intelligent Transportation Systems, vol. 26, no. 1, pp. 1286–1303, 2025

    work page 2025

  5. [5]

    Covering Underwater Shadow Zones using Acoustic Reconfigurable Intelligent Surfaces

    L. Zhao, J. Tan, J. Wang, I. F. Akyildiz, and Z. Sun, “Covering underwater shadow zones using acoustic reconfigurable intelligent surfaces,” 2025. [Online]. Available: https://arxiv.org/abs/2501.02256

    work page Pith review arXiv 2025

  6. [6]

    Reconfigurable intelligent surface system level simulations for industry standards,

    Y . Yuan, Y . Huang, X. Su, B. Duan, N. Hu, and M. D. Renzo, “Reconfigurable intelligent surface system level simulations for industry standards,”IEEE Communications Magazine, pp. 1–7, 2025

    work page 2025

  7. [7]

    Effects of realistic reradiation models in digital reconfigurable intelligent surfaces,

    M. Di Renzo, A. Mohamed, A. Zappone, G. Gradoni, M. Rossi, M. Moc- cia, G. Castaldi, and V . Galdi, “Effects of realistic reradiation models in digital reconfigurable intelligent surfaces,”Journal of Information and Intelligence, vol. 1, no. 3, pp. 238–252, 2023, special Issue: Information Metasurfaces and Reconfigurable Intelligent Surfaces

    work page 2023

  8. [8]

    Interference nulling using reconfigurable intelligent surface,

    T. Jiang and W. Yu, “Interference nulling using reconfigurable intelligent surface,”IEEE Journal on Selected Areas in Communications, vol. 40, no. 5, pp. 1392–1406, 2022

    work page 2022

Show all 19 references
  1. [9]

    Pervasive machine learning for smart radio en- vironments enabled by reconfigurable intelligent surfaces,

    G. C. Alexandropoulos, K. Stylianopoulos, C. Huang, C. Yuen, M. Ben- nis, and M. Debbah, “Pervasive machine learning for smart radio en- vironments enabled by reconfigurable intelligent surfaces,”Proc. IEEE, vol. 110, no. 9, pp. 1494–1525, 2022

    2022

  2. [10]

    MIMO MAC empowered by reconfigurable intelligent surfaces: Capacity region and large system analysis,

    A. L. Moustakas and G. C. Alexandropoulos, “MIMO MAC empowered by reconfigurable intelligent surfaces: Capacity region and large system analysis,”IEEE Trans. Wireless Commun., vol. 23, no. 12, pp. 19 245– 19 258, 2024

    2024

  3. [11]

    Intelligent spectrum learning for wireless networks with reconfig- urable intelligent surfaces,

    B. Yang, X. Cao, C. Huang, C. Yuen, L. Qian, and M. D. Renzo, “Intelligent spectrum learning for wireless networks with reconfig- urable intelligent surfaces,”IEEE Transactions on Vehicular Technology, vol. 70, no. 4, pp. 3920–3925, 2021

    2021

  4. [12]

    On discrete phase shifts optimization of RIS-aided FD systems: Are all RIS elements needed?

    A. Faisal, I. Al-Nahhal, O. A. Dobre, and T. M. N. Ngatched, “On discrete phase shifts optimization of RIS-aided FD systems: Are all RIS elements needed?” inICC 2023 - IEEE International Conference on Communications, 2023, pp. 1382–1387

    2023

  5. [13]

    To reflect or not to reflect: On–off control and number configuration for reflecting elements in RIS-aided wireless systems,

    H. Xie and D. Li, “To reflect or not to reflect: On–off control and number configuration for reflecting elements in RIS-aided wireless systems,” IEEE Transactions on Communications, vol. 71, no. 12, pp. 7409–7424, 2023

    2023

  6. [14]

    Lstm- based energy-efficient wireless communication with reconfigurable in- telligent surfaces,

    K. D. Gupta, R. Nigam, D. K. Sharma, and S. K. Dhurandher, “Lstm- based energy-efficient wireless communication with reconfigurable in- telligent surfaces,”IEEE Transactions on Green Communications and Networking, vol. 6, no. 2, pp. 704–712, 2022

    2022

  7. [15]

    3GPP release 18 overview: A world of 5G- advanced,

    W. Chen and P. Jain, “3GPP release 18 overview: A world of 5G- advanced,”Atis. Org, ATIS, vol. 2, 2023

    2023

  8. [16]

    Nwdaf in 3GPP 5G advanced: A survey,

    X. He, Z. Yang, Y . Xiang, and S. Qian, “Nwdaf in 3GPP 5G advanced: A survey,” inProc. EIECS), Changchun, China, 2023

    2023

  9. [17]

    Codebook-based solutions for reconfigurable intelligent surfaces and their open challenges,

    J. An, C. Xu, Q. Wu, D. W. K. Ng, M. Di Renzo, C. Yuen, and L. Hanzo, “Codebook-based solutions for reconfigurable intelligent surfaces and their open challenges,”IEEE Wireless Commun., vol. 31, no. 2, pp. 134–141, 2024

    2024

  10. [18]

    Coded beam training for RIS assisted wireless communications,

    Y . Chen and L. Dai, “Coded beam training for RIS assisted wireless communications,”IEEE Transactions on Wireless Communications, pp. 1–1, 2025

    2025

  11. [19]

    Geolife: A collaborative social networking service among user, location and trajectory

    Y . Zheng, X. Xie, W.-Y . Maet al., “Geolife: A collaborative social networking service among user, location and trajectory.”IEEE Data Eng. Bull., vol. 33, no. 2, pp. 32–39, 2010

    2010

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.