pith. sign in

arxiv: 2510.05438 · v5 · submitted 2025-10-06 · 📡 eess.SP

Model-based Deep Learning for Joint RIS Phase Shift Compression and WMMSE Beamforming

Alexander James Fernandes , Ioannis Psaromiligkos This is my paper

Pith reviewed 2026-05-18 08:42 UTC · model grok-4.3

classification 📡 eess.SP
keywords reconfigurable intelligent surfacephase shift compressionWMMSE beamformingmodel-based deep learningmulti-user communicationssum-rate optimizationend-to-end trainingRIS control overhead
0
0 comments X

The pith

A model-based deep learning network that unrolls WMMSE and updates beamformers with actual decompressed RIS phases sustains high sum rates even when control bits are fewer than the number of elements.

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

The paper develops a deep learning architecture for RIS-assisted multi-user systems that compresses the computed phase shifts into a short binary message for transmission to the RIS controller. It embeds the iterative weighted minimum mean square error algorithm inside the network so that beamforming is recomputed using the decompressed phases that the RIS will actually apply. End-to-end training therefore learns to compensate for the quantization errors that would otherwise degrade performance. Simulations indicate that this joint approach yields higher sum rates than conventional separate compression and beamforming steps. The result matters for large RIS deployments because transmitting full phase information to every element would otherwise consume prohibitive control bandwidth.

Core claim

The central claim is that unrolling the WMMSE algorithm into a communication-informed neural network and training it end-to-end for joint phase-shift compression and compression-aware beamforming produces substantially higher sum rates than methods that ignore compression during beamformer design, and that these gains persist even when the number of control bits is smaller than the number of RIS elements.

What carries the argument

Unrolled WMMSE iterations inside a deep network that recomputes beamformers from the decompressed RIS phases rather than the ideal phases.

If this is right

  • Beamformer mismatches caused by phase compression errors are reduced because the network sees the actual applied phases during training.
  • Sum-rate gains remain observable across a range of bit budgets and user counts in the evaluated scenarios.
  • The approach enables RIS control with limited feedback bandwidth without requiring perfect phase information at the controller.
  • Joint optimization of compression and beamforming becomes feasible through the differentiable unrolled structure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same unrolling-plus-awareness pattern could be applied to other iterative wireless algorithms that face quantization or hardware constraints.
  • Real-world performance would likely require retraining or fine-tuning on measured channel data rather than purely synthetic distributions.
  • Designers of future RIS hardware might trade element count or phase resolution against the availability of joint optimization pipelines instead of insisting on high-bit control links.

Load-bearing premise

The simulated channels and noise models used for training match the statistical conditions of the eventual deployment environment well enough for the learned solution to generalize.

What would settle it

A hardware experiment that measures sum rate on real propagation channels and shows no improvement over separate compression-plus-beamforming when the bit budget is set below the number of RIS elements.

Figures

Figures reproduced from arXiv: 2510.05438 by Alexander James Fernandes, Ioannis Psaromiligkos.

Figure 1
Figure 1. Figure 1: RIS-assisted communication system. architecture that compresses the phase shift information into a single binary control message. Compared to pre￾vious works [2], [3], [4], [5], [6], instead of individually transmitting phase shifts for each RIS element from the AP to the RIS controller, the proposed architecture takes advantage of the correlations between the phase shifts by extracting and combining relev… view at source ↗
Figure 2
Figure 2. Figure 2: DL architecture of the AQE-WMMSE network. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: RIS phase shift control message fixed to [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training/Validation loss for B = 40 control bits. D. Results In [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
read the original abstract

A model-based deep learning (DL) architecture is proposed for reconfigurable intelligent surface (RIS)-assisted multi-user communications to reduce the number of bits required for transmitting phase shift information from the access point (AP) to the RIS controller. The AP computes the phase shifts and compresses them into a binary control message that is sent to the RIS controller for element configuration. To help reduce beamformer mismatches caused by phase shift compression errors, the beamformer is updated with the actual (decompressed) RIS phase shifts. By unrolling the iterative weighted minimum mean square error (WMMSE) algorithm within the wireless communication-informed DL architecture, joint phase shift compression and WMMSE beamforming can be trained end-to-end. Simulation results demonstrate that incorporating compression-aware beamforming significantly improves sum-rate performance, even when the number of control bits is lower than the number of RIS elements.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a model-based deep learning architecture for RIS-assisted multi-user communications that jointly performs phase-shift compression at the AP and WMMSE beamforming. Phase shifts are compressed into a low-bit binary message sent to the RIS controller; the beamformer is then updated using the decompressed phases to reduce mismatch. The iterative WMMSE algorithm is unrolled into a neural network and trained end-to-end. Simulations are reported to show sum-rate gains even when the number of control bits is smaller than the number of RIS elements.

Significance. If the reported sum-rate improvements are robust, the work addresses a practical control-overhead bottleneck in RIS systems and shows that compression-aware beamforming can preserve performance with reduced signaling. The unrolled-WMMSE structure is a methodological strength that embeds domain knowledge and may improve generalization relative to purely data-driven alternatives.

major comments (2)
  1. [Simulation Results] Simulation section: the abstract and available description report performance gains but provide no information on the underlying channel model (Rician factor, spatial correlation, user geometry), training distribution, number of Monte-Carlo realizations, or statistical significance testing. Without these details the central claim that compression-aware beamforming yields reliable improvements cannot be verified.
  2. [Simulation Results] Generalization discussion: all reported results appear to be generated from the same channel distribution used for training. No cross-distribution experiments (different Rician factors, correlation structures, or user placements) or hardware-impairment injection are described. This assumption is load-bearing for the claim that the learned mapping remains effective under deployment conditions.
minor comments (2)
  1. [Abstract] The abstract uses the acronym WMMSE without spelling it out on first use; ensure the full term appears at its first occurrence in the main text as well.
  2. [Figures] Figure captions and axis labels should explicitly state the number of RIS elements, users, and the bit budgets used in each curve to allow direct comparison with the textual claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the simulation setup and generalization. We address each point below and have revised the manuscript to strengthen the presentation of results.

read point-by-point responses

  1. Referee: [Simulation Results] Simulation section: the abstract and available description report performance gains but provide no information on the underlying channel model (Rician factor, spatial correlation, user geometry), training distribution, number of Monte-Carlo realizations, or statistical significance testing. Without these details the central claim that compression-aware beamforming yields reliable improvements cannot be verified.

    Authors: We agree that these parameters are required for reproducibility and verification of the claims. The revised manuscript expands the Simulation Results section with the following details: Rician factor K=10 dB, exponential spatial correlation with coefficient 0.5, users placed uniformly in a 200 m x 200 m square centered on the AP, training distribution identical to the test distribution with 10,000 samples, 1,000 Monte-Carlo realizations, and 95% confidence intervals together with paired t-test p-values (all <0.01) confirming the statistical significance of the reported sum-rate gains. revision: yes

  2. Referee: [Simulation Results] Generalization discussion: all reported results appear to be generated from the same channel distribution used for training. No cross-distribution experiments (different Rician factors, correlation structures, or user placements) or hardware-impairment injection are described. This assumption is load-bearing for the claim that the learned mapping remains effective under deployment conditions.

    Authors: We acknowledge the value of cross-distribution testing. The original evaluation focused on matched conditions to isolate the benefit of joint compression-aware beamforming. The model-based unrolling of WMMSE embeds domain knowledge that is expected to aid robustness. In the revision we have added results for mismatched Rician factors (K=5 dB and K=20 dB) and altered user geometries, showing that sum-rate gains over the baselines are retained, although reduced. Hardware impairments lie outside the scope of the present study, which assumes ideal phase control. revision: partial

Circularity Check

0 steps flagged

No significant circularity; standard unrolled WMMSE training remains self-contained

full rationale

The paper describes a model-based DL architecture that unrolls the iterative WMMSE algorithm to jointly optimize RIS phase compression and beamforming, with end-to-end training on simulated channels. This follows conventional model-based deep learning practices where the network layers mirror known iterative steps without redefining any output as an input by construction. No load-bearing self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work are evident in the provided description; the reported sum-rate gains are presented as simulation outcomes under the training distribution rather than tautological reductions. The derivation chain is therefore independent and self-contained against external benchmarks such as standard WMMSE.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper relies on standard assumptions of wireless channel models and perfect knowledge of decompressed phase shifts at the transmitter for beamformer update. No new entities are introduced.

axioms (2)
  • domain assumption The wireless channel follows standard models used in RIS literature (e.g., Rician or Rayleigh fading).
    Invoked implicitly to generate simulation results.
  • domain assumption The RIS controller can perfectly apply the decompressed phase shifts.
    Required for the compression-aware beamforming update step.

pith-pipeline@v0.9.0 · 5682 in / 1350 out tokens · 22960 ms · 2026-05-18T08:42:58.452718+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

  1. [1]

    Intellige nt Reflecting Surface-Aided Wireless Communications: A Tutor ial,

    Q. Wu, S. Zhang, B. Zheng, C. Y ou, and R. Zhang, “Intellige nt Reflecting Surface-Aided Wireless Communications: A Tutor ial,” IEEE Trans. Commun., vol. 69, no. 5, pp. 3313–3351, May. 2021

    work page 2021

  2. [2]

    Intelligent Reflecting Surface Enhan ced Wireless Network via Joint Active and Passive Beamforming,

    Q. Wu and R. Zhang, “Intelligent Reflecting Surface Enhan ced Wireless Network via Joint Active and Passive Beamforming,” IEEE Trans. Wirel. Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019

    work page 2019

  3. [3]

    Weighted Sum-Rate Maximization for Reconfigurable Intelligent Surface Aided Wireless Networks,

    H. Guo, Y .-C. Liang, J. Chen, and E. G. Larsson, “Weighted Sum-Rate Maximization for Reconfigurable Intelligent Surface Aided Wireless Networks,” IEEE Trans. Wirel. Commun., vol. 19, no. 5, pp. 3064–3076, May. 2020

    work page 2020

  4. [4]

    Low- Complexity Joint Beamforming for RIS-Assisted MU-MISO Systems Based o n Model-Driven Deep Learning,

    W. Jin, J. Zhang, C.-K. Wen, S. Jin, X. Li, and S. Han, “Low- Complexity Joint Beamforming for RIS-Assisted MU-MISO Systems Based o n Model-Driven Deep Learning,” IEEE Trans. Wirel. Commun. , vol. 23, no. 7, pp. 6968–6982, Jul. 2024

    work page 2024

  5. [5]

    A Robust Deep Learning-Based Beam- forming Design for RIS-Assisted Multiuser MISO Communicat ions With Practical Constraints,

    W. Xu, L. Gan, and C. Huang, “A Robust Deep Learning-Based Beam- forming Design for RIS-Assisted Multiuser MISO Communicat ions With Practical Constraints,” IEEE Trans. on Cogn. Commun. Netw. , vol. 8, no. 2, pp. 694–706, Jun. 2022

    work page 2022

  6. [6]

    Deep Learning- Based Adaptive Phase Shift Compression and Feedback in IRS- Assisted Communication Systems,

    Z. Li, H. Shen, W. Xu, D. Chen, and C. Zhao, “Deep Learning- Based Adaptive Phase Shift Compression and Feedback in IRS- Assisted Communication Systems,” IEEE Wirel. Commun. Lett. , vol. 13, no. 3, pp. 766–770, 2024

    work page 2024

  7. [7]

    Re- configurable Intelligent Surface-Aided Full-Duplex mmWav e MIMO: Channel Estimation, Passive and Hybrid Beamforming,

    S. Y ang, W. Lyu, Y . Xanthos, Z. Zhang, C. Assi, and C. Y uen, “Re- configurable Intelligent Surface-Aided Full-Duplex mmWav e MIMO: Channel Estimation, Passive and Hybrid Beamforming,” IEEE Trans. Wirel. Commun., vol. 23, no. 4, pp. 2575–2590, Apr. 2024

    work page 2024

  8. [8]

    Channel Estimati on for Reconfig- urable Intelligent Surface-Assisted Full-Duplex MIMO Wit h Hardware Impairments,

    A. J. Fernandes and I. Psaromiligkos, “Channel Estimati on for Reconfig- urable Intelligent Surface-Assisted Full-Duplex MIMO Wit h Hardware Impairments,” IEEE Wirel. Commun. Lett. , vol. 12, no. 10, pp. 1697– 1701, Oct. 2023

    work page 2023

  9. [9]

    Joint Estimation of Direct and RIS-assisted Channe ls with Tensor Signal Modelling,

    ——, “Joint Estimation of Direct and RIS-assisted Channe ls with Tensor Signal Modelling,” in IEEE V eh. Technol. Conf. (VTC) . Washington DC, USA: IEEE, Oct. 2024, pp. 1–6

    work page 2024

  10. [10]

    Tensor Signal Modeling and Channel Estimation for Reconfig- urable Intelligent Surface-Assisted Full-Duplex MIMO,

    ——, “Tensor Signal Modeling and Channel Estimation for Reconfig- urable Intelligent Surface-Assisted Full-Duplex MIMO,” IEEE Open J. Commun. Soc. , vol. 5, no. November, pp. 7668–7684, 2024

    work page 2024

  11. [11]

    Reconfigu rable Intelligent Surface Assisted Multi-User Communications: How Many Reflective Elements Do We Need?

    H. Zhang, B. Di, Z. Han, H. V . Poor, and L. Song, “Reconfigu rable Intelligent Surface Assisted Multi-User Communications: How Many Reflective Elements Do We Need?” IEEE Wirel. Commun. Lett. , vol. 10, no. 5, pp. 1098–1102, May. 2021

    work page 2021

  12. [12]

    Algorithm Unrolling: I nterpretable, Efficient Deep Learning for Signal and Image Processing,

    V . Monga, Y . Li, and Y . C. Eldar, “Algorithm Unrolling: I nterpretable, Efficient Deep Learning for Signal and Image Processing,” IEEE Signal Process. Mag., vol. 38, no. 2, pp. 18–44, Mar. 2021

    work page 2021

  13. [13]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, and A. Courville, Deep Learning . MIT Press, 2016. [Online]. Available: https://www.deeplearn ingbook.org/

    work page 2016

  14. [14]

    Hard ware-Limited Task-Based Quantization,

    N. Shlezinger, Y . C. Eldar, and M. R. D. Rodrigues, “Hard ware-Limited Task-Based Quantization,” IEEE Trans. Signal Process. , vol. 67, no. 20, pp. 5223–5238, Oct. 2019

    work page 2019

  15. [15]

    Deep Quantization for MIMO Channel Estimation,

    M. Shohat, G. Tsintsadze, N. Shlezinger, and Y . C. Eldar , “Deep Quantization for MIMO Channel Estimation,” in IEEE Int. Conf. on Acoust., Speech and Signal Process. (ICASSP) , vol. 2019-May. IEEE, May. 2019, pp. 3912–3916

    work page 2019

  16. [16]

    An Iterativ ely Weighted MMSE Approach to Distributed Sum-Utility Maximization for a MIMO Interfering Broadcast Channel,

    Q. Shi, M. Razaviyayn, Z.-Q. Luo, and C. He, “An Iterativ ely Weighted MMSE Approach to Distributed Sum-Utility Maximization for a MIMO Interfering Broadcast Channel,” IEEE Trans. Signal Process. , vol. 59, no. 9, pp. 4331–4340, Sep. 2011

    work page 2011