pith. sign in

arxiv: 2604.13680 · v1 · submitted 2026-04-15 · 💻 cs.IT · math.IT

Scalable Design for RIS-Assisted Multi-User Downlink System Empowered by RSMA under Partial CSI

Yifan Fang , Bile Peng , Yingyang Chen , Qiang Li , Marwa Chafii , Eduard A. Jorswieck This is my paper

Pith reviewed 2026-05-10 12:24 UTC · model grok-4.3

classification 💻 cs.IT math.IT
keywords reconfigurable intelligent surfacesrate splitting multiple accesspartial channel state informationunsupervised neural networkdownlink precodingchannel estimation
0
0 comments X

The pith

A neural network infers complete channel information from limited observations in reconfigurable intelligent surface systems, enabling rate-splitting to preserve performance under uncertainty.

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

The paper aims to address the difficulty of obtaining accurate channel state information in large reconfigurable intelligent surface systems by using only a few anchor elements for partial CSI. It develops an unsupervised neural network called RISnet to estimate full CSI and combines it with a rate-splitting multiple access precoder for multi-user downlink transmission. This combination is designed to improve scalability and robustness. Simulations under ray-tracing channels show the scheme nearly matches full-CSI performance, with RSMA helping to limit degradation when uncertainty rises.

Core claim

By training RISnet unsupervised on partial CSI to reconstruct effective channels and then applying RSMA precoding, the system achieves downlink rates close to those with perfect CSI knowledge, while becoming more tolerant to increases in channel estimation error during the training phase.

What carries the argument

RISnet, which takes partial CSI to form effective channel features, expands them, and updates new anchors with channel information to infer the full CSI.

If this is right

  • Large-scale RIS deployments become feasible without exhaustive channel estimation.
  • Performance degradation from partial CSI is minimized through the combination of inference and rate splitting.
  • The low-complexity precoder supports real-time operation in multi-user scenarios.
  • Robustness improves specifically as training-time uncertainty grows.

Where Pith is reading between the lines

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

  • The technique might extend to mobile users or time-varying channels by adapting the network training.
  • It could reduce the need for frequent full CSI feedback in practical networks.
  • Testing against stochastic channel models would reveal if the gains hold beyond deterministic assumptions.

Load-bearing premise

That the deterministic ray-tracing channel model used in evaluations matches actual wireless propagation conditions and that the unsupervised RISnet does not overfit to the specific partial CSI patterns seen in training.

What would settle it

A measurement campaign in a real indoor or outdoor environment showing that the rate gap to full-CSI baselines exceeds the simulated approximation or that RSMA fails to reduce the loss under higher uncertainty.

Figures

Figures reproduced from arXiv: 2604.13680 by Bile Peng, Eduard A. Jorswieck, Marwa Chafii, Qiang Li, Yifan Fang, Yingyang Chen.

Figure 1
Figure 1. Figure 1: The RISnet architecture with channel feature [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The training results of RISnet with SDMA and RSMA under random [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The training and testing results of RISnet with RSMA under RT [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
read the original abstract

In large-scale reconfigurable intelligent surface (RIS) communication systems, the precise acquisition of channel state information (CSI) is challenging. Consider a practical RIS configuration where only a few reflective elements serve as anchors to estimate CSI, which are termed partial CSI. To improve the robustness against partial CSI and the scalability of RIS networks, this paper proposes an unsupervised learning-based rate-splitting multiple access (RSMA) scheme for RIS-assisted multi-user systems. Specifically, RISnet, a neural network architecture designed to infer full CSI from partial observations, is employed and integrated with a low-complexity RSMA precoder. Effective channel features are constituted from partial CSI, and the original elements with channel information contribute to new anchors after expansion in RISnet. Numerical results demonstrate that the proposed scheme approximates the performance with a full CSI of RIS under deterministic raytracing channel conditions. When channel uncertainty increases during training, RSMA has been shown to enhance RISnet robustness, significantly mitigating performance loss.

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 scalable design for RIS-assisted multi-user downlink systems using an unsupervised learning-based RSMA scheme under partial CSI. RISnet, a neural network, infers full CSI from partial observations at anchor elements and is integrated with a low-complexity RSMA precoder. Effective channel features are derived from partial CSI, with original elements contributing to new anchors after expansion. Numerical results under deterministic ray-tracing conditions show approximation to full-CSI performance, with RSMA enhancing robustness against increasing channel uncertainty during training.

Significance. If the central claims hold, the paper provides a promising approach for practical deployment of large-scale RIS systems where full CSI is hard to obtain. The combination of unsupervised channel inference via RISnet and RSMA's interference management offers potential for improved scalability and robustness. The unsupervised aspect avoids the need for labeled data, which is a strength in this context.

major comments (2)
  1. [Abstract and Numerical Results] Abstract and Numerical Results: The headline claim that the scheme approximates full-CSI performance and that RSMA significantly mitigates performance loss under rising channel uncertainty rests exclusively on simulations inside a fixed deterministic ray-tracing geometry. The partial-observation statistics, anchor-to-element mapping, and unsupervised loss landscape may not remain representative when the underlying propagation includes stochastic elements (random phases, diffuse scattering, small-scale fading). An ablation on stochastic channel generators or out-of-distribution anchor placements is required to establish that the reported approximation gap and RSMA benefit are intrinsic rather than artifacts of the evaluation environment.
  2. [Abstract] Abstract: No quantitative metrics (sum-rate values, approximation gaps, or robustness curves) are supplied to support the stated performance approximation and robustness gains, nor are architecture details, training procedure, or ablation studies on anchor count/placement referenced. This leaves the central empirical claim without visible supporting evidence in the provided description.
minor comments (2)
  1. The manuscript should explicitly state the number of anchor elements, their placement strategy, and how the unsupervised loss is formulated to allow reproducibility.
  2. Figure captions and legends (if present) would benefit from clearer indication of all baselines compared (e.g., ZF, other ML precoders) and the exact channel uncertainty levels tested.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below, providing clarifications on the evaluation scope and committing to improvements in the abstract and discussion sections where feasible.

read point-by-point responses

  1. Referee: [Abstract and Numerical Results] Abstract and Numerical Results: The headline claim that the scheme approximates full-CSI performance and that RSMA significantly mitigates performance loss under rising channel uncertainty rests exclusively on simulations inside a fixed deterministic ray-tracing geometry. The partial-observation statistics, anchor-to-element mapping, and unsupervised loss landscape may not remain representative when the underlying propagation includes stochastic elements (random phases, diffuse scattering, small-scale fading). An ablation on stochastic channel generators or out-of-distribution anchor placements is required to establish that the reported approximation gap and RSMA benefit are intrinsic rather than artifacts of the evaluation environment.

    Authors: We acknowledge the referee's point that our results are obtained under deterministic ray-tracing channels. This modeling choice is deliberate, as it captures realistic large-scale propagation effects in a controlled manner and is widely used in RIS literature to evaluate partial-CSI scenarios without additional small-scale fading variability. The unsupervised RISnet is trained to infer full CSI from partial anchor observations specifically in this setting, and the RSMA integration is shown to improve robustness as training uncertainty grows. We agree that stochastic extensions would be valuable for broader claims; however, performing new ablations on stochastic generators or out-of-distribution anchors is beyond the scope of the current work due to computational demands. We will revise the manuscript to more explicitly qualify the evaluation environment and add a limitations/future-work paragraph discussing stochastic channels. revision: partial

  2. Referee: [Abstract] Abstract: No quantitative metrics (sum-rate values, approximation gaps, or robustness curves) are supplied to support the stated performance approximation and robustness gains, nor are architecture details, training procedure, or ablation studies on anchor count/placement referenced. This leaves the central empirical claim without visible supporting evidence in the provided description.

    Authors: We agree that the abstract would benefit from more concrete support. In the revised manuscript, we will incorporate key quantitative indicators (e.g., sum-rate approximation gaps relative to full-CSI baselines and robustness improvements under increasing uncertainty) while maintaining the abstract's brevity. We will also briefly reference the RISnet architecture, the unsupervised training procedure, and the role of anchor expansion. Full details, including any anchor-count studies, remain in Sections III and IV of the paper. revision: yes

standing simulated objections not resolved
  • Requirement for ablation studies on stochastic channel generators or out-of-distribution anchor placements, as these would necessitate new extensive simulations not included in the current manuscript.

Circularity Check

0 steps flagged

No significant circularity; claims rest on external numerical validation

full rationale

The provided manuscript text contains no equations, derivations, or self-citations that reduce the central claims (RISnet inference of full CSI from partial observations, RSMA robustness under uncertainty) to fitted inputs or prior author results by construction. The performance approximation and robustness statements are presented as outcomes of numerical simulations on a deterministic ray-tracing model, which constitutes an independent evaluation step rather than a tautological re-expression of the model inputs. No load-bearing uniqueness theorems, ansatzes, or renamings are invoked in the abstract or reader-supplied excerpts.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Without the full manuscript, no explicit free parameters, axioms, or invented entities can be extracted. The approach implicitly relies on standard assumptions of neural network generalization and deterministic ray-tracing channel fidelity.

pith-pipeline@v0.9.0 · 5487 in / 1144 out tokens · 31221 ms · 2026-05-10T12:24:19.130912+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

  1. [1]

    RIS-IoE for data-driven networks: New mentalities, trends and preliminary solutions,

    B. Zhuoet al., “RIS-IoE for data-driven networks: New mentalities, trends and preliminary solutions,”IEEE Internet Things Mag., vol. 6, no. 2, pp. 102–107, 2023

    work page 2023

  2. [2]

    Deep learning-based CSI feedback for RIS-aided massive MIMO systems with time correlation,

    Z. Penget al., “Deep learning-based CSI feedback for RIS-aided massive MIMO systems with time correlation,”IEEE Wireless Commun. Lett., vol. 13, no. 8, pp. 2060–2064, 2024

    work page 2060

  3. [3]

    Channel estimation for intelligent reflecting surface as- sisted multiuser communications: Framework, algorithms, and analysis,

    Z. Wanget al., “Channel estimation for intelligent reflecting surface as- sisted multiuser communications: Framework, algorithms, and analysis,” IEEE Trans. Wireless Commun., vol. 19, no. 10, pp. 6607–6620, 2020

    work page 2020

  4. [4]

    Rate splitting for MIMO wireless networks: A promising PHY-layer strategy for LTE evolution,

    B. Clerckxet al., “Rate splitting for MIMO wireless networks: A promising PHY-layer strategy for LTE evolution,”IEEE Commun. Mag., vol. 54, no. 5, pp. 98–105, 2016

    work page 2016

  5. [5]

    Rate-splitting multiple access for downlink communi- cation systems: Bridging, generalizing, and outperforming SDMA and NOMA,

    M. Yijieet al., “Rate-splitting multiple access for downlink communi- cation systems: Bridging, generalizing, and outperforming SDMA and NOMA,”EURASIP J Wireless Com Network, vol. 2018, no. 1, 2018

    work page 2018

  6. [6]

    Resource allocation design for IRS-aided downlink MU- MISO RSMA systems,

    H. Fuet al., “Resource allocation design for IRS-aided downlink MU- MISO RSMA systems,” in2021 IEEE ICC Workshops, 2021, pp. 1–6

    work page 2021

  7. [7]

    Rate-splitting multiple access for UA V-based RIS- enabled interference-limited vehicular communication system,

    A. Bansalet al., “Rate-splitting multiple access for UA V-based RIS- enabled interference-limited vehicular communication system,”IEEE Trans. Intell. V eh., vol. 8, no. 1, pp. 936–948, 2023

    work page 2023

  8. [8]

    Reconfigurable intelligent surfaces empowered cooper- ative rate splitting with user relaying,

    K. Zhaoet al., “Reconfigurable intelligent surfaces empowered cooper- ative rate splitting with user relaying,” in2022 ISWCS, 2022, pp. 1–6

    work page 2022

  9. [9]

    Machine learning-based reconfigurable intelligent surface-aided MIMO systems,

    N. T. Nguyenet al., “Machine learning-based reconfigurable intelligent surface-aided MIMO systems,” in2021 IEEE 22nd SPA WC, 2021, pp. 101–105

    work page 2021

  10. [10]

    Learning to reflect and to beamform for intelligent reflecting surface with implicit channel estimation,

    T. Jianget al., “Learning to reflect and to beamform for intelligent reflecting surface with implicit channel estimation,”IEEE J. Sel. Areas Commun., vol. 39, no. 7, pp. 1931–1945, 2021

    work page 1931

  11. [11]

    Deep reinforcement learning based intelligent reflecting surface optimization for MISO communication systems,

    K. Fenget al., “Deep reinforcement learning based intelligent reflecting surface optimization for MISO communication systems,”IEEE Wireless Commun. Lett., vol. 9, no. 5, pp. 745–749, 2020

    work page 2020

  12. [12]

    RISnet: A scalable approach for reconfigurable intel- ligent surface optimization with partial CSI,

    B. Penget al., “RISnet: A scalable approach for reconfigurable intel- ligent surface optimization with partial CSI,” inGLOBECOM 2023 - 2023 IEEE Global Communications Conference, 2023, pp. 4810–4816

    work page 2023

  13. [13]

    Autonomous driving with RSMA-enabled finite block- length transmissions: Ergodic performance analysis and optimization,

    Y . Wanget al., “Autonomous driving with RSMA-enabled finite block- length transmissions: Ergodic performance analysis and optimization,” IEEE Trans. Wireless Commun., vol. 25, pp. 2954–2969, 2026

    work page 2026

  14. [14]

    Rate-splitting multiple access with finite blocklength and high mobility for URLLC transmissions,

    J. Zhuet al., “Rate-splitting multiple access with finite blocklength and high mobility for URLLC transmissions,”IEEE Wireless Commun. Lett., vol. 13, no. 5, pp. 1518–1522, 2024

    work page 2024

  15. [15]

    Beamforming optimization for hybrid active-passive ris assisted wireless communications: A rate-maximization perspective,

    Y . Juet al., “Beamforming optimization for hybrid active-passive ris assisted wireless communications: A rate-maximization perspective,” IEEE Trans. Commun., vol. 72, no. 9, pp. 5428–5442, 2024

    work page 2024

  16. [16]

    Rate-splitting multiple access to mitigate the curse of mobility in (massive) MIMO networks,

    O. Dizdaret al., “Rate-splitting multiple access to mitigate the curse of mobility in (massive) MIMO networks,”IEEE Trans. Commun., vol. 69, no. 10, pp. 6765–6780, 2021

    work page 2021

  17. [17]

    An iteratively weighted MMSE approach to distributed sum-utility maximization for a MIMO interfering broadcast channel,

    Q. Shiet al., “An iteratively weighted MMSE approach to distributed sum-utility maximization for a MIMO interfering broadcast channel,” in 2011 IEEE ICASSP, 2011, pp. 3060–3063

    work page 2011

  18. [18]

    RISnet: A domain-knowledge driven neural network architecture for RIS optimization with mutual coupling and partial CSI,

    B. Penget al., “RISnet: A domain-knowledge driven neural network architecture for RIS optimization with mutual coupling and partial CSI,” IEEE Trans. Wireless Commun., vol. 24, no. 5, pp. 4469–4482, 2025

    work page 2025

  19. [19]

    DeepMIMO: A generic deep learning dataset for mil- limeter wave and massive MIMO applications,

    A. Alkhateeb, “DeepMIMO: A generic deep learning dataset for mil- limeter wave and massive MIMO applications,” inProc. of ITA, San Diego, CA, Feb 2019, pp. 1–8

    work page 2019