Multi-layer RIS on Edge: Communication, Computation and Wireless Power Transfer

Baoqing Zhang; Junhong Jia; Shuyi Chen; Tianheng Xu; Weixiao Meng; Yifan Qin; Yingzhe Hui

REVIEW 3 major objections 4 minor 15 references

Multi-layer RIS on Edge: Communication, Computation and Wireless Power Transfer

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A multi-layer reconfigurable intelligent surface can act as one edge platform for MIMO communication, wave-domain computation, and wireless power transfer.

desk verdict A plausible and honestly written vision paper that integrates existing multi-layer RIS results, but its universal paradigm rests on an inter-layer channel model that is not yet shown to hold in practice. read the letter →

arxiv 2501.05780 v1 pith:JGPKUAQ5 submitted 2025-01-10 cs.IT math.IT

Shuyi Chen , Junhong Jia , Baoqing Zhang , Yingzhe Hui , Yifan Qin , Weixiao Meng , Tianheng Xu This is my paper · ORCID

classification cs.ITmath.IT

keywords multi-layerRISreconfigurableintelligentsurfacewave-domaincomputationMIMObeamformingwirelesspowertransferSWIPTdiffractivedeepneuralnetworkedgecomputing

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 universal edge platform built from a multi-layer reconfigurable intelligent surface (RIS), a sealed stack of programmable metasurface layers whose meta-atoms are tuned by a single FPGA controller. The claim is that the same passive structure can simultaneously perform MIMO communication, wave-domain computation, and wireless power transfer, replacing digital beamforming arrays, AI accelerators, and dedicated power transmitters. The motivation is that hardware-based scaling of communication and computing is too costly and power-hungry for massive IoT, while wave-based processing is nearly passive, low-latency, and scalable. Simulations reported in the paper show that MIMO capacity and SWIPT energy efficiency improve with the number of RIS layers up to a saturation point, with three layers optimal in the studied four-device scenarios.

What carries the argument

The load-bearing object is the multi-layer RIS itself: a compact, sealed stack of transmissive RIS layers, each a dense lattice of passive meta-atoms (about a tenth to a half wavelength in size) with FPGA-controlled phase and amplitude response, separated by several wavelengths. The mechanism that carries the argument is the inter-layer diffraction channel: each meta-atom output on one layer is the weighted sum of all outputs from the previous layer, with weights given by Rayleigh-Sommerfeld diffraction factors. This makes the whole stack a physical deep neural network whose adjustable weights are the meta-atom responses and whose fixed connections are the propagation factors, so beamforming matrices, computation kernels, and power-focusing phases are all realized in the wave domain during propagation. The number of layers is the central tuning parameter: more layers add degrees of freedom and beam resolution but also penetration loss, producing the saturation behavior in the simulations.

What would settle it

A direct test is to fabricate a multi-layer RIS, illuminate it with a known input field, and compare the measured output field to the field predicted from the ideal Rayleigh-Sommerfeld inter-layer model; if the residual cannot be made small by the proposed back-propagation calibration of the inter-layer coefficients, the precise-beamforming and wave-computation claims fail. A second check would be to measure the capacity of a real 3-layer stack serving four users and see whether it exceeds a single layer by the amount the simulations report.

Watch

Extended reading notes

Core claim

At the core is the idea that a multi-layer RIS, because its inter-layer propagation follows Huygens-Fresnel and Rayleigh-Sommerfeld diffraction, behaves mathematically like a fully connected neural network with the propagation factors as fixed weights and the meta-atom EM responses as adjustable parameters. By configuring those responses, the same stack can act as a MIMO beamformer, a diffractive deep neural network that executes computation tasks (classification, DOA estimation, encryption, over-the-air computation, joint source-channel coding), and an energy beamformer for wireless power transfer. The paper's simulations indicate that adding layers increases beam resolution and degrees of freedom, raising system capacity in MIMO and energy efficiency in SWIPT until penetration losses dominate, which is why the optimal layer count in the studied setups is three.

Load-bearing premise

The design assumes the inter-layer propagation between adjacent RIS layers is precisely described by Rayleigh-Sommerfeld diffraction and that those inter-layer coefficients are known when the surface is optimized or trained; if fabrication defects or assembly deformations make the real coefficients deviate, beam pointing misaligns and computation results become inaccurate, and the paper only sketches an error back-propagation calibration without showing it works.

Editorial extensions

If this is right

A base station equipped with multi-layer RIS could serve many IoT devices with MIMO beamforming performed in the wave domain, removing the need for excessive RF chains and high-precision DACs.
The same stack can execute light-speed computation tasks such as data-class-specific encryption, DOA estimation, over-the-air function computation, and joint source-channel coding, with latency independent of the number of meta-atoms and power cost dominated by the FPGA.
A standalone multi-layer RIS can act as a power beacon for WPT and SWIPT, using its higher beam resolution to focus energy on IoT devices and satisfy energy-harvest and SINR targets more easily as layers are added.
Because only the RF passage through the RIS is altered, the paradigm is backward-compatible with existing IoT devices and base stations, needing only the RIS stack and its FPGA controller as added hardware.
Adding RIS layers improves capacity and energy efficiency only up to a saturation point set by penetration loss, so layer count is a design parameter to optimize per scenario rather than something to maximize.

Reading between the lines

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

If the inter-layer channel model and calibration hold, the same wave-domain machinery could naturally extend to integrated sensing and communication, since DOA estimation and MIMO beamforming already share the diffraction structure.
The reported optimality of around three layers depends on the specific loss model and user count, so a design rule implied by the paper is to choose layer count per task and per hardware loss budget rather than maximize it universally.
The sketched error back-propagation calibration of inter-layer coefficients is directly testable: iteratively updating theoretical coefficients against measured output fields either validates the diffraction model or exposes its limits in practice.
A longer-term consequence is an energy-autonomous IoT loop, where a standalone multi-layer RIS powers devices by WPT and those devices use the harvested energy for uplink transmissions that the same RIS structure helps decode.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Quick take: this is a genuinely useful vision paper, not a rigorous research paper. It argues that a multi-layer RIS can be configured as a single nearly-passive platform for MIMO beamforming, wave-domain computation (D2NN-style), and WPT, and it maps out applications and open problems. The combination is new even though the components all come from existing prototypes. I think the central idea is plausible and the paper is honest about what is missing.

What it does well: it connects three previously separate literatures (stacked intelligent metasurfaces for beamforming, diffractive neural networks for computation, and metasurface WPT) under one architecture. The system model in Section III is a reasonable way to organize the use cases. The paper gives credit to existing prototypes and explicitly admits that the inter-layer channel is treated as known. That is more honest than many papers in this space. The self-citations are to the authors' own prior work on semantic communication and AirJSCC; that is normal practice and not a red flag.

The soft spots, in proportion: First, the load-bearing assumption is the fixed Rayleigh-Sommerfeld inter-layer channel. The stress-test concern is right to push further than the authors do: adjacent layers are only a few wavelengths apart and densely populated with tunable meta-atoms, so the field re-scattered from layer k+1 back into layer k depends on the tuning state of layer k+1. That means the effective inter-layer coupling may be state-dependent, not a known constant. The paper flags calibration but not this multiple-scattering effect. For a vision paper this is acceptable as an open problem, but it does mean the universal paradigm is conditional on solving a hard physics problem. Second, the two simulations are described in prose with no equations, no parameter values, and no single-layer baselines in the figures. They read as illustrations, not evidence. Third, 'power-free computation' is overstated; the FPGA controller consumes power, and the meta-atoms themselves need biasing. The paper does qualify this later, but the abstract and Section IV-B use the stronger phrase.

Who this is for: someone new to multi-layer RIS who wants a map of possible use cases, or a researcher looking for open problems in inter-layer channel modeling and calibration. It is not for someone who wants a demonstrated system. A serious referee could engage with it, but the right venue is a magazine, and the revisions should be about qualifiers and clearly labeling the simulations as illustrative.

Referee Report

3 major / 4 minor

Summary. The paper proposes a multi-layer RIS architecture as a universal edge platform for IoT, claiming that a single stacked-RIS structure can provide MIMO communication, wave-domain computation, and wireless power transfer (WPT) with minimal hardware changes, low cost, and low power consumption. It describes the multi-layer RIS structure, presents a system architecture with various communication/energy links, discusses the three functions in separate sections, reports two illustrative simulation results (system capacity versus number of layers, and SWIPT energy efficiency versus number of layers), and concludes with future research directions such as inter-layer channel modeling, resource allocation, channel estimation, and edge training.

Significance. If the proposed universal paradigm were validated, it could have substantial impact by replacing digital beamforming hardware, AI accelerators, and dedicated power transmitters with a single nearly passive structure at the edge. The paper usefully synthesizes prior prototypes (e.g., RIS-based MIMO, D2NN classification, RIS-based WPT) and clearly articulates a vision, while also being honest about open problems in Section V. However, the current manuscript provides no detailed derivations, no reproducible simulation models, and no experimental validation; the quantitative performance claims rest on underspecified figures. The central architectural claim is also sensitive to a physical assumption about the inter-layer channel that the paper itself flags as uncertain. As a vision/position paper the article has merit, but the strength of the claims currently exceeds the evidence provided.

major comments (3)

[II-A and V-A] The designs in Sections IV-A through IV-C assume the inter-layer channel is a fixed, known Rayleigh-Sommerfeld diffraction matrix. Because adjacent layers are separated by only a few wavelengths and each layer contains densely packed tunable meta-atoms, the field re-scattered by layer k+1 back into layer k depends on the tuning state of layer k+1. The effective inter-layer coupling is therefore state-dependent, not a known constant. The calibration procedure sketched in Section V-A (error back-propagation) would only capture a fixed channel and is not demonstrated. This concern is load-bearing because MIMO beamforming, D2NN computation, and WPT energy focusing all inherit the same channel assumption.
[IV-A, IV-C, Figs. 4 and 6] The simulation results are under-specified. No signal model, channel model, or optimization problem is stated, and no parameter values are given (e.g., number of meta-atoms per layer, inter-layer spacing, wavelength, transmit power budget, noise level, QoS thresholds). The figures do not clearly show single-layer baseline curves despite the text claiming comparison with single-layer counterparts. Without these details, the central quantitative claims—capacity increasing up to an optimal three layers and energy efficiency improving with layer count—cannot be evaluated or reproduced.
[III and IV-B] The phrase 'power-free computation' is overstated. The paper itself states in Section IV-B that power consumption is primarily due to the FPGA used for system control. The FPGA and per-meta-atom bias circuitry consume power, so the computation is not literally power-free. The claim should be qualified as 'power-efficient' or 'computation whose power consumption is limited to control circuitry.'

minor comments (4)

[III] The statement that 'only the transmission process of RF signals is altered' understates the need for per-meta-atom bias control lines and an FPGA controller; please clarify the intended meaning of 'minimal hardware changes.'
[IV-B] The claim that D2NN is a 'universal function approximator' extrapolates from the universal linear transformation result in [11] and the potential for nonlinear activation functions; please cite a specific result or temper the claim.
[IV-B] There is a typo: 'Beside, as computation' should be 'Besides, as computation.'
[I] Calling the proposal 'an unprecedented attempt' is too strong given the existing prototypes and the cited works [10] and [12]; please qualify the novelty more carefully.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the proposal rests on physical diffraction modeling and external prototypes, not on fitted parameters or self-citation chains.

full rationale

The paper is a position/proposal article, not a derivation. Its central claim that multi-layer RIS can enable MIMO, computation, and WPT is supported by citing external prototypes ([2], [3], [5], [9]) and by simulations that optimize over phase shifts under a stated physical model (Rayleigh-Sommerfeld diffraction). No fitted parameter is disguised as a prediction: the capacity and energy-efficiency curves in Figs. 4 and 6 arise from optimization simulations whose improvement with layer count is a plausible consequence of additional degrees of freedom, not a consequence of re-using the input data. The inter-layer diffraction model is explicitly identified as a known-quantity assumption, and the paper candidly flags the risk of model mismatch in Section V-A, proposing calibration rather than assuming it away. The self-citations [4] and [13] appear as supporting examples of prior diffractional neural network and over-the-air coding work, but the central architecture claim does not reduce to these citations; it is independently anchored in the cited external prototypes and in the physical layer model. The neural-network analogy is rhetorical, not a circular derivation. There is no equation in the paper that defines one claimed output in terms of another claimed output, so no circularity step can be exhibited.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claims rest on a physical model (Rayleigh-Sommerfeld inter-layer channels), an analogy to neural networks, and an assumed low-power passive hardware model. Free parameters in the simulations are not disclosed, including per-layer penetration loss, QoS thresholds, power budgets, and phase-shift resolution. No invented physical entities are introduced; SA-RIS is a deployment configuration, not a new entity.

free parameters (4)

per-layer penetration loss = not disclosed
The capacity simulation in Section IV-A assumes signals lose energy passing through each RIS layer; the optimal layer count of 3 is determined by this loss, but its value and model are not given.
QoS targets (SINR threshold theta, energy harvest threshold delta) = not fully specified
Section IV-C: strictness of QoS targets changes whether a 2-layer structure meets requirements; the threshold values are chosen for the case study and not disclosed.
maximum transmit power budget = not specified
Section IV-A and IV-C: capacity and energy efficiency results depend on the power budget at the BS or SA-RIS, which is not quantified.
phase shift resolution = assumed continuously tunable
Section IV-A states 'Assuming the phase shifts of multi-layer RIS can be continuously tuned'; practical finite-resolution phases would change the simulation results.

assumptions (5)

domain assumption Inter-layer propagation follows Rayleigh-Sommerfeld diffraction theory and is treated as a known quantity when optimizing the surface.
Section II-A and V-A: the entire optimization assumes known inter-layer channels, which the authors admit may deviate in practice.
domain assumption Multi-layer RIS can realize universal linear transformation and nonlinear activation functions, making it a universal function approximator.
Section IV-B, based on [11] and [8]; the nonlinearity demonstration is from optical neural networks, not passive RIS, so transferability is assumed.
domain assumption Meta-atoms are nearly passive and low-cost, with power consumption mainly from the FPGA controller.
Section II-A and IV-B: underpins the 'power-free computation' claim, ignoring configuration and control energy.
domain assumption The multi-layer RIS can be viewed as a deep neural network, so deep-learning training and performance properties apply.
Section IV-B: used to argue computation performance improves with layers until vanishing gradient, without direct RIS experiments.
standard math Huygens-Fresnel principle: each meta-atom acts as a point source illuminating all meta-atoms on the next layer.
Section II-A: the basis for the fully-connected inter-layer weights.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Multi-layer RIS on Edge: Communication, Computation and Wireless Power Transfer." pith.science (2026). https://pith.science/paper/JGPKUAQ5

@misc{pith2026250105780,
  author       = {Pith},
  title        = {Pith review of: Multi-layer RIS on Edge: Communication, Computation and Wireless Power Transfer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JGPKUAQ5}},
  note         = {Machine review of arXiv:2501.05780}
}

read the original abstract

The rapid expansion of Internet of Things (IoT) and its integration into various applications highlight the need for advanced communication, computation, and energy transfer techniques. However, the traditional hardware-based evolution of communication systems faces challenges due to excessive power consumption and prohibitive hardware cost. With the rapid advancement of reconfigurable intelligent surface (RIS), a new approach by parallel stacking a series of RIS, i.e., multi-layer RIS, has been proposed. Benefiting from the characteristics of scalability, passivity, low cost, and enhanced computation capability, multi-layer RIS is a promising technology for future massive IoT scenarios. Thus, this article proposes a multi-layer RIS-based universal paradigm at the network edge, enabling three functions, i.e., multiple-input multiple-output (MIMO) communication, computation, and wireless power transfer (WPT). Starting by picturing the possible applications of multi-layer RIS, we explore the potential signal transmission links, energy transmission links, and computation processes in IoT scenarios, showing its ability to handle on-edge IoT tasks and associated green challenges. Then, these three key functions are analyzed respectively in detail, showing the advantages of the proposed scheme, compared with the traditional hardware-based scheme. To facilitate the implementation of this new paradigm into reality, we list the dominant future research directions at last, such as inter-layer channel modeling, resource allocation and scheduling, channel estimation, and edge training. It is anticipated that multi-layer RIS will contribute to more energy-efficient wireless networks in the future by introducing a revolutionary paradigm shift to an all-wave-based approach.

Figures

Figures reproduced from arXiv: 2501.05780 by the authors.

**Figure 1.** The structure of multi-layer RIS. The entire structure should be enclosed within a container lined with absorbing materials to minimize [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Prototypes of RIS-based MIMO transmission, computation and WPT, where, (a) shows two prototypes of RIS-based MIMO [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** System architecture of multi-layer RIS-empowered edge IoT scenario, where multi-layer RIS is leveraged to facilitate the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: System capacity of multi-layer RIS-enabled MIMO transmis [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Multi-layer RIS-enabled computation schemes, where multi-layer RIS is utilized to perform (a) data-class-specific encryption; (b) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: Energy efficiency of the multi-layer RIS-enabled SWIPT [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

[1]

Dis- tributed artificial intelligence empowered by end-edge-cloud computing: A survey

S. Duan, D. Wang, J. Ren, F. Lyu, Y . Zhang, H. Wu and X. Shen, “Dis- tributed artificial intelligence empowered by end-edge-cloud computing: A survey”, IEEE Communications Surveys & Tutorials , vol. 25, no. 1, pp. 591-624, First quarter 2023

work page 2023
[2]

Design and implementation of MIMO transmission based on dual-polarized reconfigurable intelligent surface

X. Chen, J. Ke, W. Tang, M. Chen, J. Dai, E. Basar, S. Jin, Q. Cheng and T. Cui, “Design and implementation of MIMO transmission based on dual-polarized reconfigurable intelligent surface”, IEEE Wireless Communications Letters , vol. 10, no. 10, pp. 2155-2159, Oct. 2021

work page 2021
[3]

Multi-user ISAC through Stacked Intelligent Metasurfaces: New Algorithms and Experiments

Z. Wang, H. Liu, J. Zhang, R. Xiong, K. Wan, X. Qian, M. Di Renzo, Robert C. Qiu, “Multi-user ISAC through stacked intelli- gent metasurfaces: New algorithms and experiments”, arXiv preprint arXiv:2405.01104, 2024

work page Pith review arXiv 2024
[4]

RIS-based on-the-air semantic communications — A diffractional deep neural network approach

S. Chen, Y . Hui, Y . Qin, Y . Yuan, W. Meng, X. Luo, and H. Chen, “RIS-based on-the-air semantic communications — A diffractional deep neural network approach”, IEEE Wireless Communications , vol. 31, no. 4, pp. 115-122, Aug. 2024

work page 2024
[5]

Classification of metal handwritten digits based on microwave diffractive deep neural network

Z. Gu, Q. Ma, X. Gao, J. You, and T. Cui, “Classification of metal handwritten digits based on microwave diffractive deep neural network”, Advanced Optical Materials , vol. 12, no. 7, pp. 1-8, 2024

work page 2024
[6]

Metasurface- based diffractive optical networks with dual-channel complex amplitude modulation

G. Lu, J. Qiu, T. Liu, D. Zhang, S. Xiao and T. Yu, “Metasurface- based diffractive optical networks with dual-channel complex amplitude modulation”, Journal of Lightwave Technology , vol. 42, no. 20, pp. 7282-7290, Oct. 2024

work page 2024
[7]

Diffraction-driven parallel convolution processing with integrated photonics

H. Chen, Y . Huang, W. Liu, R. Sun, T. Fu, Y . Wang, Z. Huang and S. Yang, “Diffraction-driven parallel convolution processing with integrated photonics”, Research Square preprint , May. 2024

work page 2024
[8]

Image sensing with multilayer nonlinear optical neural networks

T. Wang, M. M. Sohoni, L. G. Wright, M. M. Stein, S. Ma, T. Onodera, M. G. Anderson and P. L. McMahon, “Image sensing with multilayer nonlinear optical neural networks”, Nature Photonics , vol. 17, pp.408- 415, May. 2023

work page 2023

Show all 15 references

[9]

Intelligent wireless power transfer via a 2-bit compact reconfigurable transmissive-metasurface-based router

W. Li, Q. Yu, J. Qiu, J. Qi, “Intelligent wireless power transfer via a 2-bit compact reconfigurable transmissive-metasurface-based router”, Nature Communications, vol. 15, pp. 1-10, Apr. 2024

2024
[10]

Stacked intelligent metasurfaces for multiuser beamforming in the wave domain

J. An, M. Di Renzo, M. Debbah and C. Yuen, “Stacked intelligent metasurfaces for multiuser beamforming in the wave domain”, in IEEE International Conference on Communications (ICC) , Rome, Italy, pp.2834-2839, 2023

2023
[11]

Data-class-specific all-optical transformations and encryption

B. Bai, H. Wei, X. Yang, T. Gan, D. Mengu, M. Jarrahi, and A. Ozcan, “Data-class-specific all-optical transformations and encryption”, Advanced Materials , vol. 35, no. 31, pp. 1-22, 2023

2023
[12]

Two-dimensional direction-of-arrival estimation using stacked intelligent metasurfaces

J. An, C. Yuen, Y . Guan, M. Di Renzo, M.Debbah, H. V . Poor and L. Hanzo, “Two-dimensional direction-of-arrival estimation using stacked intelligent metasurfaces”, IEEE Journal on Selected Areas in Communications, vol. 42, no. 10, pp. 2786-2802, Oct. 2024

2024
[13]

AirJSCC: Re- configurable intelligent surfaces based over-the-air joint source-channel coding

Y . Hui, S. Chen, W. Meng, H. Chen, W. Sun and L. Ma, “AirJSCC: Re- configurable intelligent surfaces based over-the-air joint source-channel coding”, in IEEE Global Communications Conference (GLOBECOM) , Kuala Lumpur, Malaysia, pp. 4092-4097, 2023

2023
[14]

Matrix diffractive deep neural networks merging polarization into meta-devices

Y . Wang, A. Yu, Y . Cheng and J. Qi, “Matrix diffractive deep neural networks merging polarization into meta-devices”, Laser & Photonics Reviews, vol. 12, no. 2, pp. 1-11, 2024

2024
[15]

K. Liu, Z. Zhang, L. Dai and L. Hanzo, ”Compact user-specific reconfig- urable intelligent surfaces for uplink transmission”, IEEE Transactions on Communications , vol. 70, no. 1, pp. 680-692, Jan. 2022. BIOGRAPHIES Shuyi Chen (chenshuyitina@gmail.com) received her B.Sc, M.Sc ...

2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Dis- tributed artificial intelligence empowered by end-edge-cloud computing: A survey

S. Duan, D. Wang, J. Ren, F. Lyu, Y . Zhang, H. Wu and X. Shen, “Dis- tributed artificial intelligence empowered by end-edge-cloud computing: A survey”, IEEE Communications Surveys & Tutorials , vol. 25, no. 1, pp. 591-624, First quarter 2023

work page 2023

[2] [2]

Design and implementation of MIMO transmission based on dual-polarized reconfigurable intelligent surface

X. Chen, J. Ke, W. Tang, M. Chen, J. Dai, E. Basar, S. Jin, Q. Cheng and T. Cui, “Design and implementation of MIMO transmission based on dual-polarized reconfigurable intelligent surface”, IEEE Wireless Communications Letters , vol. 10, no. 10, pp. 2155-2159, Oct. 2021

work page 2021

[3] [3]

Multi-user ISAC through Stacked Intelligent Metasurfaces: New Algorithms and Experiments

Z. Wang, H. Liu, J. Zhang, R. Xiong, K. Wan, X. Qian, M. Di Renzo, Robert C. Qiu, “Multi-user ISAC through stacked intelli- gent metasurfaces: New algorithms and experiments”, arXiv preprint arXiv:2405.01104, 2024

work page Pith review arXiv 2024

[4] [4]

RIS-based on-the-air semantic communications — A diffractional deep neural network approach

S. Chen, Y . Hui, Y . Qin, Y . Yuan, W. Meng, X. Luo, and H. Chen, “RIS-based on-the-air semantic communications — A diffractional deep neural network approach”, IEEE Wireless Communications , vol. 31, no. 4, pp. 115-122, Aug. 2024

work page 2024

[5] [5]

Classification of metal handwritten digits based on microwave diffractive deep neural network

Z. Gu, Q. Ma, X. Gao, J. You, and T. Cui, “Classification of metal handwritten digits based on microwave diffractive deep neural network”, Advanced Optical Materials , vol. 12, no. 7, pp. 1-8, 2024

work page 2024

[6] [6]

Metasurface- based diffractive optical networks with dual-channel complex amplitude modulation

G. Lu, J. Qiu, T. Liu, D. Zhang, S. Xiao and T. Yu, “Metasurface- based diffractive optical networks with dual-channel complex amplitude modulation”, Journal of Lightwave Technology , vol. 42, no. 20, pp. 7282-7290, Oct. 2024

work page 2024

[7] [7]

Diffraction-driven parallel convolution processing with integrated photonics

H. Chen, Y . Huang, W. Liu, R. Sun, T. Fu, Y . Wang, Z. Huang and S. Yang, “Diffraction-driven parallel convolution processing with integrated photonics”, Research Square preprint , May. 2024

work page 2024

[8] [8]

Image sensing with multilayer nonlinear optical neural networks

T. Wang, M. M. Sohoni, L. G. Wright, M. M. Stein, S. Ma, T. Onodera, M. G. Anderson and P. L. McMahon, “Image sensing with multilayer nonlinear optical neural networks”, Nature Photonics , vol. 17, pp.408- 415, May. 2023

work page 2023