Secure Time-Modulated Intelligent Reflecting Surface via Generative Flow Networks
Zhihao Tao, Athina P. Petropulu · 2025
Read Secure Time-Modulated Intelligent Reflecting Surface via Generative Flow Networks by Zhihao Tao, Athina P. Petropulu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-14992.txt.
Introduction
Zhihao Tao and Athina Petropulu design a secure wireless surface in a 2025 engineering paper. Their surface can reflect signals so that intended users receive a clean message while listeners in other directions get a scrambled one. Older designs mostly served a single user and followed fixed rules.
The new design uses a generative flow network, a type of AI that learns to pick good settings from many options. It handles several users at once, which fixed rules could not do well. The work is aimed at wireless systems that must keep signals private without heavy encryption at every step.
Catalog record
| CREATOR | Zhihao Tao, Athina P. Petropulu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-14992
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Secure Time-Modulated Intelligent Reflecting Surface via Generative Flow Networks Thanks: This work was supported by ARO grant W911NF2320103 and NSF grant ECCS-2320568. Zhihao Tao, Athina P. Petropulu Abstract We propose a novel directional modulation (DM) design for OFDM transmitters aided by a time-modulated intelligent reflecting surface (TM-IRS). The TM-IRS is configured to preserve the integrity of transmitted signals toward multiple legitimate users while scrambling the signal in all other directions. Existing TM-IRS design methods typically target a single user direction and follow predefined rule-based procedures, making them unsuitable for multi-user scenarios. Here, we propose a generative AI-based approach to design good sets of TM-IRS parameters out of a set of all possible quantized ranges of parameters. The design objective is to maximize the sum rate across the authorized directions. We model the TM-IRS parameter selection as a deterministic Markov decision process (MDP), where each terminal state corresponds to a specific configuration of TM-IRS parameters. GFlowNets are employed to learn a stochastic policy that samples TM-IRS parameter sets with probability proportional to their associated sum rate reward. Experimental results demonstrate that the proposed method effectively enhances the security of the TM-IRS-aided OFDM systems with multi-users. Also, despite the vast size of the TM-IRS configuration space, the GFlowNet is able to converge after training on fewer than 0.000001% of all possible configurations, demonstrating remarkable efficiency compared to exhaustive combinatorial search. Implementation code is available at https://github.com/ZhihaoTao/GFN4TM-RIS to facilitate reproducibility. I Introduction The broadcast nature of wireless propagation exposes confidential data to interception unless special precautions are taken. Physical layer security (PLS) exploits physical characteristics of the wireless medium, such as channel fading, noise, interference, and spatial diversity, to complement—or, in some circumstances, replace—higher-layer cryptographic techniques [ 1 , 2 , 3 ] . Among the many PLS mechanisms proposed, directional modulation (DM) has attracted particular interest because it embeds information in the spatial signature of the transmitted waveform: a receiver aligned with the intended steering direction observes an undistorted constellation, whereas other directions see a scrambled one [ 4 , 5 , 6 ] . Compared with other PLS approaches such as secrecy rate maximization [ 7 , 8 ] or artificial-noise injection [ 9 , 10 ] , DM can offer comparable secrecy in a more cost-effective and energy-efficient manner [ 11 ] . DM implementations have been proposed for fully digital or hybrid beamforming architectures with multiple radio-frequency (RF) chains and fine-grained phase control at each antenna element or each transmitted symbol [ 12 , 13 , 14 , 11 ] . A cost-effective single-RF-chain alternative is to apply time modulation to a phased array driven by orthogonal frequency-division multiplexing (OFDM) signals [ 15 , 16 ] . By using single-pole-single-throw (SPST) switches to periodically connect and disconnect antennas to the RF chain, a time-modulated array (TMA) generates controllable harmonics whose periods are aligned with the OFDM symbol duration. As a result, each subcarrier of the transmitted OFDM signal carries a weighted mixture of all original symbols, where the mixing coefficients depend on the TMA parameters, i.e., connection times, or on states, and on state durations. This mixing represents scrambling of the transmitted symbols in all directions. The scrambling towards a desired direction can be eliminated by a rule-based design of the TM parameters. It should be noted that the energy utilization efficiency of TMA reduces with the periodic deactivation of antenna elements [ 17 ] . Recent research [ 18 ] shifts time modulation to an intelligent reflecting surface (IRS). IRS is a passive metasurface composed of programmable elements that dynamically adjust the phase of incident electromagnetic waves to realize beamforming [ 19 ] . By exerting the periodic TM on each IRS element, the system in [ 18 ] is designed to implement a 3D directional modulation. Also, the large aperture of an IRS delivers substantial beamforming gain that compensates for power lost of TMA during element deactivation. In [ 18 ] , the TM-IRS parameters are still obtained using simple, closed-form rules. Although these rule-based patterns are straightforward to implement, they ensure undistorted signal reception in only a single user direction, while their extension to multiple users is challenging. This is particularly restrictive, as multi-user scenarios are common in contemporary wireless communication systems. This paper proposes a generative-AI framework for TM-IRS-assisted OFDM systems that replaces rule-based TM-IRS patterns with selecting good sets of TM-IRS parameters out of a set of all possible quantized ranges of parameters. The measure of goodness is the sum rate at authorized directions. We model the TM-IRS parameter selection as a deterministic Markov decision process (MDP), where each terminal state corresponds to a specific configuration of TM-IRS parameters. GFlowNets are employed to learn a stochastic policy that samples TM-IRS parameter sets with probability proportional to their associated sum rate reward. GFlowNets are unsupervised models that leverage a flow-matching principle to sample composite objects with probabilities proportional to a user-defined reward [ 20 , 21 ] . In our setting, the analytical expression for the sum rate of the legitimate users serves as the reward function. The TM-IRS parameter space is discretized, and the parameter selection task is formulated as a deterministic Markov decision process (MDP), where each terminal state corresponds to a complete TM-IRS parameter set. i.e., a full on/off and phase shift configuration of the IRS elements. A feedforward neural network-based GFlowNet is trained offline to sample TM-IRS parameter sets in proportion to their associated sum rate reward. Experimental results show that the TM-IRS patterns generated by the learned GFlowNet yield undistorted reception at multiple desired directions while effectively scrambling signals in all other directions, thereby thwarting potential eavesdroppers. Notably, the sampling policy is stochastic and remains unknown to any adversary, significantly increasing the difficulty of potential attacks. Moreover, unlike deterministic optimization methods, the GFlowNet naturally generates a diverse set of high-performing TM-IRS configurations. This diversity allows the system to randomize TM-IRS patterns over time, thereby further enhancing the security [ 16 ] . The remainder of the paper is organized as follows. Section II presents the system model and formalizes the TM-IRS parameter optimization problem. Section III details the proposed GFlowNet-based TM-IRS parameter design framework. Numerical results demonstrating the effectiveness of the proposed approach are provided in Section IV . Finally, Section V concludes the paper and discusses potential directions for future research. II System Model Consider an IRS composed of M_{x}\times M_{z} passive reflecting units that assists a uniform linear array (ULA) transmitting OFDM signals. Let \theta_{T} and \phi_{T} denote the elevation and azimuth angles of the ULA transmitter w.r.t. the IRS, respectively. From the transmitter’s perspective, the IRS is modeled as a point target due to the sub-wavelength size of individual IRS elements and the overall compactness of the surface [ 22 ] . Thus, we denote the direction of the IRS as viewed from the ULA by \theta_{I} . We consider a single legitimate user first for notational convenience and denote its direction relative to the IRS as (\theta_{c},\phi_{c}) . It is assumed that \theta_{I} is known at the transmitter, and (\theta_{T},\phi_{T}) and (\theta_{c},\phi_{c}) are known at the IRS. The location of potential eavesdroppers is assumed to be unknown. Additionally, all elements of both the ULA and the