An adaptive antenna system produced the strongest secrecy-rate results among the approaches tested in a modeled wireless network, according to an arXiv preprint. Movable-PA PASS outperformed fixed-PA PASS and conventional MIMO across the displayed security modes and transmit-power settings, and its advantage increased as transmit power rose.
The study proposes one PASS-based framework for low-, medium- and high-security transmission. It jointly optimizes beamforming and antenna positions over a long-horizon control problem. In plain terms, its physical-layer security goal is to shape the useful signal and artificial noise around the receivers that each mode treats as possible interceptors.
A controller that changes with the security goal
The simulation uses a 20 × 20 m² service region with four intended users, called Bobs, and two external Eves. It also includes four waveguides and four PAs per waveguide.
The three modes change who must be protected from leakage. Low-security transmission, or LST, has no explicit confidentiality requirement. Medium-security transmission, MST, protects against external Eves but not other Bobs. High-security transmission, HST, treats both other Bobs and Eves as potential interceptors.
The optimization jointly controls the transmission beamforming and PA positions. It seeks the best mode-dependent performance while enforcing limits on power, minimum performance, antenna location and the spacing between antenna elements.
How the control methods learned
To test the learning process, the researchers trained all deep-reinforcement-learning algorithms for 2,000 episodes, with 100 time steps in each episode. They checked performance every 50 episodes over 10 independent runs, and randomly selected the security mode at each time slot.
HSPPO changes standard PPO by standardizing violation costs across modes, adapting violation correction to the rate at which a mini-batch breaks the rules, and weighting samples using security-consistency feedback.
MRHA-DPO takes a different route. It combines a multi-relational graph representation of the PASS system, hierarchy-aware generation of coupled actions, and diffusion policy optimization.
The ordinary PPO, HSPPO and MRHA-DPO agents all reached stable reward levels under LST, MST and HST. TD3 continued to fluctuate and did not reach stable convergence. HSPPO and MRHA-DPO also consistently beat PPO across the three modes, with MRHA-DPO delivering the stronger reward enhancement among the proposed methods.
Where the gains appeared
The comparison also changed as the service area grew. As the service-region side length increased from 10 to 50 metres, all schemes lost performance, but movable-PA PASS remained best. Under LST, its absolute lead over MIMO narrowed from about 1.05 to 0.50 bit/s/Hz, while its relative gain rose from about 70% to 119%.
Component tests gave a similar message for HSPPO. Enabling both advantage correction and consistency weighting produced the highest minimum-rate values among the tested HSPPO configurations: 1.3804 bits/s/Hz in LST, 0.7549 in MST and 0.7278 in HST.
For MRHA-DPO, the highest tested minimum rates came when the graph encoder and diffusion actor were enabled together. They were 1.7826 bits/s/Hz for LST, 1.2457 for MST and 1.1264 for HST.
The energy patterns also shifted with the security objective. In LST, target-signal energy clustered around the target Bob without artificial noise aimed at a specific unintended receiver. In MST, artificial noise was directed toward external Eves. In HST, the target signal was concentrated more strongly and the artificial noise was adjusted to suppress leakage to Eves and non-target Bobs.
A result bounded by its model
Those findings are bounded by what was tested. The preprint reports numerical simulations for a defined service region with stated security modes, constraints and baseline comparisons. The reported gains therefore describe this model and these tests; they do not establish how the methods would perform in untested deployments.
Paper data and sources
Original title: Security-Aware Pinching-Antenna Systems (PASS): Physical-Layer Security Transmission
Authors: Zhaoming Hu, Xiaochen Nie, Ruikang Zhong et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text