| Title |
Multi-agent Deep Reinforcement Learning for Joint Beamforming, Active RIS Control, and Movable Antenna Positioning in ISAC Systems |
| Authors |
신승석(Seungseok Sin) ; 김재혁(Jaehyeok Kim) ; 조인식(Insik Cho) ; 김태홍(Taehong Kim) ; 문상미(Sangmi Moon) ; 황인태(Intae Hwang) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.7.3 |
| Keywords |
Alternating Optimization (AO); Active Reconfigurable Intelligent Surface (Active RIS); Integrated Sensing and Communication (ISAC); Joint Optimization; Movable Antenna (MA); Multi-Agent Deep Reinforcement Learning (MA-DRL) |
| Abstract |
Integrated Sensing and Communication (ISAC) has emerged as a key enabling technology for next-generation 6G networks, where communication and sensing functionalities are jointly supported within a unified framework. In addition, active Reconfigurable Intelligent Surfaces (RIS) and Movable Antennas (MA) provide promising degrees of freedom to shape the propagation environment and exploit geometric diversity. This paper considers an ISAC system aided by an active RIS and MA, and investigates the joint control of (i) BS transmit beamforming, (ii) RIS reflection/amplification coefficients, and (iii) MA positions to maximize radar sensing performance (Radar SINR) while satisfying user communication QoS constraints. Since the resulting optimization is highly non-convex and strongly coupled with physical constraints, we formulate the problem as a multi-agent deep reinforcement learning (MA-DRL) task. Specifically, we employ three agents for BS, RIS, and MA, respectively, each learning a continuous control policy in an off-policy manner, while sharing a global reward signal from the environment to achieve cooperative optimization. Simulation results demonstrate that the proposed MA-DRL framework achieves notable improvements in radar sensing performance compared with conventional fixed-antenna and passive-RIS baselines, while maintaining the required communication QoS. |