WorksEngineering
Joint Antenna Positioning and Dynamic Beam Tracking for MA-Enabled Full-Duplex Secure ISAC Systems: A Flow Matching DRL Approach
Abstract
Integrated sensing and communication (ISAC) improves spectral efficiency but faces significant security challenges in dynamic wireless environments, where the potential eavesdropper is treated as the sensing target. This paper proposes a movable antenna (MA)-enabled full-duplex secure ISAC system that tracks the mobile eavesdropper while safeguarding both uplink and downlink transmissions. Specifically, an extended Kalman filter (EKF) is employed to recursively estimate and predict the eavesdropper's state from radar echoes. Building on the estimated target state, we formulate a joint optimization problem to maximize the sum of the uplink and downlink secrecy rates by jointly optimizing the transmit and receive MA positions, uplink and downlink beamforming, artificial noise, and uplink transmit power. The resulting problem is challenging due to its inherent nonconvexity, strong coupling among optimization variables, and time-varying nature. Conventional iterative optimization methods may incur prohibitive computational overhead in dynamic environments, whereas existing deep reinforcement learning (DRL) methods based on unimodal Gaussian or deterministic policies may struggle to capture complex multimodal high-reward action distributions. To address these challenges, we develop a flow matching-based deep reinforcement learning (FMDRL) algorithm for sequential joint decision-making. The proposed algorithm integrates a conditional velocity-field actor with twin critics, generating continuous actions through finite-step Euler integration and employing a deterministic decoder to enforce physical constraints on antenna positions, transmit power, and positive semidefiniteness. Extensive numerical simulations confirm that the proposed FMDRL algorithm achieves superior performance in terms of both communication security and target tracking capabilities.