Physical layer security for fixed-gain UAV-mounted RIS assisted D2D communications

We study a physical-layer security control problem for fixed-gain reconfigurable intelligent surface (RIS)-UAV-assisted secure device-to-device (D2D) communications. The setting is motivated by temporary aerial access, in which a relay UAV provides short-term service and a RIS-UAV adjusts the reflected channel for local D2D reuse. In the present model, the RIS amplification coefficients are fixed, the relay beamforming vector follows maximum ratio transmission, and the learning controller selects the RIS phase shifts and D2D transmit powers. An importance-based D2D neighbor-pairing rule is used as a preprocessing step to update local interference relations. The secrecy-control objective combines normalized secrecy rate and secrecy energy efficiency (SEE) as dimensionless reward components. A problem-tailored Dual-Actor Network Softmax Twin Delayed Deep Deterministic Policy Gradient (DA-STD3) controller is used for the continuous phase-power action space. The controller retains the twin-critic, delayed-update, and target-policy-smoothing mechanisms of TD3, while two actors generate phase-power candidates that are projected to the box-constrained physical action region and evaluated through a sampled-softmax backup. Simulation results show that, over the considered operating conditions, DA-STD3 achieves higher secrecy spectral efficiency and SEE than TD3, DDPG, DQN, and random control.