Joint beamforming and phase optimization for UAV-RIS systems by gradient-based adaptive meta-learning

Abstract

With the rapid advancement of wireless communication technologies, achieving high reliability, low latency, and energy efficiency remains a critical challenge. Unmanned Aerial Vehicle-mounted Reconfigurable Intelligent Surfaces (UAV-RIS) leverage both the reconfigurability of RIS and the mobility of UAVs to enhance air-to-ground link quality. This paper investigates the joint optimization of UAV trajectory, base station (BS) beamforming, and RIS phase control in UAV-RIS-assisted systems. A density-aware positioning algorithm is first introduced to optimize the UAV trajectory and improve link robustness. Building upon this, a gradient-based adaptive meta-learning (GAML) framework is developed to jointly optimize BS beamforming and RIS phase shifts without extensive pre-training, thereby enhancing adaptability in time-varying channels. The proposed GAML scheme avoids explicit non-convex optimization and significantly reduces computational complexity. Furthermore, the integration of adaptive network structures enhances gradient propagation and mitigates vanishing gradients. Simulation results show that GAML outperforms existing learning-based and random phase methods, achieving up to 14.25% improvement in weighted sum rate (WSR), demonstrating its superior adaptability and efficiency.