Beam-Aware Computation Offloading for Dual-Tier UAV Networks via Evolutionary Game Learning

Abstract

Dual-tier unmanned aerial vehicle (UAV) networks have emerged as a promising architecture for enabling flexible and on-demand services in low-altitude wireless environments. However, the high mobility of UAVs and the dynamic nature of wireless channels introduce significant challenges for beam selection, particularly in achieving efficient and stable medium access under time-varying network conditions. Conventional centralized or static approaches are often inadequate due to their limited adaptability and high signaling overhead. To address these challenges, this paper proposes a dynamic beam selection framework based on evolutionary game learning for dual-tier UAV networks. Specifically, the beam selection process is modeled as an evolutionary game, where multiple UAVs compete for limited beam resources and iteratively adjust their strategies based on local utility. This distributed mechanism enables adaptive and scalable decision-making without requiring global coordination, making it well suited for highly dynamic environments. To further enhance convergence efficiency and robustness, a learning-based optimization method based on Proximal Policy Optimization (PPO) is incorporated to guide strategy evolution. Simulation results demonstrate that the proposed approach achieves superior performance in terms of delay reduction, energy efficiency, and convergence speed compared with baseline schemes. These results validate the effectiveness of evolutionary game learning for dynamic beam selection and highlight its potential for enabling efficient medium access in dual-tier UAV networks.