Study On Reinforcement Learning Based UAV Anti-Jamming Offloading Technique

Unmanned aerial vehicles(UAVs)are vulnerable to jamming attacks,which degrade the reliability of the UAV transmission and result in high energy consumption and latency.The existing UAV anti-jamming offloading schemes based on trajectory or power control can’t resist the smart jamming attacks,because smart jammers can dynamically adjust the jamming power.Therefore,the UAV anti-jamming offloading scheme is proposed to globally optimize the transmit power,moving trajectory and offloading rate to decrease the latency and energy consumption.The game theoretic analysis of the UAV anti-jamming transmission is proposed in this thesis.The Nash equilibria of the proposed UAV anti-jamming transmission game are derived.The optimal UAV transmission policy and the conditions ensuring the existence of NEs are provided on different system states.The deep reinforcement learning based UAV anti-jamming offloading scheme is proposed to dynamically optimize the transmit power,moving trajectory and the offloading rate of the UAV,and decrease the latency and energy consumption.The simulation results show that the proposed UAV anti-jamming offloading scheme decreases the latency by 36.6%and energy consumption by 29.5%compared with the Wolf-PDS based scheme.The experience transfer based UAV anti-jamming offloading scheme is proposed for the UAV with more computational resources and historical experiences of the antijamming transmission game.The experiences from similar environments are aggregated and transferred to further accelerate the speed of policy optimization.By transferring and mining the experiences,the proposed scheme can improve the robustness to smart attack and the time-varying channel,and decrease the latency and energy consumption of the UAV without knowing the channel model and the jamming model.The simulation results show that the proposed UAV anti-jamming offloading scheme decreases the latency by 43.1%and energy consumption by 35.3%compared with the Wolf-PDS scheme.