UAV-Assisted Hybrid Throughput Optimization Based on Deep Reinforcement Learning

Abstract

As unmanned aerial vehicles (UAVs) are employed in various areas of modern communication, cellular communication with UAV assistance continues to gain attention. However, it can be significantly complex to optimize a system that considers both the link direction between UAVs and ground users, as well as the location of UAVs. Therefore, we introduce a novel hybrid throughput optimization strategy to address the tightly coupled problem involving adjustments to both UAV height and air-ground link direction. First, the link direction is represented as a table of boolean values and optimized within a discrete space with the heuristic algorithm, where the UAV location remains fixed. Second, we optimize the UAV height with a fixed link direction through the deep reinforcement learning method, which can strengthen the robustness of the algorithm. Finally, the desired result is achieved with a hybrid iteration of the previous two processes. Experimental results demonstrate the effectiveness of the proposed algorithm by achieving a 40% gain on the throughput of the transmission link.