부산대학교 도서관

Coding-Aware routing algorithm improves the performance of Network Coding (NC) by selecting the paths with more coding opportunities, thus increasing the throughput of network transmission. However, the Unmanned Aerial Vehicle (UAV) flight in Flying Ad-hoc Network (FANET) causes network topology changes and calculating coding opportunities is NP-hard complex problem, so the performance of traditional coding-aware routing is limited. We try to solve this problem using reinforcement learning and propose a deep reinforcement learning-based coding-aware routing (RLCAR) that maximizes throughput and coding opportunities. In RLCAR, we define new concepts of coding benefits, throughput and energy distribution, and we put these parameters into the routing metric (Reward function). We evaluate the performance of RLCAR and compare the throughput, packet delivery ratio (PDR), coding/decoding rate with other coding-aware routings. The results show that RLCAR acquires more coding opportunities and improves the throughput of FANET. And, RLCAR can cope with the dynamic changes of UAV topology.