Energy-aware Routing Protocol for UAV Electronic Warfare using Graph Attention and Fuzzy Reward

The past few years have witnessed a remarkable leap forward in the tactical position of UAV swarm in aerial electronic warfare. Among them, energy-aware packet routing is one of the fundamental problems for cooperation between multiple UAVs to complete combat missions in complex battlefield environments. Recently, deep reinforcement learning (DRL) technique provides a new opportunity to networks related applications, including routing, resource allocation and network access. However, most existing DRL-based routing protocols are difficult to adapt to the changes of network scale, which have weak generalization capabilities and rely on centralized trainers. Thus, these protocols cannot be directly applied to the aerial electronic warfare. In this paper, we propose an adaptive and scalable routing protocol for UAV swarm electronic warfare with graph attention based fully distributed multi-agent reinforcement learning. Besides, a reward function design method based on fuzzy logic is proposed to reduce the probability of abnormal behaviors performed by agents. The simulation results show that our protocol can make effective routing decisions in dynamic wireless multihop networks and enhance the system performances in terms of packet delivery ratio, end-to-end delay and throughput.