Multi-hop Diffused Graph Convolution for Resilient UAV Swarm Networks

Massive destructions occurring in Unmanned Aerial Vehicle (UAV) swarm networks can cause the communication network split (CNS) issue. In this paper, we proposed a graph convolution-based framework with destruction embedding and multi-hop diffusing (DEMD), to cope with the CNS while mitigating the over-aggregation problem in existing recovery methods. First, a destruction embedding module is presented to map the features of destroyed nodes into the graph of remained nodes, providing the directions towards the empty space left by destruction. Secondly, the DEMD utilizes a multi-hop diffusing module to reconstruct the graph edges, which allows nodes to build wider connections with k-hop neighbors beyond the communication range. Thirdly, a graph convolution operation based backbone is developed to infer the graph that contains destruction information and multi-hop relationships, aiming to generate the recovery trajectories for remained UAVs. Moreover, a joint loss function is designed to take both recovery time and direction constraints into consideration. Experimental results show that the DEMD can not only shorten the recovery time but also demonstrate a significant improvement in spatial coverage ratio and topological degree distribution.