Learning-based dynamic connectivity maintenance for UAV-assisted D2D multicast communication

Abstract

Unmanned aerial vehicles (UAVs) enable flexible networking functions in emergency scenarios. However, due to the movement characteristic of ground users (GUs), it is challenging to capture the interactions among GUs. Thus, we propose a learning-based dynamic connectivity maintenance architecture to reduce the delay for the UAV-assisted device-to-device (D2D) multicast communication. In this paper, each UAV transmits information to a selected GU, and then other GUs receive the information in a multi-hop manner. To minimize the total delay while ensuring that all GUs receive the information, we decouple it into three subproblems according to the time division on the topology: For the cluster-head selection, we adopt the Whale Optimization Algorithm (WOA) to imitate the hunting behavior of whales by abstracting the UAVs and cluster-heads into whales and preys, respectively; For the D2D multi-hop link establishment, we make the best of social relationships between GUs, and propose a node mapping algorithm based on the balanced spanning tree (BST) with reconfiguration to minimize the number of hops; For the dynamic connectivity maintenance, Restricted Q-learning (RQL) is utilized to learn the optimal multicast timeslot. Finally, the simulation results show that our proposed algorithms perform better than other benchmark algorithms in the dynamic scenario.