Performance analysis and 3D position deployment for V2V-assisted UAV communications in vehicular networks

Deploying unmanned aerial vehicles (UAVs) as flying base stations (BSs) is a promising solution to alleviate the burden of communication infrastructure. However, few works have been done to theoretically analyze the system performance, especially for the vehicle-to-vehicle (V2V) assisted vehicular network, where vehicle users (VUs) can request contents from either the UAV or other VUs that the UAV has successfully served. We first consider the single hot-spot scenario and analyze the system performance in terms of the VU's successful service probability (SSP). Then we extend our analysis to the multi-hot-spot scenario where the hot-spots merge and split due to VU mobility. For this scenario, the SSP and the average number of successfully served VUs (SSVs) are theoretically derived, based on which a UAV's 3D position deployment problem is formulated. Due to the complicated formulation of SSP and the dynamic environment including hot-spot distribution and UAV's battery state, the closed-formed solutions to the UAV's deployment are intractable. To obtain the sub-optimal positions of the UAV, the proposed problem is reformulated into a deep reinforcement learning (DRL) framework, and a pre-trained deep Q-network (DQN) based scheme is proposed. Simulation results demonstrate that compared to existing schemes, our proposed scheme achieves a high number of successfully served users with moderate energy consumption.