Abstract
Unmanned Aerial Vehicles (UAVs)-assisted networks play a pivotal role in both terrestrial base stations (BSs) and non-terrestrial networks (NTNs) due to their extensive coverage and collaborative decision-making capabilities. However, the presence of diverse node types, rapidly evolving requirements, and dynamic channel conditions poses substantial challenges for ground users (GUs) access, particularly in an unknown environment within BS-UAV-NTN integrated networks. To tackle these challenges, this paper introduces a novel approach-a deep Q-learning network (DQN)-based algorithm for UAVs deployment and an adaptive and load balancing (ALB) scheme for GUs access. This paper formulates the GUs access problem in BS-UAV-NTN networks as a maximization problem, transforming it into a Markov Decision Process (MDP) problem for UAVs deployment in unknown environment. The proposed solution includes a DQN-based UAVs deployment algorithm and an access scheme that prioritizes BSs and UAVs. Simulation results convincingly show that this access scheme outperforms traditional Q-learning and random schemes in terms of rewards and the number of accessed GUs.