Spectrum Allocation via Deep Q-learning for 6G Terahertz Band Drone Communications

Efficient resource allocation in Terahertz (THz) drone-to-drone communications is a critical challenge for 6G systems, as the high path loss inherent to the THz band severely hinder the establishment of reliable high-capacity links. Current approaches achieve high capacity but fail in practice due to excessive complexity. In this paper, we introduce a novel framework based on dueling double deep Q-learning that optimizes channel selection, substantially reducing computational overhead while maintaining competitive capacity performance. Simulations under both linearly-aligned and real drone trace scenarios show that our method matches state-of-the-art capacity while reducing complexity by 104, proving its viability for 6G aerial communications.