A Generalized Deep Hierarchical Reinforcement Learning for HAPS Resource Scheduler
Hesam Khoshkbari, Georges Kaddoum, Majid L Altamimi · 2025
In this paper, we introduce a novel generalized deep hierarchical reinforcement learning (DHRL) approach to address the joint user association and channel assignment problem in high-altitude platform station (HAPS)-integrated wireless networks. The proposed method partitions the action space into sub-action spaces, treating each user as an agent. We leverage a deep neural network structure, incorporating convolutional neural network (CNN) layers as a feature extractor, and demonstrate the model’s generalization capability across varying numbers of users without fine-tuning and its ability to adapt to new scenarios through transfer learning. Our proposed DHRL model exhibits strong performance, surpassing the genetic algorithm and achieves nearly identical results as the exhaustive search action selection method in terms of sum-rate maximization.