Optimizing Spectrum Allocation Using Deep Reinforcement Learning in Heterogeneous Wireless Networks

ABSTRACT Especially in heterogeneous wireless networks (HetNets), the demand for efficient spectrum utilization has increased exponentially. Traditional spectrum allocation methods struggle to cope with dynamic environments and underutilized frequency bands. DRQNs are used to intelligently manage spectrum access by secondary users without interfering with that of primary users in this paper. The system model incorporates various types of nodes, including legacy, greedy, and learning‐based agents. DRL's adaptive decision‐making and memory capabilities allow the proposed method to minimize collisions and maximize performance. A simulation study shows that the proposed DRQN‐based method outperforms traditional models, including LTE, SUM, DLMA, and TSRA, on multiple metrics, including packet expiry durations and throughput over episodes.