Cache-Enabled Dynamic Spectrum Access via Deep Recurrent Q-Networks with Partial Observation
Yue Xu, Jiabin Yu, Richard Michael Buehrer · 2019
This paper investigates deep reinforcement learning (DRL) based on Recurrent Neural Networks for Dynamic Spectrum Access (DSA), referred to as a Deep Recurrent Q-Networks (DRQN). The approach uses sensing and cache occupancy as observations. Specifically, we consider a scenario with multiple independent channels and multiple different Primary Users (PU). Three key challenges in our problem formulation are (1) no prior knowledge; (2) prediction based on partial observations, and (3) multi-rate transmission capability. The goal of the DRQN is to learn an optimal channel access strategy to achieve a global objective.