Double Deep Q-Network-Based Energy-Efficient Resource Allocation in Cloud Radio Access Network

Cloud radio access network (CRAN) has been shown as an effective means to boost performance. Such gain stems from the intelligent management of remote heads (RRHs) in terms on/off operation mode and power consumption. Most conventional resource allocation (RA) methods, however, optimize utility without considering switching overhead RRHs adjacent time intervals. When environment becomes time-correlated, mathematical optimization is not directly applicable. In this paper, we aim energy efficiency (EE) subject constraints on per-RRH transmission user data rates. To end, formulate EE problem a Markov decision process (MDP) subsequently adopt deep reinforcement learning (DRL) technique reap cumulative rewards. Our starting point Q (DQN), which combination Q-learning. each slot, DQN configures status yielding largest Q-value (known state-action value) prior solving minimization for active RRHs. overcome overestimation issue DQN, propose Double (DDQN) framework that obtains optimal reward better than by separating selected action target generator. Simulation results validate DDQN-based RA method more energy-efficient DQN-based algorithm baseline solution.