Deep Deterministic Policy Gradient Based Dynamic Power Control for Self-Powered Ultra-Dense Networks

By densely deploying the base stations (BSs), Ultra Dense Network (UDN) exhibits strong potential to enhance the network capacity, while leading to huge power consumption and a great deal of greenhouse emissions. To this end, power control is regraded as a promising solution to enhance energy efficiency (EE). Without prior knowledge about energy arrival, user arrival and channel state information, we propose a Deep Deterministic Policy Gradient (DDPG)-based EE optimization problem in energy harvesting UDN (EH-UDN), aiming to obtain the optimal power control scheme. The proposed DDPG-based optimization framework is evaluated by comparing with the well-known RL algorithms, i.e., Deep Q-learning Network and Q learning. Numerical results show that the proposed DDPG-based framework is able to enhance EE significantly, and shows strong potential to deal with complicated optimization problems.