Energy-Saving Predictive Video Streaming with Deep Reinforcement Learning

Dong Liu, Jianyu Zhao, Chenyang Yang

Abstract: In this paper, we propose a policy to optimize predictive power allocation for video streaming over mobile networks with deep reinforcement learning. The objective is to minimize the average energy consumption for video transmission under the quality of service constraint that avoids video stalling. To handle the continuous state and action spaces, we resort to deep deterministic policy gradient to solve the formulated problem. In contrast to previous predictive resource policies for video streaming, the proposed policy operates in an on- line and end-to-end manner. By judiciously designing action and state, the policy can exploit future information without explicit prediction. Simulation results show that the proposed policy can converge closely to the optimal policy with perfect prediction of future large-scale channel gains and outperforms the prediction-based optimal policy when prediction errors exist.

External IDs:dblp:conf/globecom/0003ZY19