Caching popular contents at the edge of mobile networks is an effective technology to relieve burden on backhaul links and reduce average transmission delay. However, the effectiveness of content caching depends on the cache-hit probability, and thus the content caching strategy becomes an important topic. This paper jointly addresses user scheduling and content caching issues with the purpose of minimizing the average transmission delay in the mobile edge network. Since network states are unknown random variables, the reinforcement learning (RL) is adopted to learn the optimal stochastic policy through interactions with the network environment. The actor of RL agent uses the Gibbs distribution as the probabilistic caching policy, and the parameters of the policy are updated with the gradient ascent method. The critic of RL agent uses a deep neural network to approximate the value function and helps the actor estimate the gradient of the policy. The techniques of experience replay and fixed target network are used to guarantee convergence and stability of the learning algorithm. Simulation results are shown to illustrate the superior performance of the proposed approach.

Additional Metadata
Persistent URL dx.doi.org/10.1109/ICCW.2018.8403711
Conference 2018 IEEE International Conference on Communications Workshops, ICC Workshops 2018
Citation
Wei, Y. (Yifei), Zhang, Z. (Zhiqiang), Yu, F.R, & Han, Z. (Zhu). (2018). Joint user scheduling and content caching strategy for mobile edge networks using deep reinforcement learning. In 2018 IEEE International Conference on Communications Workshops, ICC Workshops 2018 - Proceedings (pp. 1–6). doi:10.1109/ICCW.2018.8403711