Machine-Learning Approach for User Association and Content Placement in Fog Radio Access Networks

摘要

The joint user association and cache placement problem is challenging in fog radio access networks (F-RANs) due to its difficulty to present the optimal solution with low complexity. Motivated by the recent development of artificial intelligence, we divide the original optimization problem into two subproblems. In particular, the user association problem is solved by a reinforcement-learning-based algorithm in which the enhanced fog access point content placement profiles and the fronthaul constraint are considered. On the other hand, since the popularity profile of the contents is hard to acquire in practice, a stacked autoencoder-based scheme is presented to predict the content popularity, which considers both the local and global user request status within a specified time interval. Based on the popularity prediction, the edge content placement problem is solved by a deep-reinforcement-learning-based algorithm, aiming at maximizing the F-RAN network payoff. Moreover, the complicated interactions and the cyclic dependency among the short time-scale user association and the long time-scale content popularity prediction and placement problems are studied by applying the Stackelberg game theory. The simulation validates the accuracy of the analytical results and proves that the proposal can further improve the performance of F-RANs.