A Game-Theoretic Approach to Cache and Radio Resource Management in Fog Radio Access Networks

Authors:
Publisher website
Google Scholar
Abstract
Fog radio access networks (F-RANs) have been seen as promising paradigms to handle the stringent requirements in the 5G era by utilizing the cache and resource management capabilities of fog access points (FAPs). To achieve better system performance, cache resource and radio resource should be jointly optimized. However, a fully centralized optimization can put heavy burden on the resource manager in the cloud. Faced with this issue, a hierarchical resource management architecture is adopted. Specifically, the resource manager in the upper layer maximizes a long-term utility by optimizing cache resource, which is adaptive to the statistics of channel gains and user content requests. In the lower layer, FAPs self-organize into multiple clusters to mitigate inter-FAP interference in each transmission interval given user content requests, channel gains and cache configuration, and interactions among FAPs are further modeled by a coalition formation game under per-FAP fronthaul capacity constraints. Considering the coupling of FAPs' and the resource manager's strategies and the hierarchy of resource management, the joint cache and radio resource management is formulated as a Stackelberg game with the resource manager and FAPs being the leader and followers, respectively. To achieve Stackelberg equilibrium, a distributed coalition formation algorithm is first developed for FAPs to achieve a stable state. Since there is no closed-form for the leader's objective and the leader's strategy is discontinuous, two model-free reinforcement learning (RL) algorithms are utilized, which can approach a global and a local optimal caching strategy, respectively, taking into account the cluster formation behavior of FAPs. Simulation results show that the proposed cluster formation scheme and multi-agent RL based caching scheme outperform the baselines.
Funding Information
  • State Major Science and Technology Special (2017ZX03001025-006, 2018ZX03001023-005)
  • National Natural Science Foundation of China (61831002, 61728101)
  • NSF (ECCS-1923717)

This publication has 48 references indexed in Scilit: