Xiao Tuo, Cui Taiping, Islam S M Riazul, Chen Qianbin
School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Nan-An District, Chongqing 400065, China.
Chongqing Key Labs of Mobile Communications, Chongqing 400065, China.
Sensors (Basel). 2020 Dec 31;21(1):215. doi: 10.3390/s21010215.
With the rapid development of mobile communication and the sharp increase of smart mobile devices, wireless data traffic has experienced explosive growth in recent years, thus injecting tremendous traffic into the network. Fog Radio Access Network (F-RAN) is a promising wireless network architecture to accommodate the fast growing data traffic and improve the performance of network service. By deploying content caching in F-RAN, fast and repeatable data access can be achieved, which reduces network traffic and transmission latency. Due to the capacity limit of caches, it is essential to predict the popularity of the content and pre-cache them in edge nodes. In general, the classic prediction approaches require the gathering of users' personal information at a central unit, giving rise to users' privacy issues. In this paper, we propose an intelligent F-RANs framework based on federated learning (FL), which does not require gathering user data centrally on the server for training, so it can effectively ensure the privacy of users. In the work, federated learning is applied to user demand prediction, which can accurately predict the content popularity distribution in the network. In addition, to minimize the total traffic cost of the network in consideration of user content requests, we address the allocation of storage resources and content placement in the network as an integrated model and formulate it as an Integer Linear Programming (ILP) problem. Due to the high computational complexity of the ILP problem, two heuristic algorithms are designed to solve it. Simulation results show that the performance of our proposed algorithm is close to the optimal solution.
随着移动通信的快速发展和智能移动设备的急剧增加，近年来无线数据流量呈爆发式增长，从而给网络注入了巨大的流量。雾无线接入网络（F-RAN）是一种很有前途的无线网络架构，可适应快速增长的数据流量并提高网络服务性能。通过在F-RAN中部署内容缓存，可以实现快速且可重复的数据访问，这减少了网络流量和传输延迟。由于缓存容量有限，预测内容的流行度并在边缘节点中进行预缓存至关重要。一般来说，经典的预测方法需要在中央单元收集用户的个人信息，这引发了用户隐私问题。在本文中，我们提出了一种基于联邦学习（FL）的智能F-RAN框架，该框架不需要在服务器上集中收集用户数据进行训练，因此可以有效确保用户隐私。在这项工作中，联邦学习被应用于用户需求预测，它可以准确预测网络中的内容流行度分布。此外，为了在考虑用户内容请求的情况下最小化网络的总流量成本，我们将网络中的存储资源分配和内容放置作为一个综合模型来处理，并将其表述为整数线性规划（ILP）问题。由于ILP问题的计算复杂度很高，我们设计了两种启发式算法来解决它。仿真结果表明，我们提出的算法性能接近最优解。
