DeepCachNet: A Proactive Caching Framework Based on Deep Learning in Cellular Networks

Abstract

Content caching at the network edge is considered to be a suitable technique for enhancing the efficacy of content delivery in cellular networks. Caching the strategic content at the SBS is critical due to storage constraints; however, it requires information about popularity dissemination that is not known in advance. Furthermore, the popularity of content is varied due to the fact that every mobile user connected to the SBSs has a different preference for content. Thus, the nature of the content a user prefers depends on the features of both user and content. This article proposes a novel deep learning-based proactive caching framework in cellular networks, called DeepCachNet, in which a vast amount of data is collected from the mobile devices of users connected to SBSs. The deep-learning methods called auto-encoder and stacked denoising autoencoders are applied to the collected data to extract the features of users and content, respectively. The extracted features are then used to estimate the content popularity at the core network. Based on the estimated content popularity, the strategic content is cached at SBSs to obtain higher backhaul offloading and user satisfaction. To validate the effectiveness of the proposed framework, a case study is carried out in which mobile data are gathered from connected mobile devices by using a developed android mobile application, and a simulation of the proposed framework is performed on the collected data. The results of the simulation show that the framework resolves the cold-start and data sparsity problem, and yields significant improvements in terms of backhaul offloading and the user satisfaction ratio. It achieves gains of up to 6.2 percent and 30 percent for backhaul offloading and user satisfaction, respectively.