Clustered device-to-device caching based on file preferences

Proactive caching at the mobile network edge has been considered as a promising technology for enhancing users' Quality of Experience (QoE) and reducing redundant transmissions over the already overburdened cellular networks. The problem of video file caching in wireless Device-to-Device (D2D) communication networks, in which mobile users designated as helper users store popular video files and serve other requesting users via D2D localized transmissions, is studied in this paper. As personalized video recommendation systems are widely applied in video sites such as YouTube and Netflix, they cause mobile users' diversification and individuation in file preferences and users may make selfish caching decisions. Moreover, designing the file placement in caches is a task of hugely computational complexity due to the vast number of involved files and users. In this paper, we simultaneously cluster users and files into different interest groups and then propose a greedy intra-cluster caching scheme to greatly reduce its complexity. And we also compare the performance of each clustering algorithm while the file preferences matrix becomes high dimensional, sparse and highly asymmetric. Simulation results confirm that, with markedly reduced complexity, our proposed greedy caching scheme with spectral clustering using cosine similarity as the distance measure achieves near-optimal delay performance.