Energy Efficiency in Cache-Enabled Small Cell Networks With Adaptive User Clustering

Using a network of cache enabled small cells, traffic during peak hours can be reduced by proactively fetching the content that is most likely to be requested. In this paper, we aim to explore the impact of proactive caching on an important metric for future generation networks, namely, energy efficiency (EE). We argue that, exploiting the spatial repartitions of users in addition to the correlation in their content popularity profiles, can result in considerable improvement of the achievable EE. In this paper, the optimization of EE is decoupled into two related subproblems. The first one addresses the issue of content popularity modeling. While most existing works assume similar popularity profiles for all users, we consider an alternative framework in which, users are clustered according to their popularity profiles. In order to showcase the utility of the proposed clustering, we use a statistical model selection criterion, namely, Akaike information criterion. Using stochastic geometry, we derive a closed-form expression of the achievable EE and we find the optimal active small cell density vector that maximizes it. The second subproblem investigates the impact of exploiting the spatial repartitions of users. After considering a snapshot of the network, we formulate a combinatorial problem that optimizes content placement in order to minimize the transmission power. Numerical results show that the clustering scheme considerably improves the cache hit probability and consequently the EE, compared with an unclustered approach. Simulations also show that the small base station allocation algorithm improves the energy efficiency and hit probability.