Distributed Caching Methods in Small Cell Networks

Publication Type:



CentraleSupélec, Université Paris-Saclay, Volume PhD, p.170 (2015)


This thesis explores one of the key enablers of $5$G wireless networks leveraging small cell network deployments, namely proactive caching. Endowed with predictive capabilities and harnessing recent developments in storage, context-awareness and social networks, peak traffic demands can be substantially reduced by proactively serving predictable user demands, via caching at base stations and users' devices. In order to show the effectiveness of proactive caching techniques, we tackle the problem from two different perspectives, namely theoretical and practical ones.

In the first part of this thesis, we use tools from stochastic geometry to model and analyse the theoretical gains of caching at base stations. In particular, we focus on 1) single-tier networks where small base stations with limited storage are deployed, 2) multi-tier networks with limited backhaul, and) multi-tier clustered networks with two different topologies, namely coverage-aided and capacity-aided deployments. Therein, we characterize the gains of caching in terms of average delivery rate and mean delay, and show several trade-offs as a function of the number of base stations, storage size, content popularity behaviour and target content bitrate.

In the second part of the thesis, we take a more practical approach of proactive caching and focus on content popularity estimation and algorithmic aspects. In particular: 1) We first investigate the gains of proactive caching both at base stations and user terminals, by exploiting recent tools from machine learning and enabling social-network aware device-to-device (D2D) communications; 2) we propose a transfer learning approach by exploiting the rich contextual information extracted from D2D interactions (referred to as source domain) in order to better estimate the content popularity and cache strategic contents at the base stations (referred to as target domain); 3) finally, to estimate the content popularity in practice, we collect users' real mobile traffic data from a telecom operator from several base stations in hours of time interval. This amount of large data falls into the framework of big data and requires novel machine learning mechanisms to handle. Therein, we propose a parallelized architecture in which content popularity estimation from this data and caching at the base stations are done simultaneously.

Our results and analysis provide key insights into the deployment of cache-enabled small base stations, which are seen as a promising solution for $5$G heterogeneous cellular networks.