Institut d'Électronique et de Télécommunications de Rennes
UMR CNRS 6164







Accueil du site > Accueil > Actualités

Soutenance de thèse

Yifan ZHOU soutient sa thèse intitulée :

Stratégies d’adaptation du trafic de données à la répartition des stations de base dans un réseaux cellulaires.

La soutenance aura lieu mardi 03 juillet 2018 à 10h30 en salle "UEB" à CentraleSupélec, campus de Rennes.

- Résumé : Stratégies d’adaptation du trafic de données à la répartition des stations de base dans un réseaux cellulaires

L’évolution des réseaux radio-mobiles montre que les gains liés à la couche physique ne pourront pas permettre à eux seuls de supporter l’accroissement très rapides des débits. Une façon complémentaire d’accompagner cette évolution est de prendre finement en compte les profils de trafic des utilisateurs (débit demandé, environnement, type de service, …) de façon à adapter en temps réel les caractéristiques des réseaux en fonction des besoins.

La collecte de données issues du terrain est donc dans un premier temps indispensable pour modéliser le trafic de données, qu’il soit temporel, fréquentiel ou spatial et selon les types de services. La première partie de cette thèse se focalise donc sur des mesures de trafic en téléphonie mobile récoltées en Chine (zones urbaines et rurales) et sur les modèles associés. Ces mesures ont alors permis de constater que la loi de classique de répartition des stations de base (selon un processus de Type Poisson Point Process -PPP) n’est pas adaptée.

En recherchant un modèle conjoint liant la distribution des stations de base et le trafic de données (sur des dimensions spatiales et temporelles), la loi α-stable est celle qui s’accorde le mieux avec les mesures du terrain. La deuxième partie de la thèse est plus algorithmique : en utilisant les résultats sur la répartition spatiale des stations de base et la loi α-stable, il a alors été proposé une stratégie coopérative visant à répartir de façon flexible (au niveau temporel et spatial) les contenus les plus demandés par les utilisateurs dans les stations de base et à les échanger entre stations de base de façon à réduire non seulement les temps de transfert (téléchargement) pour les utilisateurs mais aussi la consommation des stations de base.

- Abstract : Clustering Nature of Base Station and Traffic Demand in Cellular Networks and Corresponding Caching and Multicast Strategies

Traditional cellular networks have evolved from the first generation of analog communications to the current fourth generation of digital communications where iteratively enhanced physical layer technologies have greatly increased network capacity. According to the Shannon’s theory, the technical gains brought by the physical layer gradually become saturated, which cannot match the rapid increase of user traffic demand in the current mobile internet era, thus calls for another path of evolution, i.e., digging into the traffic demand of mobile users. In recent years, the academic communities have begun to use the real data to analyze the infrastructure deployment of wireless networks and the traffic demand of mobile users, in order to make benefits from the underlying service patterns. At the same time, along with the recent rise of machine learning technics, data-driven service is considered as the next economic and technical growth point, thus the industry is putting more and more attention to data accumulation and knowledge mining related services. In cellular networks, operators are coming to realize the increasing importance of the recorded data from their own networks.

Therefore, the real-data-driven technology advancing is considered as a promising direction for the next evolution of cellular networks. In this thesis, we firstly give a comprehensive review of the state-of-the-art real data measurement in Chapter 2 which not only sheds light on the importance of real data analysis, but also paves way for its reasonable usage to improve the service performance of cellular networks. From the survey, we conclude that there exhibits a periodic pattern of the temporal traffic assumption for large coverage area in cellular networks, while for single cell, a heavy-tailed distribution is widespread across the temporal and spatial characterization. Furthermore, this imbalance phenomenon emerges more significantly in the call duration, request arrivals and content preference. Then, based on large amount of real data collected from on-operating cellular networks, we conduct a large-scale identification on spatial modeling of BS in Chapter 3. According to the fitting results, we verify the inaccuracy of Poisson distribution for BS locations, and uncover the clustering nature of BS deployment in cellular networks. Although typical clustering models have improved the modeling accuracy but are still not qualified to accurately reproduce the practical BSs distribution scenario, which leads to the spatial density characterization of BS. In Chapter 4, we try to characterize the distribution density of BS deployment and traffic demand, in both spatial domain and temporal dimension. In accordance with the heavy-tailed phenomenon in Chapter 2, we find that the α-Stable distribution is the most accurate model for the BS spatial density, traffic spatial density where a linear dependence is revealed through real data examination. Moreover, the power-law and lognormal distribution for packet length and inter-arrival time are verified, respectively, which again leads to the α-Stable distribution of aggregated traffic volume in cellular networks.

To make benefit from the findings in previous chapters, we proposed a cooperative caching strategy in RAN based on the spatial aggregation of BSs and a unicast/multicast hybrid strategy based on the temporal aggregation of content requests in Chapter 5. According to the theoretical and simulation results, we find that the proposed `Caching as a Cluster’ strategy can significantly reduce the average delay of users especially in the inhomogeneous BS deployment scenario, and the unicast/multicast hybrid transmission strategy can not only reduce the average latency of content requests but also diminishing the average power consumption of BSs especially under the bursty request arrival pattern.

To implement the large-amount real data analysis and dynamically efficient serving mechanism, we propose an intelligent SDN-based heterogeneous cellular network architecture in Chapter 5. With the introduction of an intelligence center, the brand new architecture is able to trace the demand variation in real time, and therefore simultaneously satisfy the operational requirements of the entire network and QoE of all users by deploying fexible and efficient algorithms upon it.






Départements et équipes de recherche
>
Top