Méthodes d'Accès au Canal pour les Réseaux Dédiés à l'Internet des Objets
Abderrahman Ben Khalifa · HAL (Le Centre pour la Communication Scientifique Directe) · 2020
Dedicated networks for the Internet of Things appeared with the promise of connecting thousands of nodes, or even more, to a single base station in a star topology. This new logic represents a fundamental change in the way of thinking about networks, after decades during which research work mainly focused on multi-hop networks.Internet of Things networks are characterized by long transmission range, wide geographic coverage, low energy consumption and low set-up costs. This made it necessary to adapt the protocols at different architectural layers in order to meet the needs of these networks.Several players compete in the Internet of Things market, each trying to establish the most efficient solution. These players are mostly focused on modifying the physical layer, on the hardware part or through proposing new modulations. However, with regard to the channel access control solution (known as the MAC protocol), all the solutions proposed by these players are based on classic approaches such as Aloha and CSMA.The objective of this thesis is to propose a dynamic MAC solution for networks dedicated to the Internet of Things. The proposed solution has the ability to adapt to network conditions. This solution is based on a machine learning algorithm that learns from network history in order to establish the relationship between network conditions, MAC layer parameters and network performance in terms of reliability and energy consumption. The solution also has the originality of making possible the coexistence of nodes using different MAC configurations within the same network. The results of simulations have shown that a MAC solution based on machine learning could take advantage of the good properties of different conventional MAC protocols. The results also show that a cognitive MAC solution always offers the best compromise between reliability and energy consumption, while taking into account the fairness between the nodes of the network. The cognitive MAC solution tested for high density networks has proven better scalability compared to conventional MAC protocols, which is another important advantage of our solution.