TY: THES T1 - Intelligent and transparent resource management A1 - Capela, Nelson Filipe N2 - Wireless access networks have become available almost everywhere. In the same area we can have access to several networks of the same or di erent technologies that can present di erent characteristics. Alongside the evolution of access networks, we have the evolution of terminals. These are currently equipped with a multitude of wireless interfaces, easier to carry and more accessible to users. In uenced by these features, users began to change the way they use their devices to obtain information. The introduction of these new terminals in environments with several networks brings up a high number of new opportunities. The way that available resources are addressed by the users' terminals needs to evolve to a new level. Users' requirements are increasingly stringent and di cult to ful l; however, even with several available networks at terminals' range, they only take advantage of one of them. In this sense, this Thesis proposes a new reliable, exible, context-, resource-, and mobility-aware architecture that can provide e cient communication in heterogeneous and dynamic environments, using all the available networks resources. The Thesis starts with an analytical analysis in terms of the multihoming and network coding impact on typical wireless networks. It is proposed a new approach that combines these concepts to improve the allocation of network resources and the communication process in heterogeneous technologies. The outcome of this study shows the advantages of using multihoming and network coding in terms of the system performance. Next, we evolve this analytical analysis and propose a dynamic architecture capable of integrating multihoming, mobility and context-information. This is then implemented and evaluated in real scenarios with both Wi-Fi and cellular networks. After the implementation of a functional prototype based on information extracted from the environment, we propose a machine learning process able to predict information based on previous actions. The learning mechanism extracts the required information, creates its owns databases in a dynamic way and identi es when the existing information is enough to perform a good prediction. The outcome of this study demonstrates the signi cant reduction of the overhead, enabling a more scalable process. Finally, we go further in the resources management and integrate and evolve our approach to a real vehicular environment, assuming both the single-hop and multihop con gurations. It is developed the capability to split the tra c through the di erent networks, based on the type of tra c, the networks and vehicles characteristics. Furthermore, it is proposed both a downlink and an uplink multihoming approach and integrated the use of network coding. Through this study, we can observe the positive impact of our resource management approach in the network performance and in the vehicles mobility. UR - https://ria.ua.pt/handle/10773/22720 Y1 - 2017 PB - Universidade de Aveiro