L'utilisation de TCP sur des liens asymétriques entraîne fréquemment des débits plus faibles qu'attendus au point de nuire sensiblement à la qualité de service ressentie par l'utilisateur. Ces baisses de performances peuvent prendre diverses formes parmi lesquelles une forte latence en début de connexion, une sous utilisation de la capacité du lien ou encore des latences excessivement hautes pour l'ensemble de la connexion. Afin de contrer ces effets, plusieurs approches sont possibles, que ce soit de bout-en-bout par des modifications de la pile TCP/IP ou en cœur de réseau avec divers mécanismes d'ordonnancement. L'objectif de cette thèse est d'explorer comment un résultat similaire à celui obtenu par des méthodes d'ordonnancement au goulot d'étranglement peut être obtenu en travaillant de bout-en-bout, c'est à dire là où les ressources de calcul et de mémoire sont les plus abondantes. Ce questionnement est accompagné par une analyse en profondeur des phénomènes causant une dégradation des performances, ainsi que l'évaluation des solutions existantes. Finalement, des solutions nouvelles, en cœur de réseau ainsi que de bout en bout, ont été apportées et testées sur banc d'essai. / Using TCP on asymmetric links may lead to unexpected and significant performance drops, severely degrading user experience. Those performance drops can come in various forms, among which a huge latency at the beginning of a connection, under-utilization of link capacities, or even excessive delays for the whole connection.In order to prevent those effects to happen, various approaches exist, either end-to-end through changes in the TCP/IP stack, or in the network core with a collection of scheduling algorithms.The first goal of this thesis is to explore if and how an end-to-end policy (i.e where CPU and memory resources are the most abundant) can achieve similar results as buffering policies in the core of the network. Secondly, we provide an in-depth analysis of the root cause of the performance drops, and evaluate existing algorithms. Finally, new solutions, both end-to-end and in the core of the network, are brought and tested in real life networks.

This dissertation investigates Ultra-Wideband (UWB) techniques for the next generation Radio Frequency Identification (RFID) towards the Internet-of-Things (IoT). In particular, an ultra-high frequency (UHF) wireless-powered UWB radio (UHF/UWB hybrid) with asymmetric links is explored from system architecture to circuit implementation. Context-aware, location-aware, and energy-aware computing for the IoT demands future micro-devices (e.g., RFID tags) with capabilities of sensing, processing, communication, and positioning, which can be integrated into everyday objects including paper documents, as well as food and pharmaceutical packages. To this end, reliable-operating and maintenance-free wireless networks with low-power and low-cost radio transceivers are essential. In this context, state-of-the-art passive RFID technologies provide limited data rate and positioning accuracy, whereas active radios suffer from high complexity and power-hungry transceivers. Impulse Radio UWB (IR-UWB) exhibits significant advantages that are expected to overcome these limitations. Wideband signals offer robust communications and high-precision positioning; duty-cycled operations allow link scalability; and baseband-like architecture facilitates extremely simple and low-power transmitters. However, the implementation of the IR-UWB receiver is still power-hungry and complex, and thus is unacceptable for self-powered or passive tags. To cope with μW level power budget in wireless-powered systems, this dissertation proposes an UHF/UWB hybrid radio architecture with asymmetric links. It combines the passive UHF RFID and the IR-UWB transmitter. In the downlink (reader-tag), the tag is powered and controlled by UHF signals as conventional passive UHF tags, whereas it uses an IR-UWB transmitter to send data for a short time at a high rate in the uplink (tag-reader). Such an innovative architecture takes advantage of UWB transmissions, while the tag avoids the complex UWB receiver by shifting the burden to the reader. A wireless-powered tag providing -18.5 dBm sensitivity UHF downlink and 10 Mb/s UWB uplink is implemented in 180 nm CMOS. At the reader side, a non-coherent energy detection IR-UWB receiver is designed to pair the tag. The receiver is featured by high energy-efficiency and flexibility that supports multi-mode operations. A novel synchronization scheme based on the energy offset is suggested. It allows fast synchronization between the reader and tags, without increasing the hardware complexity. Time-of-Arrival (TOA) estimation schemes are analyzed and developed for the reader, which enables tag localization. The receiver prototype is fabricated in 90 nm CMOS with 16.3 mW power consumption and -79 dBm sensitivity at 10 Mb/s data rate. The system concept is verified by the link measurement between the tag and the reader. Compared with current passive UHF RFID systems, the UHF/UWB hybrid solution provides an order of magnitude improvement in terms of the data rate and positioning accuracy brought by the IR-UWB uplink. / QC 20120110