Data intensive applications, requiring reliability and strict delay constraints, have emerged recently and they necessitate a different approach to analyzing system performance. In my work, I establish a framework that relates physical channel parameters to the queueing performance for a single-user wireless system. I then seek to assess the potential benefits of multirate techniques, such as hybrid-ARQ (Automatic Repeat reQuest), in the context of delay-sensitive communications. Present methods of analysis in an information theoretic paradigm define capacity assuming that long codewords can be used to take advantage of the ergodic properties of the fading wireless channel. This definition provides only a limited characterization of the channel in the light of delay constraints. The assumption of independent and identically distributed channel realizations tends to over-estimate the system performance by not considering the inherent time correlation. A finite-state continuous time Markov channel model that I formulate enables me to partition the instantaneous data-rate received at the destination into a finite number of states, representing layers in a hybrid-ARQ scheme. The correlation of channel has been incorporated through level crossing rates as transition rates in the Markov model. The large deviation principle governing the buffer overflow of the Markov model, is very sensitive to channel memory, is tractable, and gives a good estimate of the system performance. Metrics such as effective capacity and probability of buffer overflow, that are obtained through large deviations have been related to the wireless physical layer parameters through the model. Using the above metrics under QoS constraints, I establish the quantitative performance advantage of using hybrid-ARQ over traditional systems. I conduct this inquiry by restricting attention to the case where the expected transmit power is fixed at the transmitter. The results show that hybrid-ARQ helps us in obtaining higher effective capacity, but it is very difficult to support delay sensitive communication over wireless channel in the absence of channel knowledge and dynamic power allocation strategies.

No description available.

Dans cette thèse, nous nous sommes intéressés à l'analyse et la conception de réseaux étudiés sont les réseaux de fibres optiques et les réseaux de capteurs. Les problématiques étudiées sont: pour les réseaux de fibres optiques, minimiser le coût de déploiement et assurer la qualité de service; dans les réseaux de capteurs, garantir la sécurité des transmissions et l'énergie consommée par le routage des communications. Pour résoudre ces problèmes nous utilisons des techniques de théorie des graphes, de complexité, de programmation linéaire. Le premier problème consiste à concevoir un plan d'installation de fibres optiques de coût minimal permettant de connecter un ensemble de clients à un noeud de raccordement via un ensemble de coupleurs en respectant les contraintes technologiques imposées par la norme. Nous proposons une modélisation de ce problème ainsi qu'une méthode de résolution. Le deuxième problème est un problème de flot avec contrainte de délai où le délai pour traverser une liaison est proportionnel à la quantité de flot quicircule sur celle-ci. Nous proposons une preuve de NP-complétude dans le cas général, un algorithme d'approximation facteur 2 dans le cas où le graphe support est un chemin et une heuristique évaluée de façon expérimentale qui calcule en un temps raisonnable de bonnes solutions pour des instances de tailles réelles. Enfin, nous proposons deux protocoles concernant les réseaux de capteurs. Le premier, basé sur un algorithme distribué, calcule un ensemble de chemins disjoints entre les terminaux. Le second maximise la durée de vie d'un réseau de capteurs alimentés par batteries. Des résultats s'expérimentations numériques sont présentés. / In this thesis, we study some optimization problem applied to telecommunication networks. We study fiber optical networks and sensor networks. We are interested to analysis and design for these types of networks. The issues studied are: for fiber optic networks, minimize the cost of deployement and ensure quality of service; for sensors network, ensure the safety of transmissions and the energy consumed. To solve these problems we use techniques as graphs theory, complexity, linearprogramming, generalized flows and paths with resource constraints. The first problem is to minimize the cost to deploy a fiber optical network which connect a set of customers to a connection node through a set of splitter and deal about technological constraints imposed by the standard. We propose a model and a method of resolution for this problem. The second problem is a flow problem with delay constraint where time to cross a edge is proportional to the amount of flow that flows thereon. We offer a proof of NP-completeness in the general case, an approximation algorithm factor 2 in the case where the support graph is a path and an estimated experimentally an heuristic that calculates good solutions for instances of real sizes. Finally, we propose two protocols for sensor networks, which resulted in two patents. The first, based on a distributed algorithm, calculates a set of disjoint paths between terminals. The second maximizing the lifetime of sensors powered by batteries. The results of numerical experiments are also presented.