This thesis explores the queueing performance of a wireless communication system that transmits packets over a correlated erasure channel using the IEEE 802.11 protocol suit. The channel states and the queue length together form a Markov chain. Exploiting this mathematical structure, the probability of the queue exceeding a certain threshold can be obtained. Most previous contributions in this area treat code-rate selection, channel erasure probability and network congestion separately. In this thesis, a simple integrated approach, which jointly considers these factors, is introduced. This approach becomes especially valuable for capturing the performance of delay-sensitive communication systems over time-varying channels. This thesis starts with a review of related work about correlated bit-erasure wireless channel models. A numerical study is then conducted to demonstrate the importance of optimizing overall system performance, and how this process impacts error-control coding at the physical layer. Following this exercise, a packet-erasure channel model with a Poisson arrival process is analyzed. The Baum-Welch algorithm is subsequently presented as a means to estimate the parameters of wireless communication systems. Furthermore, a matrix geometric method for obtaining the stationary distribution of the ensuing Markov chain is discussed. This offers a new perspective on wireless communication in the context of delay-sensitive applications. To complement the analysis platform put forth in this work, illustrative numerical results are contained in the last section of the thesis. From these results, design guidelines for improving the performance of delay-sensitive wireless communication systems are established. Although these results are obtained under simplifying assumptions, the overall methodology applies to more general situations, especially for wide-band delay-sensitive wireless communication applications.

The morphology, composition, and magnetic properties of NiFe thin films were characterized. Films with thicknesses up to 137 nm were deposited in an RF induction evaporator at high vacuum (10^-8 mbar). Time resolved magneto-optic Kerr effect microscopy (TR-MOKE) was used to measure the Gilbert damping constant, an important dynamic magnetic property with applications to magnetic data storage. The composition of each film was measured with energy-dispersive X-ray (EDX) microscopy and used to determine the weight percent of Ni and Fe in each film. A trend of increased damping with increased thickness was found, in agreement with published results. Magnetic properties and roughness were found to differ significantly from previous films grown in the same vacuum chamber by Rudge, and are attributed to different growth modes produced by differing deposition conditions. However, the weight percent of Ni in each film was found to be inconsistent, deviating by up to 7% from the Ni80Fe20 evaporation source. Inconsistent composition, caused by the inability to control deposition parameters, prevents insight into Gilbert damping from being drawn from the analysis. / Graduate

Thesis (M.A. (Art History)) -- University of Adelaide, 2002. / Bibliography: p. leaves 62-68.

Theses (M.A.)--Carleton University, 2003. / Includes bibliographical references (p. 134-148). Also available in electronic format on the Internet.

Thesis (Ph. D.)--University of Washington, 1996. / Vita. Includes bibliographical references (leaves 512-520).

Thesis (Ph. D.)--Florida State University, 1973. / Typescript. eContent provider-neutral record in process. Description based on print version record. Includes bibliographical references (leaves [170]-177).

Thesis (Ph.D.)--McMaster University, 2004. / Advisor: Peter Walmsley. Includes bibliographical references (leaves 206-216). Also available via World Wide Web.

Th. 3e cycle--Litt. fr.--Toulouse 2, 1985.

Orientador: Arlete Moyses Rodrigues / Tese (doutorado) - Universidade Estadual de Campinas, Instituo de Filosofia e Ciencias Humanas / Made available in DSpace on 2018-07-28T15:55:40Z (GMT). No. of bitstreams: 1 Santos_MariaElisabetePereirados_D.pdf: 21542720 bytes, checksum: 5dc0ff31b4d337c2623ded8b2d984cda (MD5) Previous issue date: 1999 / Doutorado

Cette recherche se donne comme défi de retracer les enjeux normatifs liés au développement du sous-domaine de l’intelligence artificielle appelé machine learning. Bien que celui-ci connaisse un impressionnant regain d’intérêt depuis le début du XXIe siècle, et ce à travers un nombre croissant d’activités sociales, son émergence remonte au moins à la première moitié du XXe siècle. En effet, l’idée de concevoir des machines capables de modéliser l’apprentissage organique est largement concomitante au projet cybernétique de fonder une science du contrôle et de la communication. A force de modéliser l’apprentissage, il devient progressivement possible d’envisager à partir des années 1940 et 50 que les machines elles-mêmes seraient capables d’apprendre. C’est par l’observation des comportements des machines qu’une formalisation algorithmique de l’apprentissage s’impose. A partir de cette mise en scène, nous chercherons à montrer en quoi l’apprentissage algorithmique n’est pas nécessairement une question d’automatisation de l’apprentissage et que le comportement permet, au contraire, de problématiser la normativité des algorithmes apprenants en ce qu’ils participent activement à certaines activités sociales (travail, enseignement, guerre, etc.). Il s’agira, à partir des philosophies de Georges Canguilhem et Gilbert Simondon, de montrer en quoi les effets normatifs induits par l’apprentissage algorithmique, comme mode de gouvernement au sein d’activités sociales données, peuvent se comprendre à l’aune du concept de "répertoire de comportements potentiels". / Doctorat en Philosophie / info:eu-repo/semantics/nonPublished