By deploying small cells (i.e., femtocells) within the same service area of regular macrocells, a much higher area spectral efficiency, better indoor coverage, and significant mobile data offloading can be achieved at low cost. Because femtocells reuse the radio spectrum already assigned to macrocells in an uncoordinated fashion, new cell boundaries are created and the signal interference situation becomes far more unpredictable than in conventional networks. In this heterogeneous network setting, adaptive power allocation and dynamic spectrum access are needed to ensure a harmonized coexistence of network entities with diverse design specifications. Since femtocells are deployed by end users without any network planning, autonomous solutions are desirable to effectively control the severe intra-tier and cross-tier interferences. In this study, we develop and evaluate distributed radio resource allocation algorithms for wireless heterogeneous networks employing code-division multiple access (CDMA) and orthogonal frequency division multiple access (OFDMA). Requiring no central coordination, the devised solutions robustly protect the ongoing operation of all existing macrocell users, while optimally exploiting the residual network capacity for femtocells. In CDMA-based networks, we propose a dynamic pricing scheme combined with femtocell user admission control to indirectly manage the cross-tier interference. The proposed joint power and admission control algorithms can be locally executed at each link to offer a maximum utility to individual users. For the total network utility maximization, we develop joint Pareto-optimal power control and signal-to-interference-plus-noise ratio (SINR) assignment algorithms that can fairly share radio resources among users. Upon applying convex optimization methods, the minimum SINRs prescribed by macrocell users are effectively enforced, whereas the sum network utility is globally maximized. In OFDMA-based networks, to resolve the highly nonconvex and combinatorial problem of joint power and subchannel allocation, we propose a dynamic spectrum management scheme that alternatively optimizes the powers and assigns the subchannels. With the adopted successive convex approximation approach, the total throughput of all femtocells is maximized whilst the macrocell network capacity is always protected. In cognitive femtocells where femtocell users opportunistically access the spectrum licensed to macrocells, we apply Lagrangian duality to devise jointly optimal power and subchannel allocation algorithms. The distributed solutions are shown to achieve their global optimality with low complexity. / En déployant de petites cellules (dénommées les femtocells) au sein de la même zone de service que les cellules de tailles régulières (dénommées cellules macro), une efficacité spectrale zone beaucoup plus élevé, une meilleure couverture à l'intérieur, et d'importantes données mobiles de déchargement entre les deux cellules peuvent être réalisé tout en gardant faible coût. Vu que les femtocells réutilisent le spectre de fréquence déjà consacré à la cellule macro, auxquelles ils sont assignés, d'une manière non-coordonnée, de nouvelles limites de cellules sont créées et l'interférence devient beaucoup plus imprévisible que dans les réseaux traditionnels. Dans ce contexte réseau hétérogène, une allocation adaptative de puissance et des méthodes d'accès dynamiques au spectre sont nécessaires pour assurer une coexistence harmonisée des entités du réseau avec les nouvelles spécifications imposées par les femtocells. Depuis que les femtocells sont déployées par les terminaux sans aucune planification au préalable du réseau, des solutions qui s'adaptent automatiquement sont toujours désirable pour contrôler efficacement les sévères interférences entre les différents niveaux du réseau sans fil hétérogène.Dans cette étude, nous développons et évaluons des algorithmes distribués pour l'allocation de ressources radio dans les réseaux sans fil hétérogènes employant l accès multiple par répartition en code (CDMA) et Accès multiple par répartition en fréquence (OFDMA). En évitant une coordination centralisée, les solutions proposées protègent le fonctionnement de tous les utilisateurs de la cellule macro existantes, tout en exploitant de manière optimale la capacité résiduelle du réseau pour les utilisateurs du femtocells. Dans les réseaux CDMA, nous proposons un schéma de tarification dynamique associé à un contrôle d'admission des utilisateurs de la femtocell nous permettant de gérer indirectement l'interférence inter-niveaux (entre cellule macro et femtocell). Le contrôle simultané de la puissance et les algorithmes de contrôle d'admission proposés peut être exécuté localement sur chaque lien pour offrir un maximum d'utilité pour les utilisateurs individuels. Pour maximiser l'utilité totale du réseau, nous développons un algorithme de contrôle simultané de puissance basant sur l'optimalité de Pareto et le rapport signal sur interférence plus bruit (SINR) qui peut partager équitablement les ressources radio entre les utilisateurs. En appliquant d'une méthode d'optimisation, les SINR minimaux prescrits par les utilisateurs des cellules macro sont garantis, alors que le maximal global la somme de l'utilité du réseau est trouvé.Dans les réseaux OFDMA, afin de résoudre le problème non convexe et combinatoire de l'allocation conjointe de la puissance et des sous-porteuses, nous proposons un schéma alternatif de gestion dynamique du spectre qui optimise la distribution de puissance et des sous-porteuses. Avec l'approche par approximations successives convexe adoptée, le débit total de tous les femtocells est maximisé alors que la capacité du réseau de la cellule macro est toujours protégée. En femtocells cognitives où les utilisateurs du femtocell accèdent au spectre autorisé à la cellule macro d'une manière opportuniste, nous appliquons la dualité lagrangienne pour optimiser la distribution de la puissance et des sous-porteuses. Nous prouvons que les solutions distribuées proposées atteignent leur optimal global avec une faible complexité.

Many questions of interest in various fields ranging from machine learning to computational biology and finance require the solution of an optimization problem. Frequently such problems are classified as large scale in the sense that they involve complex computations over very large datasets. The increasing interest in distributed optimization algorithms is motivated by two main reasons. First, the problem complexity pushes today's processors to their limits and the need for distributed algorithms arises quite naturally. A second and more practical reason is that sometimes the data is collected in a distributed manner and transmitting it to a single location is either too costly or violates privacy. The starting point of this thesis is the simple realization that the main difference between a serial and a distributed algorithm is that in the latter, a processor needs to exchange messages over a network to access another processor's information. Network communication is in general less reliable and orders of magnitude slower than local disk accesses. Furthermore, in an arbitrary network topology to achieve global performance messages might need to travel over multiple hops. Finally, the hardware's capabilities also limit the ways in which a distributed algorithm may be implemented. All these factors highlight the important role of the network in distributed optimization algorithms. To understand that role we focus on the class of consensus-based distributed optimization algorithms. Those algorithms admit an elegant theoretical analysis while remaining easy to implement. In addition they tend to be scalable and robust to communication delays. The contributions of this work can be grouped into four important areas: 1) understanding the communication/computation tradeoff and its effect on scalability with the network size, 2) understanding the limitations of the network and the necessary features that distributed algorithms need to possess to be practical, 3) understanding the effects on convergence of network-induced communication delays and 4) understanding the theoretically achievable convergence rates of distributed algorithms. These areas impact the design and deployment of any consensus-based distributed optimization algorithm. / Il y a une grande variété de domaines d'apprentissage automatique, de la biologie à la finance, où l'on doit résoudre des problèmes d'optimisation. Souvent ces problèmes impliquent des calculs complexes sur de vastes ensembles de données. Pour résoudre ces problèmes le développement d'algorithmes distribués est devenu très populaire pour deux raisons. Premièrement, la complexité des problèmes pousse les processeurs actuels à leurs limites. Naturellement, il devient essentiel d'utiliser des systèmes distribués. La deuxième raison est que la collecte des données est parfois distribuée, et il est difficile, coûteux ou en violation d'accords de confidentialité de transférer toutes les données au même endroit. La fondation de cette thèse est la réalisation simple que la grande différence entre un algorithme centralisé et un algorithme distribué est que ce dernier utilise un réseau pour permettre l'échange d'information d'un processeur à un autre. Généralement, la communication sur un réseau est moins fiable et beaucoup plus lente que l'accès d'information sur un disque local. De plus, pour une topologie de réseau arbitraire, la communication de messages nécessite plusieurs sauts. Finalement, les capacités matérielles aussi limitent les façons par lequelles les algorithmes distribués théoriques peuvent être mis en oeuvre. Tous ces facteurs, rappellent l'importance du réseau. Pour comprendre cette importance, nous nous concentrons, sur la classe des algorithmes de consensus pour optimisation distribuée. Ces algorithmes aussi possèdent des analyses théoriques très élégantes tout en restant faciles a mettre en oeuvre. Aussi, ils sont robustes aux délais de communication et extensibles. Les contributions de cette thèse peuvent être classifiées selon les quatres catégories suivantes: 1) comprendre le compromis entre communication et calculs locaux, et l'extensibilité avec la taille du réseau, 2) comprendre les limites posées par le réseau aux fonctionnalités nécessaires que chaque algorithme distribué doit possèder en pratique, 3) comprendre les effets de délais de communication et les propriétés de convergence de ces algorithmes en présence de délais, 4) comprendre les taux théoriques de convergence des algorithmes d'optimisation distribués. Tous ces domaines affectent la conception et le déploiement de chaque algorithme de consensus d'optimisation distribué.

Building novel Digital Music Instruments (DMIs) requires the use of a variety of sensors to transduce human actions to electronic signals that will control sound synthesis variables. Among them, contact (force/pressure) sensors such as the Force Sensing Resistors (FSR) are the most commonly used in the design of DMIs. Similar commercial sensor designs based on conductive materials also enable to detect position, displacement or flexion. Unfortunately, although commercial sensors can be easily found at electronic resellers, only a few standardised models with predefined sizes, shapes and electric characteristics can be purchased. Moreover, the quality of the sensor appears to depend on their shape and dimension. These limitations have direct implications on the design of novel DMIs, which actually need to be adapted to the existing offer of commercial sensors. After various investigations with polymers, textiles, adhesives and glues, we finally chose to focus on conductive paper as a generic material for producing sustainable, flexible and customisable contact sensors. This thesis study presents two directions on paper sensors investigations: We first began by studying the potential of building paper contact sensors using industrial paper. This type of paper is originally produced to provide optimal colour properties and is not optimised for use as a conductive material. Nevertheless, it has been shown that it provides a very efficient choice for the design of custom sensors. Several sensor prototypes were built using samples of four industrial papers from three main manufacturers (ArjoWiggins (Canson), Fabriano, and PASCO) and have been used in the design of various gesture controllers. In order to verify the electrical behaviour of these prototypes, we developed a special test machine that enabled us to characterise the sensor's properties in terms of repeatability, drift and hysteresis, and to compare the results with commercial sensors. The second step was to consider how to produce such a paper. Laboratory experiments were performed to evaluate the influence of the papermaking process as well as the chemicals to be added to optimise the retention and formation of a uniform mix of pigments and fibres, enabling the fabrication of a rough, porous, compressible but still elastic material. Papers with various amounts of pigment and different sheet structures were made and characterised in order to evaluate the influence of the paper characteristics onto their electrical resistance. Resulting materials were also characterised with the test machine and compared to other types of resistive materials. This research contributes to the promotion of alternative solutions for the development of contact sensors. In particular, it shows that conductive paper and inks are an ecological alternative to conducting polymers for producing custom contact sensors. Paper sensors can replace industrial sensors in many applications, as they are equally efficient. Moreover they are recyclable or reusable, flexible and customisable, therefore extending the capabilities of industrial solutions. The production of such a paper will provide researchers with raw materials to produce their own sensors and realise new DMIs, such as the ones presented in this thesis. Apart from their usefulness in acquiring expert musical gestures, paper sensors have potential applications in many other fields such as medical (e.g. as hospital bed sheet), security (e.g. sensitive floors), intelligent packaging, etc. Multi-disciplinary research is a profitable way for creating new technological outputs. Conductive paper can provide new ways of building DMIs at moderate cost and with an environmental-friendly label. Music and arts can in return offer an efficient and sensitive proof-of-work to convince industries that conductive papers have a major potential in other applications. / La fabrication de nouveaux instruments musicaux numérique (IMN) nécessite l'utilisation de plusieurs types de capteurs afin de convertir les actions humaines en signaux électroniques pour le contrôle de variables de synthèse sonore. Parmi eux les capteurs de contact (force/pression) tels que les "Force Sensing Resistors" (FSR) sont les capteurs les plus utilisés. Même si des capteurs commerciaux peuvent facilement êtres trouvés chez des détaillants en électronique, seul un nombre limité de modèles standardisés, avec des tailles, des formes et des caractéristiques électriques prédéfinies, peuvent êtres achetés. Ces limitations ont des conséquences directes sur le design de nouveaux IMN qui doivent ainsi se limiter aux offres commerciales existantes.Après divers essais utilisant des polymères souples, des textiles, et des colles, nous avons choisi de nous concentrer sur les papiers conducteurs comme matières premières pour produire des capteurs de contact sensibles et adaptables. Ces recherches de thèse se sont alors orientées vers deux directions.Premièrement, nous avons commencé par fabriquer des capteurs à partir de papiers industriels à l'origine produit pour fournir des papiers colorés, et non pour une utilisation en tant que matériau conducteur. Nous avons montré néanmoins que ce papier conducteur est un matériaux sensible et exploitable. Plusieurs prototypes de capteurs ont ainsi été fabriqués en utilisant des échantillons de quatre types de papiers industriels provenant de trois manufacturiers (ArjoWiggins (Canson), Fabriano et PASCO). Ces prototypes ont ensuite été utilisés pour le développement de divers contrôleurs gestuels. Afin de vérifier le comportement des capteurs en action, nous avons développé un banc d'essai permettant de caractériser les propriétés électriques de ces capteurs en termes de répétabilité, dérive et hystérésis, et de les comparer à ceux des capteurs commerciaux.La seconde direction consistait à rechercher comment produire de tels papiers. Des expériences de laboratoires ont permis d'évaluer réglages des processus de fabrication à privilégier et les additifs chimiques à ajouter pour optimiser la rétention et la formation d'un mélange uniforme de pigments et de fibres et la formation d'un papier rugueux, poreux et compressible mais toujours élastique. Plusieurs types de papiers ont été réalisés en faisant varier la quantité de pigments et la structure des feuilles afin d'évaluer l'influence des caractéristiques d'un papier sur sa résistance électrique. Cette recherche contribue à la promotion de solutions alternatives pour le développement de capteurs de contact. Elle montre en particulier que les papiers conducteurs et les encres conductrices sont une alternative écologique aux polymères conducteurs pour la production de capteurs de contact. Les capteurs en papiers peuvent remplacer les capteurs industriels dans de nombreuses applications. De plus, ils sont recyclables et adaptables à tous types de design. La production d'un tel papier va procurer un stock de matières premières aux chercheurs afin qu'ils produisent leurs propres capteurs et conçoivent de nouveaux IMN tels que ceux qui sont présentés dans cette thèse. De plus, mis à part leurs capacités pour le contrôle de gestes musicaux experts, les capteurs en papier ont des applications potentielles dans d'autres domaines tels qu'en médecine (monitoring de lits d'hôpitaux), sécurité et domotiques (sols et murs sensibles), emballages intelligents, etc…La recherche multidisciplinaire est un moyen susceptible de générer de nouvelles technologies. Le papier conducteur offre de nouveaux moyens de produire des IMN pour un coût modéré en plus d'un label environnemental. La musique et les arts permettent en retour d'offrir une application ludique, attirante qui à fait ses preuves afin de convaincre les industries que les papiers conducteurs ont un gros potentiel dans d'autres applications.

Reliability is a key aspect of power systems. In this research, the minimum cut set method is adapted for assessing the reliability of reconfigurable open-ring distribution networks, where we have considered both sustained and momentary interruptions. Open-ring distribution network topologies can significantly improve system reliability through load transferring, and consequently the outages suffered at load points are reduced. The impact of various transfer switch technologies (i.e. manual or automatic mechanical switches and static switches) on the reliability level is investigated. The placement of the transfer switch and its effect on the reliability level are also evaluated using two open-ring distribution network configurations. The value of reliability plays an important role in the design and planning of power systems. We present an assessment of the costs and benefits associated with different transfer switch technologies and placements. This assessment incorporates customer concerns in the analysis while responding to economic constraints of the distribution network owner. One of the economic evaluation methods based on customer perception relating the level of system reliability is the customer interruption cost (CIC). The net present value method has been utilized to combine CIC, capital investment in transfer switches, and the impact of lost utility revenues during outages to evaluate the economic viability of switch technologies and placements. It is found that the application of all transfer switches significantly reduces annual outage time and associated CIC. However, the benefit of using static switches cannot be clearly observed in annual outage reduction compared to automatic mechanical switches. Also, results show that placing a transfer switch at the end of two adjacent radial feeders can contribute to reduce the most the annual outage time suffered by customers. On the other hand, placing a transfer switch at the feeding bus contributes most to lower the outage frequency seen in the parallel feeders. / La fiabilité est un élément clé des réseaux de distribution électriques. Dans cette recherche, la méthode de l'ensemble de coupure minimal est présentée pour l'évaluation de la fiabilité des réseaux de distribution en boucle ouverte. Un réseau de distribution en boucle ouverte en soi peut améliorer considérablement la fiabilité du système en rendant possible le transfert de charge lors de pannes. Dans ces cas, les points normalement ouverts du réseau sont fermés afin d'effectuer ce transfert de charge. Dans cette thèse, on évalue l'impact sur la fiabilité de la technologie utilisée pour le commutateur de transfert (i.e. commutateur mécanique manuel ou automatique et commutateur statique). L'effet du placement du commutateur de transfert sur le niveau de fiabilité est également évalué sur deux configurations de réseau de distribution en boucle ouverte. On effectue ensuite une analyse coût-bénéfice des différentes technologies de commutateurs de transfert. Cette analyse intègre les préoccupations vis-à-vis la fiabilité des clients tout en répondant aux contraintes économiques du distributeur. Cette évaluation technico-économique est basée sur le coût d'interruption client (CIC), qui mesure l'impact économique équivalent des pannes sur les clients. Le CIC est calculé pour diverses technologies de commutateur de transfert et configurations du réseau de distribution. Simultanément, on calcule la valeur actualisée nette de ses coûts pour chaque cas ainsi que celle des coûts et bénéfices logeant du côté du distributeur, c'est-à-dire le coût d'installation des commutateurs et les réductions dans les ventes d'électricité perdues lors les pannes.On démontre ainsi que l'application de toutes les technologies de commutateurs de transfert permettent de réduire la durée des pannes et du CIC associé. Cependant, on trouve qu'il y a peu de benefices additionnels à passer d'un commutateur mécanique automatique à un commutateur statique. De plus, les résultats démontrent que le placement des commutateurs de transfert joue un rôle important dans la fiabilité des réseaux. Ainsi si on place un commutateur de transfert à la fin de deux artères parallèles, on maximise la réduction de la durée des pannes chez les clients. En contrepartie, dans le cas où le commutateur est placé en tête d'artère, ceci permet de maximiser la réduction de la fréquence de défaillance des artères parallèles.

<p> To get reliable location and time information for specific objects, the U. S. Department of Defense (DOD) developed the Global Positioning System (GPS). In spite of its enormous advantages, it encounters limitations in GPS-denied or compromised environments. To address these limitations, Inertial Motion Units (IMUs), consisting of gyroscopic devices, have been substituted for GPS in these harsh environments. However, they are also limited by bias, which increases error growth by integrating signals. This research project allows for tracking a pedestrian in GPS limited environments by adding another portable localization system to the IMU, the Ground Reaction Sensor Cluster (GRSC). The GRSC measures the precise velocity of an object by applying zero velocity updating (ZUPTing) events, and seeks to compensate for the limitations of the IMU to give the precise position of the pedestrian. This project addresses the testing environment developed to evaluate the reliability of a shoe mounted GRSC. This environment consisted of both physical measurements and analysis of the data. Through this testing we check whether this device would meet the Defense Advanced Research Projects Agency's(DARPA) requirement to resolve the zero-velocity error biases of 4000, 250 and 20 &mgr;m/s with a typical 0.3s mid-stance measurement time. To ensure an adequate amount of data, the process was automated and run for several different cycle lengths.</p>

Techniques are presented for use in FFT network analyzers to improve estimates of long lags in the impulse response. A key step involves short-term correlation of time-shifted copies of the input-output data. We introduce new Banach signal spaces for the analysis of such data. As a preliminary result, new transfer function estimation algorithms are given with bounds on the effect of undermodelling, noise and lack of empirical data in the Hinfinity norm. Comparisons with optimality are given using metric-complexity concepts. The algorithms and error calculations are implemented in MATLAB and C to realize a software network analyzer which includes error bars. As a test case, the algorithm is applied to data from the compliant hip joint of a hopping robot.

The main contribution of this work is to extend the present multi-path fading channel (MFC) models in two directions. / First, the wireless communication system is viewed as a dynamical system where the channel constitutes the main plant of such a system. The random variables characterizing the magnitude of each multi-path component are generalized to random processes with time-varying statistics. It is claimed that the dynamics of both the log-normal shadowing and short-term fading channels is captured by the use of stochastic diffusion processes which emerge from mean-reverting Ornstein-Uhlenbeck processes, implying that signal envelope variations can be modeled using stochastic differential equations. Explicit expressions for signal envelope distributions and second order statistics of the received signal are derived. / Further, the short-term dynamical model is cast into a state-space representation and a nominal model is derived, based on which robust analysis and design can be performed in order to capture additional channel uncertainties. / The dynamical channel models introduced here are in accordance of the physical principles of electromagnetic wave propagation, they are parametric and able to describe diverse propagation environments. / Second, a unified statistical analysis framework is introduced in order to compute the various statistics, second-order properties and densities of the overall received signal of the MFC's. To this end, the MFC model is formulated as a general shot-noise process brought forward by Rice. By extending and adapting the shot-noise analysis to complex signals the statistics of the overall received signal are derived, with explicit computations of the various moments and second-order statistics. The analysis includes the case of moving scatters (i.e. rain droplets) which it is shown to affect the dynamical temporal characteristics of the channel. It is also shown here that samples of the received signal approach a jointly Gaussian density, as certain parameters tend to their limits. Further, when the density of the received signal does not approach the Gaussian density, the Edgeworth series expansion is employed to compute the non-Gaussian density. This analysis brings forward the parameters responsible for Gaussianity of the received signal, parameters which designers can use in order to tailor their transmitter/receiver design. (Abstract shortened by UMI.)

The reduction of the kinematic error for precise trajectory tracking of a robotic manipulator's end effector pose is best achieved by taskspace sensory feedback. To this end, a new taskspace control scheme is introduced which has two main features: Firstly, a hierarchical control structure which feeds back both, joint and taskspace variables and thereby reduces uncertainty better than other schemes. Secondly, a nonlinear decoupling scheme which best linearizes the taskspace loop. / The manipulator's end effector pose is measured by a 3D ultrasound range finder, for which a very detailed deterministic and stochastic model is obtained from experimental data. Some of this information is then used in an extended Kalman filter to compensate for the range finder's imperfections. / Finally, the 3D ultrasound range finder, the extended Kalman filter and the new taskspace control scheme are simulated in concert in a realistic environment, to assess the control system's ability to reduce the kinematic tracking error.

Based upon the COCOLOG (Conditional Observer and Controller Logic) logic control formalism, we present a framework, called a system of Markovian fragments, which permits efficient updating of the axiom sets of a COCOLOG system. The key idea is that one should only retain those axioms and theorems concerning the most recent state estimate of the controlled machine. This information is sufficient to evaluate the truth of conditional control formulas which only depend upon the current state estimate of the controlled machine. As a result, a Markovian fragment system is demonstrated to have virtually the same control power as a full COCOLOG system. Further, it is shown that a Markovian fragmented system has a complexity of logical deduction which is time-dependent while a full COCOLOG system has a complexity which increases monotonically with time. / In order to formulate the analysis and synthesis of hierarchical systems, a hierarchical structure based upon the notion of dynamical consistency is presented. The underlying motivation for this notion is that high level dynamics should be consistent with low level dynamics. The major part of this investigation is carried out for the so-called in-block controllable partition lattices, which are the lattices of partitions for which each block element constitutes a controllable submachine. We show that most controllable finite state machines have at least one non-trivial in-block controllable partition machine. It is also shown that, subject to in-block controllability, the resulting partition machines are controllable if and only if the base machines are controllable. A consequence of this construction is that a state-to-state reachability problem can be decomposed into reachability problems at different levels of abstraction. In terms of this theory, a hierarchical control system is a set of control systems associated with a chain from the top element to the bottom element of the in-block controllable partition lattice of a controlled machine. / As an extension of regular COCOLOG, we give the HICOLOG formulation of hierarchical logic control. The basic contribution is the construction of a system wherein a high level logic controller makes its control decisions with respect to the information available to it and communicates its commands as control objectives to the associated low level logic control systems; this is continued down the chain of systems in a given hierarchical lattice. Information flows up the hierarchy in the form of abstract statements concerning the system state. / The notion of dynamical consistency is extended to hybrid systems so as to define the set of dynamically consistent hybrid partition machines associated with a continuous system ${ cal S}$. It is shown that it is possible to define an in-block controllable hybrid partition lattice for ${ cal S}$ and this permits the hybrid control of a continuous system ${ cal S}$ to be formulated within a lattice theoretic hierarchical control theory.

Using polycrystalline thin films of CuInSe$ sb2$ (thickness 1.5 $ mu$m) prepared by an electrodeposition method, photovoltaic cells of the form ZnO/CdS/CuInSe$ sb2$ have been fabricated and studied. Before the cell fabrication, properties of the electrodeposited CuInSe$ sb2$ were first studied. Conditions for the preparation of high resistivity CdS thin films (thickness 360 A) by a chemical bath deposition method were then established. This was achieved by varying the deposition temperature and amount of NH$ sb4$OH in the solution. It was observed that the quality of CdS films deposited at a temperature of 60$ sp circ$C and an NH$ sb4$OH concentration of 0.48 M was the best. Thermal stability of low resistivity, In- or Ar-doped, rf-sputtered ZnO thin films (thickness 1$ mu$m) was also studied. This was done by heat treating the ZnO films in air, O$ sb2$ and N$ sb2$ at temperature in a range from 200 to 350$ sp circ$C. It was observed that the amount of increase of resistivity at a given temperature decreased as the doping concentration was increased from 0.5 to 5.3 wt.%. For the films containing 2 wt.% In or Al, the resistivity increased as the treating temperature was increased from 200 to 350$ sp circ$C. For the samples treated at 200$ sp circ$C, the increase in resistivity was about 1 order of magnitude. / The high resistivity CdS and low resistivity ZnO thin films were then deposited on electrodeposited CuInSe$ sb2$ to fabricate cells of the form ZnO/CdS(high $ rho$)/CuInSe$ sb2$. For comparison, cells of CdS(low $ rho$)/CdS(high $ rho$)/CuInSe$ sb2$ were also fabricated by evaporation of low resistivity CdS. The CuInSe$ sb2$ films used were treated either in vacuum or Ar. For those treated in vacuum, very poor properties were observed. The properties improved after a post fabrication heat treatment in air, however, the efficiency of these cells was below 2%. The low conversion efficiency was due to the low open circuit voltage. From capacitance-voltage measurements, this was found to be due to a high acceptor concentration on the surface of the vacuum treated CuInSe$ sb2$ films (10$ sp{17}$ cm$ sp{-3}$). For the cells fabricated on the CuInSe$ sb2$ films treated in Ar, photovoltaic effects were present before the air heat treatment. An efficiency of 6.8% was obtained for one of the best cells, sample J8-4 (with low resistivity CdS window). For cells with ZnO window, a conversion efficiency of 6.3% was obtained (cell O51). For these cells, the acceptor concentration in CuInSe$ sb2$ was 10$ sp{16}$ cm$ sp{-3}$, which was one order of magnitude lower than that of CuInSe$ sb2$ films treated in vacuum. / The diffusion length of minority carriers (electrons) in the electrodeposited, p-type CuInSe$ sb2$ was first measured using the photocurrent and capacitance methods. For the vacuum treated CuInSe$ sb2$ films, the electron diffusion length was small (less than 0.1 $ mu$m). For those treated in Ar, values of the electron diffusion length were about 0.5 $ mu$m. These values are close to those reported for evaporated CuInSe$ sb2$ thin films. / Some of the fabricated cells were also studied using an electron beam induced current (EBIC) method. From the EBIC experiments, the effective diffusion lengths of electrons with values greater than 1 $ mu$m were obtained. Considering the surface recombination effect, the electron diffusion length of the electrodeposited CuInSe$ sb2$ was finally found to be 2.4 $ mu$m. This large electron diffusion length was consistent with the high short circuit current density observed in I-V measurements of the electrodeposited CuInSe$ sb2$ cells.