Service orientation is a design paradigm consisting of a set of principles governed by a service-oriented architecture (SOA) to support the creation of software systems as a composition of interoperable services. The ability to effectively compose services is not a trivial task due to the dynamic nature of the execution environment of service compositions. In this context, dynamic service selection and composition is a critical requirement and one of the major research challenges for service-based systems. This research investigates the identification, detection and prediction of the need for adaptation as well as ways to autonomously reconfigure the service composition during its execution time in order to improve service reliability and conformance with systems requirements and policies. We propose a framework for proactive adaptation of service compositions that extends current approaches for dynamic service composition by proactively and individually identifying the need for adaptation for each parallel running instance of service composition while avoiding unnecessary changes and distributing load request among different service operations when necessary. Our framework has been tested and validated using different prototypes implemented in both simulated and real environments. The results were favourable with the research objectives and indicates a major gain in the use of the proposed proactive techniques in the execution and adaptation of web service compositions.

Among the several numerical methods which are available for solving complex problems in many areas of engineering and science such as structural analysis, fluid flow and bio-mechanics, the Finite Element Method (FEM) is the most prominent. In the context of these methods, high quality meshes can be crucial to obtaining accurate results. Finite Element meshes are composed of elements and the quality of an element can be described as a numerical measure which estimates the effect that the size/shape of an element will have on the accuracy of an analysis. In this thesis, the strong link between mesh geometry and the accuracy and efficiency of a simulation is explored and it is shown that poor quality elements cause both interpolation errors and poor conditioning of the global stiffness matrix. Numerical optimisation is the process of maximising or minimising an objective func- tion, subject to constraints on the solution. When this is applied to a finite element mesh it is referred to as mesh optimisation, where the quality of the mesh is the objec- tive function and the constraints include, for example, the domain geometry, maximum element size, etc. A mesh optimisation strategy is developed with a particular focus on optimising the quality of the worst elements in a mesh. Using both two and three dimensional examples, the most efficient and effective combination of element quality measure and objective function is found. Many of the problems under consideration are characterised by very complex geometries. The nodes lying on the surfaces of such meshes are typically treated as unmovable by most mesh optimisation software. Techniques exist for moving such nodes as part of the mesh optimisation process, however, the resulting mesh geometry and area/volume is often not conserved. This means that the optimised mesh is no longer an accurate discretisation of the original domain. Therefore, a method is developed and demonstrated which optimises the positions of surface nodes while respecting the geometry and area/volume of a domain. At the heart of many of the problems being considered is the Arbitrary Lagrangian Eulerian (ALE) formulation where the need to ensure mesh quality in an evolving mesh is very important. In such a formulation, a method of determining the updated nodal positions is required. Such a method is developed using mesh optimisation techniques as part of the FE solution process and this is demonstrated using a two-dimensional, axisymmetric simulation of a micro-fluid droplet subject to external excitation. While better quality meshes were observed using this method, the time step collapsed resulting in simulations requiring significantly more time to complete. The extension of this method to incorporate adaptive re-meshing is also discussed.

In this thesis the theory of both linear differential equations with periodic coefficients and linear differential equations with random coefficients is applied to investigate the stability and accuracy of parameter adaptation of sinusoidal perturbation and model reference adaptive control systems. Throughout dimensional analysis applied so that all the results are presented in a non-dimensional form. The first part of the thesis t s devoted to investigating the stability of such differential equations. In chapter 1 a system of linear homogeneous differential equations with periodic coefficients is considered and a numerical procedure, based on Floquet theory and well suited for use on a digital computer, is presented for obtaining necessary and sufficient conditions for asymptotic stability of the null solution. Also considered in this chapter is the so called infinite determinant method of obtaining the stability boundaries for a restricted class of linear differential equations with periodic coefficients. Chapter 2 is devoted to reviewing the current state of the stability theory of linear differential equations with random coefficients. In chapter 3 a theoretical analysis of the stability and accuracy of parameter adaptation of a single input, sinusoidal perturbation, extremum control system with output lag is considered. Using the principle of harmonic balance it is shown that various stable harmonic and sub-harmonic steady state solutions are possible in certain regions of the parameter space. By examining the domains of attraction, corresponding to the stable solutions, regions in three dimensional space are obtained within .which initial conditions will lead to a given steady state stable oscillation. It is also shown that the subharmonic steady state solutions do not correspond to the optimum solution, so that, for certain initial conditions and parameter values, it is possible for the system to reach a steady state solution which is not the optimum solution. All the theoretical results are verified by direct analogue computer simulation of the system. The remainder of the thesis is devoted to investigating the stability and accuracy of parameter adapt.at ion of model reference adaptive control systems. In order to develop a mathematical analysis, and to illustrate the difficulties involved, a stability analysis of a first order M.I.T. type system with controllable gain, when the input varies with time in both a periodic and random manner, is first carried out. Also considered are the effects of (a) random disturbances at the system output (b) and (b) periodic and random variations, with time, of the controlled process environmental parameters, on the stability of the system and the accuracy of its parameter adaptation. When the input varies sinusoidally with time stability boundaries are obtained using both a numerical implementation of Floquet theory and the infinite determinant method; the relative merits of the two methods is discussed. The theoretical results are compared with stability boundaries obtained by analogue computer simulation of the system. It is shown that the stability boundaries are complex in nature and that some knowledge of such boundaries is desirable before embarking on an analogue computer investigation of the system. When the input varies randomly with time the stability problem reduces to one of investigating the stability of a system of linear differential equations with random coefficients. Both the theory of Markov processes, involving use of the Fokker-Planck equation, and the second method of Liapunov are used to investigate the problem; limitations and difficulty of applications of the theory is discussed. The theoretical results obtained are compared with those obtained by digital simulation of the system. If the controlled process environmental parameter is allowed to become time varying then it is shown that this effects both the stability of the system and the accuracy of its parameter adaptation. Theoretical results are obtained for the cases of the parameter varying both sinusoidally and randomly with time; some of the results are compared with those obtained by digital simulation of the system. It is also shown that noise disturbance at the system output has no effect on the system stability but does effect the accuracy of the parameter adaptation. The doubts concerning the stability and the difficulty of analysis of the M.I.T. , type system have led.:researchers to think about redesigning the model reference system from the point of view of stability. In particular we have the Liapunov synthesis method where the resulting system is guaranteed stable for all possible inputs. However, in designing such systems the controlled process environmental parameters are assumed constant and, by considering the Liapunov redesign scheme of the first order M.I.T. system previously discussed, it will be shown that the effect of making such parameters time varying is to introduce a stability problem. In chapter 6 the methods developed for analysing the first order system are extended to examine the stability of a higher order M.I.T. type system. The system considered has a third order process and a second order model and a stability analysis is presented for both sinusoidal and random input. Steady state values of the adapting parameters are first obtained and the linearized variational equations , for small disturbances about such steady states, examined to answer the stability problem. Theoretical results are compared with those obtained by direct analogue computer simulation of the system. The effect, on the mathematical analysis , of replacing the system multipliers by diode switching units is also considered in this chapter. The chapter concludes by presenting a method of obtaining a Liapunov redesign scheme for the system under discussion.

This thesis describes an investigation of polarisation conversion effects in gallium arsenide optical waveguides. The research was carried out with the aims of predicting, preventing and harnessing such effects. Experimental results are presented to demonstrate changes in the polarisation state of light propagating in passive deep-etched waveguides. The results are described by established modelling techniques. The effect due to process-dependent features of waveguide cross-section geometry, in particular asymmetry resulting form non-vertical etching, is investigated. The polarisation angles of hybrid waveguide modes are measured, and a novel technique is presented for the measurement of the differences been the effective indices of orthogonally polarised modes. The measurements obtained are used to analyse the evolution of elliptical polarisation states during propagation, and to provide an account of the physical origin of the polarisation conversion. Details of the nature of the optical modes predicted by rigorous numerical method simulations are demonstrated by the experimental results, while quantitative agreement between the simulated and measured data is shown. A simplified account of the behaviour is provided using a coupled-mode formulation. The influence of the linear electrooptic effect in modifying the polarisation conversion behaviour is explored experimentally, and is described using established theory. Waveguide designs are obtained which prevent unintended polarisation conversion in the presence of identified causes, while maintaining the main waveguide parameters of material composition, optical mode size and shape, electrooptic performance, and fabrication process. The polarisation behaviour in waveguides fabricated to these designs is evaluated, and the expected performance benefits are confirmed. A novel waveguide device which provides electrooptic control and switching of the optical polarisation state is presented. The device is capable of converting any input polarisation state into an arbitrary output state using the linear electrooptic effect. A working design is obtained and the fabrication of devices is described. Experimental results are presented which demonstrate the concept. Further developments of the polarisation controller device are proposed, including the realisation of the potential for switching speeds at frequencies of tens of GHz.

Generating high-fidelity images in real-time at reasonable frame rates, still remains one of the main challenges in computer graphics. Furthermore, visuals remain only one of the multiple sensory cues that are required to be delivered simultaneously in a multi-sensory virtual environment. The most frequently used sense, besides vision, in virtual environments and entertainment, is audio. While the rendering community focuses on solving the rendering equation more quickly using various algorithmic and hardware improvements, the exploitation of human limitations to assist in this process remain largely unexplored. Many findings in the research literature prove the existence of physical and psychological limitations of humans, including attentional, perceptual and limitations of the Human Sensory System (HSS). Knowledge of the Human Visual System (HVS) may be exploited in computer graphics to significantly reduce rendering times without the viewer being aware of any resultant image quality difference. Furthermore, cross-modal effects, that is the influence of one sensory input on another, for example sound and visuals, have also recently been shown to have a substantial impact on viewer perception of virtual environment. In this thesis, auditory-visual cross-modal interaction research findings have been investigated and adapted to graphics rendering purposes. The results from five psychophysical experiments, involving 233 participants, showed that, even in the realm of computer graphics, there is a strong relationship between vision and audition in both spatial and temporal domains. The first experiment, investigating the auditory-visual cross-modal interaction within spatial domain, showed that unrelated sound effects reduce perceived rendering quality threshold. In the following experiments, the effect of audio on temporal visual perception was investigated. The results obtained indicate that audio with certain beat rates can be used in order to reduce the amount of rendering required to achieve a perceptual high quality. Furthermore, introducing the sound effect of footsteps to walking animations increased the visual smoothness perception. These results suggest that for certain conditions the number of frames that need to be rendered each second can be reduced, saving valuable computation time, without the viewer being aware of this reduction. This is another step towards a comprehensive understanding of auditory-visual cross-modal interaction and its use in high-fidelity interactive multi-sensory virtual environments.

The primary outcome of this research project is the development of a methodology enabling fast automated early-stage power and energy analysis of configurable processors for system-on-chip platforms. Such capability is essential to the process of selecting energy efficient processors during design-space exploration, when potential savings are highest. This has been achieved by developing dynamic and static energy consumption models for the constituent blocks within the processors. Several optimisations have been identified, specifically targeting the most significant blocks in terms of energy consumption. Instruction encoding mechanism reduces both the energy and area requirements of the instruction cache; modifications to the multiplier unit reduce energy consumption during inactive cycles. Both techniques are demonstrated to offer substantial energy savings. The aforementioned techniques have undergone detailed evaluation and, based on the positive outcomes obtained, have been incorporated into Cascade, a system-on-chip coprocessor synthesis tool developed by Critical Blue, to provide automated analysis and optimisation of processor energy requirements. This thesis details the process of identifying and examining each method, along with the results obtained. Finally, a case study demonstrates the benefits of the developed functionality, from the perspective of someone using Cascade to automate the creation of an energy-efficient configurable processor for system-on-chip platforms.

This thesis details the analysis and design of ultra-low power radio transceivers operating at microwave frequencies. Hybrid prototypes and Monolithic Microwave Integrated Circuits (MMICs) which achieve power consumptions of less than 1 mW and theoretical operating ranges of over 10 m are described. The motivation behind the design of circuits exhibiting ultra low power consumption and, in the case of the MMICs, small size is the emerging technology of Wireless Sensor Networks (WSN). WSNs consist of spatially distributed ‘nodes’ or ‘specks’ each with their own renewable energy source, one or more sensors, limited memory, processing capability and radio or optical link. The idea is that specks within a ‘speckzone’ cooperate and share computational resources to perform complex tasks such as monitoring fire hazards, radiation levels or for motion tracking. The radio section must be ultra low power e.g. sub 1 mW in order not to drain the limited battery capacity. The radio must also be small in size e.g. less than 5 x 5 mm so that the overall speck size is small. Also, the radio must still be able to operate over a range of at least a metre so as to allow radio contact between, for example, rooms or relatively distant specks. The unsuitability of conventional homodyne topologies to WSNs is discussed and more efficient methods of modulation (On-Off Keying) and demodulation (non-coherent) are presented. Furthermore, it is shown how Super-Regenerative Receivers (SRR) can be used to achieve relatively large output voltages for small input powers. This is important because baseband Op-Amps connected at the RF receiver output generally cannot amplify small signals at the input without the output being saturated in noise (10mV is the smallest measured input for 741 Op-Amp). Instrumentation amplifiers are used in this work as they can amplify signals below 1mV. The thesis details the analysis and design of basic RF building blocks: amplifiers, oscillators, switches and detectors. It also details how the circuits can be put together to make transceivers as well as describing various strategies to lower power consumption. In addition, novel techniques in both circuit and system design are presented which allow the power consumption of the radio to be reduced by as much as 97% whilst still retaining adequate performance. These techniques are based on duty cycling the transmitter and receiver and are possible because of the discontinuous nature of the On-Off Keying signal. In order to ease the sensitivity requirements of the baseband receive amplifier a design methodology for large output voltage receivers is presented. The designed receiver is measured to give a 5 mV output for an input power of -90 dBm and yet consumes less than 0.7 mW. There is also an appendix on the non linear modelling of the Glasgow University 50nm InP meta-morphic High Electron Mobility Transistor (50nm mHEMT) and one on the non linear modelling of a commercial Step Recovery diode (SRD). Models for the 50 nm mHEMT and the SRD are useful in the analysis, simulation and design of oscillators and pulse generators respectively.

Despite the success of connectionist systems in prediction and classification problems, critics argue that the lack of symbol processing and explanation capability makes them less competitive than symbolic systems. Rule extraction from neural networks makes the interpretation of the behaviour of connectionist networks possible by relating sub-symbolic and symbolic processing. However, most rule extraction methods focus only on specific neural network architectures and present limited generalization performance. Support Vector Machine is an unsupervised learning method that has been recently applied successfully in many areas, and offers excellent generalization ability in comparison with other neural network, statistical, or symbolic machine learning models. In this thesis, an algorithm called Geometric and Oracle-Based Support Vector Machines Rule Extraction (GOSE) has been proposed to overcome the limitations of other rule-extraction methods by extracting comprehensible models from Support Vector Machines (SVM). This algorithm views the extraction as a geometric task. Given a trained SVM network, GOSE queries the synthetic instances and draws conjunction rules by approximating the optimization problem. The extracted rule set also represents the approximation of the SVM classification boundary. Unlike previous works in SVM rule-extraction, GOSE is broadly applicable to different networks and problems because it need not rely on training examples and network architectures. Theoretical proof guarantees that GOSE is capable of approximating the behavior of SVM networks. Empirical experiments are conducted on different SVM networks from binary classification networks to multi-class networks in various classification domains. The result of experiments demonstrates that GOSE can extract comprehensible rules with high levels of accuracy and fidelity for their corresponding networks. GOSE also exhibits superior consistency. After analyzing and applying several optimizing measures, the complexity of GOSE was improved. In brief, GOSE provides a novel way to explain how an SVM network functions.

This thesis proposes a new, highly generalised and fundamental, information-modelling framework called the TRM (Ternary Relations Model). The TRM was designed to be a model for converging a number of differing paradigms of information management, some of which are quite isolated. These include areas such as: hypertext navigation; relational databases; semi-structured databases; the Semantic Web; ZigZag and workflow modelling. While many related works model linking by the connection of two ends, the TRM adds a third element to this, thereby enriching the links with associative meanings. The TRM is a formal description of a technique that establishes bi-directional and dynamic node-link structures in which each link is an ordered triple of three other nodes. The key features that makes the TRM distinct from other triple-based models (such as RDF) is the integration of bi-directionality, functional links and simplicity in the definition and elements hierarchy. There are two useful applications of the TRM. Firstly it may be used as a tool for the analysis of information models, to elucidate connections and parallels. Secondly, it may be used as a “construction kit” to build new paradigms and/or applications in information management. The TRM may be used to provide a substrate for building diverse systems, such as adaptive hypertext, schemaless database, query languages, hyperlink models and workflow management systems. It is, however, highly generalised and is by no means limited to these purposes.

The ever increasing demand for higher power devices at higher frequencies has prompted much research recently into the aluminium nitride/gallium nitride high electron mobility transistors (AlN/GaN HEMTs) in response to theoretical predictions of higher performance devices. Despite having superior material properties such as higher two-dimensional electron gas (2DEG) densities and larger breakdown field as compared to the conventional aluminium gallium nitride (AlGaN)/GaN HEMTs, the AlN/GaN devices suffer from surface sensitivity, high leakage currents and high Ohmic contact resistances. Having very thin AlN barrier layer of ∼ 3 nm makes the epilayers very sensitive to liquids coming in contact with the surface. Exposure to any chemical solutions during device processing degrades the surface properties, resulting in poor device performance. To overcome the problems, a protective layer is employed during fabrication of AlN/GaN-based devices. However, in the presence of the protective/passivation layers, formation of low Ohmic resistance source and drain contact becomes even more difficult. In this work, thermally grown aluminium oxide (Al2O3) was used as a gate di- electric and surface passivation for AlN/GaN metal-oxide-semiconductor (MOS)-HEMTs. Most importantly, the Al2O3 acts as a protection layer during device processing. The developed technique allows for a simple and effective wet etching optimisation using 16H3PO4:HNO3:2H2O solution to remove Al from the Ohmic contact regions prior to the formation of Al2O3 and Ohmic metallisation. Low Ohmic contact resistance (0.76Ω.mm) as well as low sheet resistance (318Ω/square) were obtained after optimisation. Significant reduction in the gate leakage currents was observed when employing an additional layer of thermally grown Al2O3 on the mesa sidewalls, particularly in the region where the gate metallisation overlaps with the exposed channel edge. A high peak current ∼1.5 A/mm at VGS=+3 V and a current-gain cutoff frequency, fT , and maximum oscillation frequency, fMAX , of 50 GHz and 40 GHz, respectively, were obtained for a device with 0.2 μm gate length and 100 μm gate width. The measured breakdown voltage, VBR, of a two-finger MOS-HEMT with 0.5μm gate length and 100 μm gate width was 58 V. Additionally, an approach based on an accurate estimate of all the small-signal equivalent circuit elements followed by optimisation of these to get the actual element values was also developed for AlN/GaN MOS-HEMTs. The extracted element values provide feedback for further device process optimisation. The achieved results indicate the suitability of thermally grown Al2O3 for AlN/GaN-based MOS-HEMT technology for future high frequency power applications.