Thesis (PhD)--University of Stellenbosch, 2004. / ENGLISH ABSTRACT: The implementation and verification of a recently proposed Huy gens' principle in hollow lossless waveguide is described. The extended Huygens' principle is applicable to the scattering effect of arbitrary obstacles in waveguide and allows the coupling of a volume Finite Element discretization with a quasi Method of Moments surface element approach. A review of the technique is given and expressions for the two port scattering parameters are derived. The implementation is restricted to posts in rectangular waveguide, although the technique may also be applied to general scattering problems in waveguide. Finite Elements for the volume of the obstacle are selected and a basis set is proposed. The elemental finite element matrices are derived for a low and higher order basis set, while the validity of the derivations is established by considering a simple waveguide example. A coupled set of matrix equations is constructed that can be solved for the unknown surface electric and magnetic fields and the volume magnetic fields. A set of test examples is chosen to verify the implementation. The convergence behaviour is examined for two test cases. The dissertation concludes with an evaluation of the technique and recommendations for future work. / AFRIKAANSE OPSOMMING: Die implementasie en verifiëring van 'n onlangs voorgestelde Huygens beginsel in leë golfleier word beskryf. Die uitgebreide beginsel van Huygens is van toepassing op die strooiingseffek van arbitrêre voorwerpe in verlieslose golfleier en beskryf die koppeling tussen 'n volume eindige element diskretisasie met 'n oppervlakelement kwasi-moment metode. 'n Oorsig oor die tegniek word gegee en uitdrukkings vir die tweepoort strooiingsparameters word afgelei. Die implementering word beperk tot penne in reghoekige golfleier, alhoewel die tegniek op algemene golfleier strooiingsprobleme van toepassing is. Eindige elemente word gekies vir die volume van die voorwerp en 'n basisfunksie versameling afgelei van bestaande basisfunksies. Lae en hoë orde element matrikse word afgelei en die geldigheid van die afleidings word nagegaan deur 'n eenvoudige golfleier probleem op te los. 'n Gekoppelde stel matriksvergelykings word opgestel in terme van die onbekende oppervlak elektriese en magnetiese velde. 'n Aantal toetsvoorbeelde word gekies om die implementasie te verifieer. Konvergensiegedrag word ondersoek vir twee toetsgevalle. Die proefskrif sluit af met 'n evaluasie van die tegniek en aanbevelings vir verdere werk.

Thesis (MScEng)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: Markerless mobile Augmented Reality (AR) is currently being limited by the processing power available on smartphones and tablet-like devices, which also limits the research that can be done using these devices. To overcome the problem of limited processing power and the problem of being limited to the device's original hardware, the development of a laptop-based prototype platform was proposed. The use of a laptop was motivated by the processing power it o ers and because it is self-contained, but still allows for expansion. The implemented system consists of a handheld tablet-like display with touch input, attached to a backpack which houses the processing unit of the laptop. The tablet portion of the device is equipped with stereo cameras and inertial sensors to allow for optical inertial hybrid tracking. A basic optical inertial Extended Kalman Filter based tracking system was implemented as proof of the concept. A modular design was implemented that allows for components to be added to or removed from the prototype, thereby allowing for rapid prototyping with various combinations of sensors and cameras. Furthermore, multi-core and parallel processing, on both the GPU and CPU, was used to achieve fast processing with only minimal optimisations. The conceptual design, practical implementation, and testing of the prototype platform are covered in this document, along with recommendations for both continued research and the creation of similar systems. / AFRIKAANSE OPSOMMING: Die veld van merkervrye mobiele Toegevoegde Realiteit (TR) word tans beperk deur die verwerkingsvermo e van slimfone en tablette. Die beperking a ekteer ook die vermo e om navorsing te doen op hierdie toestelle. Nog 'n struikelblok met navorsing op hierdie tipe toestel is dat die hardeware nie verander of aangepas kan word nie. Om die beperkings in verwerkingsvermo e en die hardewarebeperking te oorkom, word daar voorgestel dat 'n mobiele stelsel ontwerp word wat oor genoegsame verwerkingvermo e beskik. Die voorgestelde prototipe-stelsel word gebaseer op 'n kragtige skootrekenaar, omdat dit die nodigde verwerkingvermo e besit en mobiliteit verskaf in 'n selfstandigeeenheid. Die voorgestelde stelsel is ge mplementeer deur 'n handdraagbare aanraakskerm te skep, wat verbind is met die rugsak wat die verwerkingse enheid van die skootrekenaar huisves. Die handdraagbare gedeelte van die skerm is toegerus met stereovisie-kameras en inersiesensors. Hierdie sensors en kameras is dan gebruik om 'n basiese hibriede optiese-inersie uitgebreide Kalman lter-gebaseerde posisievolgingsstelsel te skep, as bewys van die voorgestelde konsep. 'n Modul^ere ontwerp is gebruik, omdat dit toelaat dat komponente maklik vervang of bygevoeg kan word. Die modul^ere ontwerp maak dit moontlik om verskeie sensors spoedig te toets en te vergelyk. Verder is beide die SVE- en GVE-eenhede van die skootrekenaar se parallelle verwerkingsvermo e benut. Sodoende is 'n ho e verwerkingspoed bereik, deur slegs minimale optimering toe te pas. Die konsepontwerp, -implementering en -toetsing van die stelsel word in hierdie dokument beskryf. Saam met die beskrywing word daar ook aanbevelings gemaak vir opvolgende navorsing, asook vir die ontwerp van soortgelyke stelsels.

This thesis is concerned with the automatic transcription of chords from audio, with an emphasis on modern popular music. Musical context such as the key and the structural segmentation aid the interpretation of chords in human beings. In this thesis we propose computational models that integrate such musical context into the automatic chord estimation process. We present a novel dynamic Bayesian network (DBN) which integrates models of metric position, key, chord, bass note and two beat-synchronous audio features (bass and treble chroma) into a single high-level musical context model. We simultaneously infer the most probable sequence of metric positions, keys, chords and bass notes via Viterbi inference. Several experiments with real world data show that adding context parameters results in a significant increase in chord recognition accuracy and faithfulness of chord segmentation. The proposed, most complex method transcribes chords with a state-of-the-art accuracy of 73% on the song collection used for the 2009 MIREX Chord Detection tasks. This method is used as a baseline method for two further enhancements. Firstly, we aim to improve chord confusion behaviour by modifying the audio front end processing. We compare the effect of learning chord profiles as Gaussian mixtures to the effect of using chromagrams generated from an approximate pitch transcription method. We show that using chromagrams from approximate transcription results in the most substantial increase in accuracy. The best method achieves 79% accuracy and significantly outperforms the state of the art. Secondly, we propose a method by which chromagram information is shared between repeated structural segments (such as verses) in a song. This can be done fully automatically using a novel structural segmentation algorithm tailored to this task. We show that the technique leads to a significant increase in accuracy and readability. The segmentation algorithm itself also obtains state-of-the-art results. A method that combines both of the above enhancements reaches an accuracy of 81%, a statistically significant improvement over the best result (74%) in the 2009 MIREX Chord Detection tasks.

Wireless communications have been growing with an astonishing rate over the past few years and wireless terminals for future applications are required to provide diverse services. This rising demand prompts the needs for antennas able to cover multiple bandwidths or an ultrawide bandwidth for various systems. Since the release by the Federal Communications Commission (FCC) of a bandwidth of 7.5 GHz (from 3.1 GHz to 10.6 GHz) for ultra wideband (UWB) wireless communications, UWB has been rapidly evolving as a potential wireless technology and UWB antennas have consequently drawn more and more attention from both academia and industries worldwide. Unlike traditional narrow band antennas, design and analysis of UWB antennas are facing more challenges and difficulties. A competent UWB antenna should be capable of operating over an ultra wide bandwidth as assigned by the FCC. At the same time, a small and compact antenna size is highly desired, due to the integration requirement of entire UWB systems. Another key requirement of UWB antennas is the good time domain behaviour, i.e. a good impulse response with minimal distortion. This thesis focuses on UWB antenna miniaturisation and analysis. Studies have been undertaken to cover the aspects of UWB fundamentals and antenna theory. Extensive investigations are also conducted on three different types of miniaturised UWB antennas. 5 The first type of miniaturised UWB antenna studied in this thesis is the loaded orthogonal half disc monopole antenna. An inductive load is introduced to broaden the impedance bandwidth as well as the pattern bandwidth, in other words, an equivalent size reduction is realised. The second type of miniaturised UWB antenna is the printed half disc monopole antenna. By simply halving the original antenna and tuning the width of the coplanar ground plane, a significant more than 50% size reduction is achieved. The third type of miniaturised UWB antenna is the printed quasi-self-complementary antenna. By exploiting a quasi-self-complementary structure and a built-in matching section, a small and compact antenna dimension is achieved. The performances and characteristics of the three types of miniaturised UWB antennas are studied both numerically and experimentally and the design parameters for achieving optimal operation of the antennas are also analysed extensively in order to understand the antenna operations. Also, time domain performance of the Coplanar Waveguide (CPW)-fed disc monopole antenna is examined in this thesis to demonstrate the importance of time domain study on UWB antennas. Over the past few years of my PhD study, I feel honoured and lucky to work with some of the most prestigious researchers in the Department of Electronic Engineering, Queen Mary, University of London. I would like to show my most cordial gratitude to those who have been helping me during the past few years. There would be no any progress without their generous and sincere support. First of all, I would like to thank my supervisors Professor Clive Parini and Professor Xiaodong Chen, for their kind supervision and encouragement. I am impressed by their notable academic background and profound understanding of the subjects, which have proved to be immense benefits to me. It has been my great pleasure and honour to be under their supervision and work with them. Second of all, I would like to thank Mr John Dupuy for his help in the fabrication and measurement of antennas I have designed during my PhD study. Also, a special acknowledgement goes to all of the staff for all the assistance throughout my graduate program.

The work presented in this thesis is focused at associating a semantics to the content of an image, linking the content to high level semantic categories. The process can take place at two levels: either at image level, towards image categorisation, or at pixel level, in se- mantic segmentation or semantic labelling. To this end, an analysis framework is proposed, and the different steps of part (or patch) extraction, description and probabilistic modelling are detailed. Parts of different nature are used, and one of the contributions is a method to complement information associated to them. Context for parts has to be considered at different scales. Short range pixel dependences are accounted by associating pixels to larger patches. A Conditional Random Field, that is, a probabilistic discriminative graphical model, is used to model medium range dependences between neighbouring patches. Another contribution is an efficient method to consider rich neighbourhoods without having loops in the inference graph. To this end, weak neighbours are introduced, that is, neighbours whose label probability distribution is pre-estimated rather than mutable during the inference. Longer range dependences, that tend to make the inference problem intractable, are addressed as well. A novel descriptor based on local histograms of visual words has been proposed, meant to both complement the feature descriptor of the patches and augment the context awareness in the patch labelling process. Finally, an alternative approach to consider multiple scales in a hierarchical framework based on image pyramids is proposed. An image pyramid is a compositional representation of the image based on hierarchical clustering. All the presented contributions are extensively detailed throughout the thesis, and experimental results performed on publicly available datasets are reported to assess their validity. A critical comparison with the state of the art in this research area is also presented, and the advantage in adopting the proposed improvements are clearly highlighted.

In today’s working world the elderly who are dependent can sometimes be neglected by society. Statistically, after toddlers it is the elderly who are observed to have higher accident rates while performing everyday activities. Alzheimer’s disease is one of the major impairments that elderly people suffer from, and leads to the elderly person not being able to live an independent life due to forgetfulness. One way to support elderly people who aspire to live an independent life and remain safe in their home is to find out what activities the elderly person is carrying out at a given time and provide appropriate assistance or institute safeguards. The aim of this research is to create improved methods to identify tasks related to activities of daily life and determine a person’s current intentions and so reason about that person’s future intentions. A novel hierarchal framework has been developed, which recognises sensor events and maps them to significant activities and intentions. As privacy is becoming a growing concern, the monitoring of an individual’s behaviour can be seen as intrusive. Hence, the monitoring is based around using simple non intrusive sensors and tags on everyday objects that are used to perform daily activities around the home. Specifically there is no use of any cameras or visual surveillance equipment, though the techniques developed are still relevant in such a situation. Models for task recognition and plan recognition have been developed and tested on scenarios where the plans can be interwoven. Potential targets are people in the first stages of Alzheimer’s disease and in the structuring of the library of kernel plan sequences, typical routines used to sustain meaningful activity have been used. Evaluations have been carried out using volunteers conducting activities of daily life in an experimental home environment. The results generated from the sensors have been interpreted and analysis of developed algorithms has been made. The outcomes and findings of these experiments demonstrate that the developed hierarchal framework is capable of carrying activity recognition as well as being able to carry out intention analysis, e.g. predicting what activity they are most likely to carry out next.

This thesis investigates the use of latent semantic models for annotation and retrieval from collections of musical audio tracks. In particular latent semantic analysis (LSA) and aspect models (or probabilistic latent semantic analysis, pLSA) are used to index words in descriptions of music drawn from hundreds of thousands of social tags. A new discrete audio feature representation is introduced to encode musical characteristics of automatically-identified regions of interest within each track, using a vocabulary of audio muswords. Finally a joint aspect model is developed that can learn from both tagged and untagged tracks by indexing both conventional words and muswords. This model is used as the basis of a music search system that supports query by example and by keyword, and of a simple probabilistic machine annotation system. The models are evaluated by their performance in a variety of realistic retrieval and annotation tasks, motivated by applications including playlist generation, internet radio streaming, music recommendation and catalogue search.

The overlapping of the different wireless network technologies creates heterogeneous communication environments. Future mobile communication system considers the technological and operational services of heterogeneous communication environments. Based on its packet switched core, the access to future mobile communication system will not be restricted to the mobile cellular networks but may be via other wireless or even wired technologies. Such universal access can enable service convergence, joint resource management, and adaptive quality of service. However, in order to realise the universal access, there are still many pending challenges to solve. One of them is the selection of the most appropriate radio access network. Previous work on the network selection has concentrated on serving the requesting user, but the existing users and the consumption of the network resources were not the main focus. Such network selection decision might only be able to benefit a limited number of users while the satisfaction levels of some users are compromised, and the network resources might be consumed in an ineffective way. Solutions are needed to handle the radio access network selection in a manner that both of the satisfaction levels of all users and the network resource consumption are considered. This thesis proposes an intelligent radio access network selection and optimisation system. The work in this thesis includes the proposal of an architecture for the radio access network selection and optimisation system and the creation of novel adaptive algorithms that are employed by the network selection system. The proposed algorithms solve the limitations of previous work and adaptively optimise network resource consumption and implement different policies to cope with different scenarios, network conditions, and aims of operators. Furthermore, this thesis also presents novel network resource availability evaluation models. The proposed models study the physical principles of the considered radio access network and avoid employing assumptions which are too stringent abstractions of real network scenarios. They enable the implementation of call level simulations for the comparison and evaluation of the performance of the network selection and optimisation algorithms.

My thesis takes steps towards understanding the role technology can play in supporting multisession creative collaborative work. This is achieved by exploring the relationship between the outcomes of a session of work and the resources available within the environment where work takes place. My domain of study is Joint Music Composition, which is a form of collaborative work that requires participants to generate, share, develop and remember information about a musical composition across a number of sessions. Although musical instrument and recording technology have advanced, there appears to be little understanding of how technology can be used to support collaboration in Joint Music Composition. To investigate this, I used the Distributed Cognition framework (Hutchins, 1995a), which has traditionally been employed to study work activities within socio-technological settings, to better understand how to support collaboration and coordination within my domain of study. The findings of my thesis are based on studies conducted in real life settings (i.e., field) and in environments that I helped to organise (i.e., laboratory). Research from the field describes how groups naturally organise their session, their physical setting, and their communication. It also helps to highlight a number of issues relating to the cognitive burden associated with compositions when they are in development. The first laboratory study illustrates the distributed nature of problem solving in Joint Music Composition by giving examples of different ways knowledge is shared within the group and across sessions. The second laboratory study describes how a shared work space appears to change the way knowledge is represented and distributed within two different rehearsal set-ups. Overall, the main insights that are applicable to informing design relate to the way practitioners of Joint Music Composition manage the distributed nature of problem solving using transient representations across multiple sessions of work.

Self-repairs, where speakers repeat themselves, reformulate or restart what they are saying, are pervasive in human dialogue. These phenomena provide a window into real-time human language processing. For explanatory adequacy, a model of dialogue must include mechanisms that account for them. Artificial dialogue agents also need this capability for more natural interaction with human users. This thesis investigates the structure of self-repair and its function in the incremental construction of meaning in interaction. A corpus study shows how the range of self-repairs seen in dialogue cannot be accounted for by looking at surface form alone. More particularly it analyses a string-alignment approach and shows how it is insufficient, provides requirements for a suitable model of incremental context and an ontology of self-repair function. An information-theoretic model is developed which addresses these issues along with a system that automatically detects self-repairs and edit terms on transcripts incrementally with minimal latency, achieving state-of-the-art results. Additionally it is shown to have practical use in the psychiatric domain. The thesis goes on to present a dialogue model to interpret and generate repaired utterances incrementally. When processing repaired rather than fluent utterances, it achieves the same degree of incremental interpretation and incremental representation. Practical implementation methods are presented for an existing dialogue system. Finally, a more pragmatically oriented approach is presented to model self-repairs in a psycholinguistically plausible way. This is achieved through extending the dialogue model to include a probabilistic semantic framework to perform incremental inference in a reference resolution domain. The thesis concludes that at least as fine-grained a model of context as word-by-word is required for realistic models of self-repair, and context must include linguistic action sequences and information update effects. The way dialogue participants process self-repairs to make inferences in real time, rather than filter out their disfluency effects, has been modelled formally and in practical systems.