In spatial diversity, independently fading signals can be received by antenna elements separated by a small distance. Combining of the received signals can lead to an increase in the capacity and reliability of communications over the wireless channel. Spatial diversity can be achieved in a cellular network by sharing information between Mobile Terminals (MTs) where one MT acts as a relay supporting the data transmitted from the source in the first instance by forwarding information to the destination in the second instance. Due to the cooperation between MTs this is termed cooperative diversity. Initially this thesis considers the effect of cooperative diversity in an environment where MTs are equipped with two antenna elements, effectively combining the fast-fading combating techniques of cooperative diversity and multiple-antennas. Cooperative diversity transmission can be performed by a number of different protocols, which are termed Protocols I-V. Imposing system constraints on the network in order to make a fair comparison between the protocols, in-particular the traditional single-hop channel, allows the benefits of cooperative diversity to fully be established. An information theory approach is developed using multiple antenna techniques to provide a framework for cooperative diversity. It is shown that cooperative diversity can offer significant improvements in terms of probability of outage and capacity. In-particular, an adaptive cooperative diversity protocol is developed to select the optimal protocol dependent on channel conditions which shows a 4.25dB increase in capacity, at the 5% outage level, for a single user.

This thesis investigates the management of radio resources in the physical layer (PHY and MAC layers of multi-carrier CDMA (MC-CDMA)) systems and how the estimation of metric sin the various layers may be used in performing a cross layer management of resources to provide increased QoS whilst making optimal usage of the radio resource. At the PHY layer, the grouping and subcarrier allocation problem for a grouped MC-CDMA system is formulated as an integer linear programming problem. Two algorithms are proposed to solve this problem, namely a Branch and Bound based algorithm and a mixed probabilistic-greedy Local Search algorithm. The Local Search algorithm is found to offer increased QoS (in terms of BER) for more users at a lower complexity than any of the other algorithms. At the MAC layer, a new multi-rate model - multi-group MC-CDMA (MG-MC-CDMA) - is introduced and the performance of power-control and multi-group allocation algorithms in the MG-MC-CDMA system examined. A weighted fair queuing scheduler that takes advantage of the particular features of the MG-MC-CDMA system is proposed. A capacity model, incorporating an interference analysis and that takes into account the nature of the traffic types carried in the system, is outlined. In addition to MAC layer metrics characterising the traffic in the system, the capacity model has, as some of its required metrics, PHY layer parameters such as the ratio of inter-cell interference to total received power and information of whether or not a mobile is in a cell’s edge region. New techniques are proposed to estimate these metrics. The final contribution of the thesis is the use of the proposed dynamic capacity estimation formwork to develop new radio resource management algorithms that work across the PHY and MAC layers to deliver enhanced QoS.

Automated speech recognition technology is increasingly used in the mass-market domain of self-service telephone applications. A dilemma facing designers of menu-based applications is where to place new or less frequently requested service options within the call-flow and how to incorporate these with the existing dialogue interface design. The research detailed in this thesis proposes the use of system-initiated digressions as an alternative strategy to that of explicitly adding all options to the main menu listing of a speech-driven automated service. The purpose of these digressions was to deliver information about the availability of a new product or service option that could be triggered by using the relevant spoken keyword at the main menu. The keyword itself, however, was not explicitly mentioned (i.e. remained ‘hidden’) as an option in the existing main menu listing; therefore, callers had to infer that the option was available and then initiate the request themselves rather than passively select it from the menu listing. The dialogue engineering investigation presented here centres on three main themes: the location of the digression in the dialogue, the turn-taking strategy employed and the type of register (wording) adopted.  Contrasting system-initiated digressions were introduced into the dialogue of an existing real-world automated telephone banking service. In a series of four progressive empirical experiments, participants were invited to use the automated service to carry out banking tasks and were subjected to digressive dialogues in the form of banking product offers. The purpose of the experiments was to evaluate the impact of deploying system-initiated digressions on user attitudes toward the usability of the core service. Furthermore, detailed analyses were also performed to determine the effect that varying location, strategy and register in the digressions may have on participant attitudes. The conclusions drawn from this research support the introduction of system-initiated digressions in automated services. However, issues regarding users’ mental models of menu-driven automated services and their expectations of the computer’s social behaviour were identified in the research: participants had difficulties with correctly interpreting the concept of ‘hidden’ menu options and were sensitive to the more forceful registers adopted in the system-initiated digressions. The findings from the four experiments are presented.

Surname capture via automatic speech recognition has many potential applications, including automated directory assistance and travel reservation services. It is, however, a difficult challenge, firstly because of the large set of names involved in many of the potential applications and secondly because of the lack of standardised pronunciations for many of these names. Previous work has explored a variety of approaches to proper name recognition but has focused on recognition accuracy alone, with few attempts to assess user reaction to the various strategies investigated. The work presented in this thesis addresses this by examining the problem of automated surname capture from a user perspective. In doing so it seeks to advance knowledge in the field of spoken language dialogue services through examination of a particular problem that nonetheless has wide applicability. Data from a series of three controlled experiments are presented in which the usability of three different strategies for surname capture is empirically evaluated in both automated telephony and multimodal contexts. The focus of the multimodal work is on spoken language dialogue services in which graphical output is employed in the form of an embodied conversational agent. The underlying thesis of the work is that, through careful dialogue engineering, automated surname capture using current speech recognition technology (and by extension other proper name tasks) can be made highly usable. The evaluation methodology employed throughout provides both quantitative and qualitative data on user attitudes, together with objective measures of performance. It thus provides a comprehensive measure of usability that is missing not only from work on proper name recognition, but from the wider field of spoken language dialogue services as a whole. In particular, systematic usability evaluations of embodied conversational agent applications in which realistic speech recognition is employed are rare, hence the need for the evaluations presented here. The results show that in order to achieve a high level of usability the use of spelling information is vital in strategies for automated surname capture. This is true in both automated telephony contexts and multimodal interfaces of the type examined in the research. Moreover, where text output is available this can also improve the usability of the process.

No description available.

The frequency-domain space alternating generalised expectation-maximisation (FD-SAGE) algorithm is proposed and used in conjunction with the serial interference cancellation (SIC) technique for joint detection and estimation of multipath channel parameters. The SIC technique demonstrates more stable performance than the parallel interference cancellation (PIC) technique used in the time-domain SAGE algorithm especially in a multipath rich environment. The performance of the FD-SAGE algorithm is demonstrated by using real indoor channel measurement data and its functionality is verified through comparison with unitary estimation of signal parameter via rotational invariance technique (ESPRIT) algorithm. The first channel model is derived from data collected during a static measurement campaign. This model incorporates both the clustering of MPCs and the correlation between the spatial and temporal domains. The clustering effect relies on two classes of parameters (<i>intercluster </i>and <i>intracluster parameters)</i> and two classes of power density spectra (PDS) (<i>intercluster </i>and <i>intracluster</i> PDS) which characterise the cluster and MPC, respectively. All parameters are described by empirical probability density (pdfs) derived from the measured data and the correlation properties are incorporated in two joint pdfs for cluster and MPC positions. Data analysis shows that the intercluster and intracluster PDS exhibit exponential and Laplacian functions in the delay and angular domains, respectively. The second channel model is derived based on data collected during a dynamic measurement campaign. This model incorporates both the spatial-temporal properties as well as the dynamic evolution of paths due to motion of the MT. An <i>M</i>-step, 4-state Markov channel model (MCM) is proposed in order to account for the correlation between the number of births and deaths and multiple births and deaths that can occur at any time instant. The power and spatio-temporal variation of paths within their lifespan are modelled by a low-pass-filter and a Gaussian distributed spatio-temporal vector, respectively. Due to the distinction in the birth-death statistics and the spatio-temporal dispersion and correlation properties for line-of-sight (LOS) and non-LOS (NLOS) scenarios, the model can be generalised, and parameterised by two sets of Markov parameters for these two scenarios.

Until recently, videophony and videoconferencing have not been commercially successful due to the lack of standards in the image compression algorithms employed. In 1990, the CCITT released a draft of such an algorithm for videophony, known as recommendation H.261. This algorithm took several years to develop, and already several superior algorithms are known. Tools are required to aid the development stage of algorithm design. The commercial success of videophones relies upon standard algorithms and the availability of a ubiquitous digital switched network. The latter has been provided in the form of the CCITT ISDN standard. This project investigates parallel techniques for the real-time simulation of videophone image compression algorithms. The simulations are flexible in the sense that the algorithm being simulated can be changed with little effort. Also, they provide a high degree of algorithm visualisation - <i>i.e.</i> video output from <i>any</i> of the algorithm can be easily viewed. In this thesis a survey of past and present videophone image compression algorithms is presented and the H.261 algorithm is described in detail. The complexity of current algorithms dictates that they cannot be simulated in real-time using conventional sequential computers. Affordable supercomputer power is now available in the form of MIMD distributed memory parallel computers. Such machines can be used to develop complex image processing algorithms in <i>software</i>. The use of parallel computers is in its infancy. It is not clear how the processor networks should be interconnected, what languages should be used, how processes should be mapped to processors and how communications should be managed. The first simulations presented are written in occam with a topology independent routing harness, <i>Tiny</i>, and run on a transputer-based Meiko Computing Surface. 'Compact graph' process network topologies are employed to reduce communications overheads. These simulations are compared with related work. In so doing, various topologies are analysed using meaasures from graph theory.

The topic of this thesis is the use of multi-carrier modulation with code division multiple access (CDMA). The motivation of this work is to establish if the combination of multi-carrier modulation with CDMA has a performance advantage over a conventional direct sequence CDMA (DS-CDMA) communication system. In this thesis three types of multi-carrier CDMA are identified and the main work is concentrated on one particular combination, which is referred to as one chip per carrier multi-carrier CDMA system. This system itself, however can be split into different variations and an examination of two of these is made. The first of these one chip per carrier multi-carrier CDMA systems utilises the same number of carriers as the spreading sequence length. The carriers overlap and adjacent chips of the spreading sequence modulate adjacent carriers. There is no guard interval and therefore intercarrier interference occurs. If the receiver is synchronised and has a perfect estimate of the channel, it is shown that this multi-carrier CDMA system has comparable performance to a DS-CDMA system of the same bandwidth. It is further shown that it is simple to compute the minimum mean square error criteria as the equaliser consists of <I>N</I> one tap equalisers, where <I>N</I> is the number of carriers. The second system utilises many overlapping low data rate orthogonal carriers. The orthogonality of the carriers is maintained due to cyclically extended guard interval and the number of carriers is much higher than the spreading sequence length. After spreading, the data streams are interleaved onto the carriers to maximise diversity. A practical form of maximum likelihood detection for 64 users is described. It is shown from simulation results that when the system is used in conjunction with ½ rate (constraint length 7) coding and equal gain combining the system can support 64 users at 6 dB <I>E<SUB>6</SUB>/N<SUB>6</SUB></I> for a bit error rate of 2 x 10<SUP>-3</SUP>. This compares with an equivalent DS-CDMA system which can only support 16 users for the same bit error rate and <I>E<SUB>6</SUB>/N<SUB>o</SUB></I>. These results assume perfect channel knowledge and synchronisation. It is further shown that to provide high spectral efficiency in a coded system a high rate convolutional coding scheme is needed. A combined decoder/canceller is also presented. Finally, techniques to achieve synchronisation and channel estimation algorithms are presented. These algorithms are considered in conjunction with the second system. In the framework of synchronisation, methods are presented for frequency and timing synchronisation. For channel estimation, simulation results are presented for a simple channel estimator.

In the contemporary world communication technology has an immense influence on the way we work and behave. For many years now the telephone has been the most commonly used means of interactive communication. Recently, due to the standardisation and commercial application of moving image compression techniques (MPEG, MPEG-II, H.263, H.263+) and the ever increasing power of personal computers, the interest in interactive video communication (videophone) applications has grown considerably. Most of the research concerning video compression techniques avoids the topic of extremely low bit rates (required e.g. for mobile communication). There is currently no technique dedicated to encoding the video for such low data rates (below 10 kbit/s). In most cases a reduction in frame-rate and heavy quantisation would be applied to an existing algorithm designed for a higher data rate. The resulting artefacts would in many cases prevent the recognition of the speaker in a head-and-shoulders scene. In recent years, due to the development of the MPEG-IV standard, there has been a growing interest in model based video coding techniques with algorithms utilising semantic knowledge (wire-frame models) about the scene offering the highest compression ratio. This thesis describes an investigation into the topic of semantic model based coding of typical videophone scenes (head-and-shoulders and head-only). New techniques for automatic fitting of the semantic wire-frame are described and tested. Finally a new algorithm for automatic tracking and a unified approach to both fitting and tracking are presented. Due to very encouraging feedback from other researchers working in the same area, it was possible to publish the results of investigations described in this thesis in 14 journal and conference papers. These are listed at the end of this thesis and one of the published papers is included.

In modern wireless networks the demand for high-speed transmissions is ever increasing to provide access to data and enable new services anywhere and anytime. Mobile internet, video telephony, music and video on demand are examples for the possible applications which demand high data rates. However, the available frequency spectrum is limited and expensive. To satisfy the demand for high data-rates, turbo-encoded multiple-input multiple-output (MIMO) radio links have been recently proposed for the support of high-speed downlink packet access (HSDPA) in UMTS, where the re-use of spreading codes across the transmitter antennas results in high levels of interference. In this thesis, low complexity MIMO receiver architectures and their components are investigated to enable high-speed receivers capable of dealing with high-order modulations. For detection, multi-stage partial parallel interference cancellation (MS-PPIC) and matched-filter based ordered serial interference cancellation (MF-SIC) are proposed as low complexity alternatives to the <i>a posteriori</i> probability (APP) detector and its Max-Log-APP variant. Non-linear cancellation metrics are derived for the MS-PPIC and the performance of the proposed detectors is investigated for different channel conditions. It is shown that the MS-PPIC can provide similar performance compared to the APP and, for low coding rates, superior performance compared to the Max-Log-APP, at a substantially lower computational complexity. While the MF-SIC cannot compete with the APP and MS-PPIC detectors in non-iterative receivers, it provides impressive performance when used in an iterative receiver architecture where <i>a priori</i> information from a decoder is available for ordering and interference cancellation. For decoding, a novel modification of the turbo decoder using the Max-Log-MAP algorithm is proposed, which results in a performance approaching that of a turbo decoder using the optimum Log-MAP or MAP algorithms. The approach aims to maximise the mutual information at the input of each component decoder by correcting the bias in the <i>a priori</i> information caused by the Max-Log approximation in the previous component decoder. This is performed by scaling the <i>a priori</i> information by optimised iteration-specific weight factors at each turbo iteration. Another contribution is a method for the off-line computation of the optimal weights according to the maximum mutual information criterion. Subsequently, different versions of non-iterative and iterative MIMO receiver architectures are proposed and compared in terms of performance and computational complexity for a wide range of detection algorithms using 4-QAM modulation. For iterative receivers which rely on hard cancellation, a soft-output combining scheme which maximises the mutual information at each iteration is proposed and a corresponding method for the offline computation of the optimal combining weights is presented. Finally a novel layered encoding scheme is proposed which overcomes the problem of exponential growth in complexity of the APP detector when higher order modulations such as 16- and 64-QAM are employed. This could be achieved without any loss in performance. Layered encoding also solves the performance and convergence problems of low complexity detectors such as the proposed MS-PPIC and the MF-SIC which occur at higher order modulations. In addition, the applicability of high order modulations in MIMO systems is investigated using system-level simulations for a 2-cell indoor and a 7-cell urban scenario. The results indicate that high-order modulations could be used in a substantial area of the cell.