Markov decision processes form an important class of dynamic programming problems because they are widely applicable. However solving real applications of Markov decision processes on serial computers is often impractical due to constraints on memory and processing time. Parallel processing has long been considered a potential solution to the computational intractability of these problems on serial machines, but prior to this work no detailed theoretical or practical studies in this area had been carried out. This thesis examines several successful serial iterative solution methods for infinite horizon, time invariant, discounted Markov decision processes and develops efficient analogous parallel algorithms. Particular consideration is given to the two classes of iterative solution methods known as value iteration methods and reward revision, but the techniques developed and the conclusions drawn are applicable to other iterative methods for Markov decision processes (for example policy iteration methods) and also to iterative methods in general. Iterative methods are applied to many other problem areas including dynamic programming and the solution of linear and differential equations. The main thrust of this thesis is concerned with the optimisation of the performance of the parallel algorithms developed. A detailed analysis of the implementation of several parallel iterative solution methods on a distributed memory, multiple instruction, multiple data, parallel processor reveals the key issues involved in optimising performance. Timing models are developed for processor communication time, processor calculation time and overall run time. These models guide the choice of the connection topology, the communication protocols and the degree of overlapping of communication and calculation. This leads to the development of a phased pipeline algorithm which yields 60 fold speed-ups when a ring of 121 transputers is used to solve problems with 60,000 states and sparse transition structures.

Ferroelectric liquid crystal over silicon (FLCOS) technology, initially developed for coherent optical computing applications, is identified within this thesis as a highly suitable technology for miniature colour display development. This is evident as surface stabilised ferroelectric liquid crystals exhibit high switching speeds, a wide field of view and a digital switching mechanism, while very large scale integrated circuits offer high frequency, low power operation and versatile functionality. This thesis introduces the use of FLCOS combined with fast, high-intensity, primary-coloured light sources in order to achieve colour images with no colour separation. The digital nature of this technology permits high speed temporal multiplexing of illuminated binary images to produce a visual grey- or colour-level sensation. Accurate colour reproduction is achieved consistently as the silicon backplane, ferroelectric liquid crystal, illumination sources and overall electronic drive schemes are digitally controlled. Optical analysis and characterisation of available DRAM (Dynamic Random Access Memory) based FLCOS backplanes had led to the development of a number of optimised device-dependent electronic driving schemes. One such scheme aims to reduce the visual effects of image degradation due to the photoelectric effect which was apparent in early DRAM based devices. A generic interfacing platform, real-time video processing architecture and a variety of image manipulation hardware functions for FLCOS displays are presented. These functions have been implemented on fast programmable logic thus verifying correct operation and functionality. This technology has great potential for head mounted applications, where portability and high quality imagery are required. A collaborative research project 'SLIMDIS' has produced a colour head mounted unit incorporating a 1024x768 pixellated FLCOS device - a world first. The first colour images obtained on this device use driving electronics and addressing schemes which are presented in this thesis.

In this thesis, we propose to use matching pursuit (MP) to specify the basis expansion model (BEM) coefficients for fast time-varying flat-fading channels. Further, sequential matching pursuit (SMP) is proposed by removing the selection process of MP so that the computational complexity is significantly reduced. Via computer-based simulations, it is shown that MP and SMP are able to provide a robust channel estimate when using more basis functions than necessary. A mathematical expression of eth SMP performance is approximately established, and can be used to represent the MP performance to a certain extent. Besides, MP and SMP even perform well without the information of signal-to-noise ratio (SNR). The applications of MP and SMP in flat-fading channels are extended to an orthogonal frequency division multiplexing (OFDM) system. For OFDM, there are channel variations in the time and frequency directions. Firstly, flat-fading subcarrier channels in OFDM are represented by the BEM in the time direction. The studies of MP and SMP for flat-fading channels are suggested to directly apply to the subchannels in OFDM. Secondly, a non-sample-spaced channel with a continuous power delay profile is described by the BEM along the frequency direction in OFDM. Both MP and SMP are also employed to determine the BEM coefficients in such an occasion. Finally, a cascading channel estimator formed by combining the above two channel estimation schemes in the time and frequency directions is studied. The effectiveness of these channel characterization schemes is verified by the simulation study. The robustness of MP and SMP, when the BEM uses a larger model order than the optimal one, is confirmed by the simulation result.

No description available.

The use of personal, sensitive information, such as privileges and attributes, to gain access to computer resources in distributed environments raises an interesting paradox. On one hand, in order to make the services and resources accessible to legitimate users, access control infrastructure requires valid and provable service clients' identities or attributes to make decisions. On the other hand, the service clients may not be prepared to disclose their identity information or attributes to a remote party without determining in advance whether the service provider can be trusted with such sensitive information. Moreover, when clients give out personal information, they still are unsure of the extent of propagation and use of the information. This thesis describes an investigation of privacy preserving options in access control infrastructures, and proposes a security model to support the management of those options, based on extensible Access Control Markup Language (XACML) and Security Access Markup Language (SAML), both of which are OASIS security standards. Existing access control systems are typically unilateral in that the enterprise service provider assigns the access rights and makes the access control decisions, and there is no negotiation between the client and the service provider. As access control management systems lean towards being user-centric or federated, unilateral approaches can no longer adequately preserve the client's privacy, particularly where communicating parties have no pre-existing trust relationship. As a result, a unified approach that significantly improves privacy and confidentiality protection in distributed environments was considered. This resulted in the development of XACML Trust Management Authorization Infrastructure (XTMAI) designed to handle privacy and confidentiality mutually and simultaneously using the concept of Obligation of Trust (OoT) protocol. The OoT enables two or more transaction parties to exchange Notice of Obligations (NoB) (obligating constraints) as well as Signed Acceptance of Obligation (SAO), a proof of acceptance, as security assurances before exchange of sensitive resources.

Wireless Sensor Networks (WSNs) are computer networks consisting of miniaturised electronic devices that aim to gather and report information about their environment. The devices are limited in computational, data storage and communication ability. Furthermore, the devices communicate via a wireless, unregulated medium and usually operate on finite power sources. Security in Wireless Sensor Networks is the research area that seeks to provide adequate and energy-efficient security mechanisms for WSNs. Such provision is required in order to increase their range of possible applications and allow them to be deployed in critical and valuable environments. Existing security mechanisms for larger computer networks are inappropriate since they were not designed for the resourceconstrained environment of WSNs. There are some purpose-built solutions but this research has found potential security or efficiency problems with each of them. This thesis contributes SecRose, a security mechanism for the data-transportation layer of Wireless Sensor Networks. The solution attempts to provide higher level of security than currently provided, without introduction of significant energy overheads and by retaining backwards compatibility. SecRose achieves its security objectives by introducing a number of innovations and improvements. SecRose innovates in the provision of freshness and semantic security by altering the secret cryptographic keys. The process is managed at the transportation level by the basic key management mechanism. The integrity and safety of the key-changing operation is achieved by authenticating all packets and their acknowledgements. This behaviour contrasts with other proposals, which are based on openly transmitted Initialisation Vectors, and allows SecRose to provide better security than most of them, including TinySec, the accepted standard. In addition, measurements show that SecRose provides better energy-efficiency than other proposals. In particular, the solution requires less energy than TinySec in all cases and it can even be more efficient than the base Operating System, the TinyOS, which does not provide any security at all.

No description available.

The quest for achieving an efficient, reliable and cost-effective network infrastructure in support of innovative and rich communication services has resulted in the advent and popularity of IP based converged Next Generation Networks (NGN). According to the ITU-T, the NGN has significant advantages such as support for end to end Quality of Service (QoS), generalised mobility, converged services between fixed & mobile networks and interworking with legacy networks. These networks require Network Management Systems (NMS), which play a key role in monitoring and administering them, to ensure smooth running of services and optimum utilisation of network resources. However, it has been observed that a disparity exists in the pace of development of the NGN and the existing NMSs. This thesis takes up the challenge of addressing this disparity by providing autonomic, intelligent and scalable NM solutions with focus on QoS Management, Distributed Monitoring & Control and Traffic Engineering. Traditional methods of network management do have shortcomings in providing the level of dependability and reliability which is required in the current NGN environment. Machine Learning (ML) approaches have gained popularity as the foundation of intelligent and automated systems, since they are capable of modelling the system behaviour through the process of learning, based on the observation of the system over a period of time. Once appropriately trained they can automatically estimate and predict future system behaviour with high accuracy and speed. Hence, the overall objective of this research is design, development and evaluation of ML based network management mechanisms for Admission Control , Distributed Monitoring & Control and Intelligent Traffic Engineering. In the pursuit of fulfilling the stated objective, the Bayesian Networks (BN) approach emerged as a viable and efficient solution to the identified specific problems.

This thesis aims to examine ways of increasing the spectral efficiency of a communication system while meeting the QoS needs of the users. One of the most prominent antenna techniques that has received a lot of research recently is multiple input multiple output (MIMO) antenna array systems, where both transmitter and receiver are equipped with multiple antenna elements exploiting space (antenna) diversity. Another interesting technology for realizing very high data rates over the rich scattering wireless channel is the VBLAST (Vertical Bell Labs Space Time) architecture. These techniques are thoroughly described, and their BER (Bit Error Rate) versus SNR (Signal-to-Noise Ratio) performance is evaluated by means of computer simulations in various channel environments and using different numbers of transmit antenna elements in the base station. These results of the techniques, along with other characteristics, are compared to examine their performance in various channel environments and investigate which technique is most suitable for each channel environment. Next, we concentrate on round robin packet scheduling. A Packet Scheduler (PS) is aiming to effectively allocate radio resources to users in different channel conditions. The objectives are to maximize spectral efficiency and to provide fairness among users. We investigate the performance of various schemes based on the round robin scheme (RRS) and we propose some new algorithms. Computer simulation are conducted to compare the different scheduling schemes in terms of cell throughputs and outage capacities. The degree of fairness (in terms of time delay and data rates) among the users for these schemes is also investigated. Finally, we focus on Proportional Fair (PF) scheduling. New algorithms using MIMO antennas and VBLAST architecture are proposed to improve the performance of the PF scheme. The system efficiency (in terms of cell throughput) and the QoS (in terms of time delay and data rates) achieved by these schemes are investigated through computer simulations. The main objective of this work is to investigate the interplay of spatial multiplexing techniques (MIMO arrays and VBLAST architecture) with packet scheduling schemes, as well as the tradeoffs between the QoS provided to the end user and the achievable network throughput.

No description available.