Whole document restricted, see Access Instructions file below for details of how to access the print copy. / Future telecommunication networks will employ digital transmission techniques. Such networks will provide a number of benefits including the ability to integrate voice and non-voice messages. The transmission channel of this network can be represented by a cascaded channel composed of a number of elementary channels connected in series. Therefore the modelling of such a channel is of particular interest. The influence of noise and other impairments in the cascaded binary channel cause errors which may be represented by a binary signal called the error sequence. Consequently, an important step in digital channel modelling is estimation of parameters and distribution functions which characterise the statistical properties of error sequences in the channel. Thus, the development of efficient methods for this estimation is a problem of long term interest which should be properly solved. This thesis presents methods and techniques for parameter (primarily the probability of error) and distribution function (primarily the error gap complementary distribution function) estimation using the error sequences obtained by measurement or simulation in elementary or cascaded channels. Theoretical analysis and testing confirm that it is possible to control the accuracy and reliability of estimation. Two principal and practical methods for the probability of error estimation are developed: the modified Monte Carlo method (MMC); and the method based on Chebyshev inequality (MCI). In contrast to the traditional Monte Carlo method based on classical statistics, the methods developed in this thesis aim to specify the sample size required to achieve the desired accuracy. The methods developed are based on the dependence of the sample size on the estimated value of a parameter being estimated. Hence the sample size is a random variable and the confidence limits factor (which specifies the width of confidence interval in respect to the estimated value) is a constant. Based on these methods, this thesis proposes and demonstrates two techniques for parameter estimation. The traditional Monte Carlo method has been primarily used for the probability of error estimation in channels without memory. In this thesis the capabilities of this method are extended to the case of estimating the probability of error in channels with memory and cascaded channels. However, even with this extension, this method is not practical due to its complexity and limitations on the qualification and quantification of the accuracy and reliability of estimation. Also, the extended method is unable to satisfactorily estimate the probability of error in cascaded channels with memory; nor could it improve the speed of the estimation process. Two methods and two techniques for distribution function estimation are developed in this thesis. They are demonstrated by estimating the error gap complementary functions of simulated data. For this purpose, simulators of binary channels with and without memory have been developed. The methods and techniques are characterised by their simplicity in application; ability to quantify the accuracy and reliability; time efficiency; and real time capability. The wider application of the methods and techniques developed in this thesis are demonstrated on three examples: a distribution function estimation using data obtained by indoor wideband radio propagation measurement; BER characteristics measurement; and measurement of the residual probability of error in transmission systems using error correcting codes. From the results obtained in the thesis some recommendations for future work in the field of digital channel modelling and simulation are discussed.

This thesis describes a novel approach for adaptive optimal control and demonstrates its application to a variety of systems, including motion control learning for legged robots. The new controller, called “FOX”, uses a modified form of Albus’s CMAC neural network. It is trained to generate control signals that minimize a system’s performance error. A theoretical consideration of the adaptive control problem is used to show that FOX must assign each CMAC weight an “eligibility” value which controls how that weight is updated. FOX thus implements a kind of reinforcement learning which makes it functionally similar to the cerebellum (a part of the brain that modulates movement). A highly efficient implementation is described which makes FOX suitable for on-line control. FOX requires a small amount of dynamical information about the system being controlled: the system’s impulse response is used to choose the rules that update the eligibility values. A FOX-based controller design methodology is developed, and FOX is tested on four control problems: controlling a simulated linear system, controlling a model gantry crane, balancing an inverted pendulum on a cart, and making a wheeled robot follow a path. In each case FOX is effective: it associates sensor values with (and anticipates) the correct control actions, it compensates for system nonlinearities, and it provides robust control as long as the training is comprehensive enough. FOX is also applied to the control of a simulated hopping monoped, and a walking biped. FOX learns parameters that fine tune the movements of pre-programmed controllers, in a manner analogous to the cerebellar modulation of spinal cord reflexes in human movement. The robots are successfully taught how to move with a steady gait along flat ground, in any direction, and how to climb and descend slopes.

In February 1965, a research programme in radio astronomy was started by the Electrical Engineering Department of the University of Auckland. One of the main purposes of the programme was to provide a unified theme for post-graduate research in the department. The initial aim of the programme was to develop through graduate research, sufficient facilities for radio astronomy observations at frequencies below 100Mhz. Being among the first group of graduate students in the programme, the author was given the problem of studying the antenna requirements for the programme. At the frequencies concerned, the antenna systems are often large and expensive. As the programme is supported at present, only by funds for ordinary graduate research, there is a great need for an antenna array with good performance at minimal cost. This has led the author into his main field of study, viz. the synthesis of arrays with non-uniformly spaced directional elements. The use of directional elements together with non-uniform spacing technique permits larger inter-element spacings in the array without resulting in large sidelobes in the response pattern. Available synthesis methods are inadequate despite the large number of papers published on the subject since its introduction 7 years ago. The synthesis problem involves the determination of a set of element positions to give a desirable response pattern. Because the element position variables lie in the arguments of the cosine terms in the pattern function, the problem becomes highly non-linear. For simplicity, most of the published works have assumed isotropic elements. The methods proposed have been mainly centred on some form of linear approximations to the problem. Consequently, these methods are only effective over a limited region of space*. As a result of this limitation, good pattern characteristics can only be achieved with impractically small spacings. For arrays with less than about 50 elements, the element spacings can best be determined by an optimisation procedure. This method involves the repeated application of small pertubations to the element positions of a starting array until maximum improvement to the sidelobe levels of the array is achieved. An efficient perturbation method has been proposed by Baklanov et al using a matrix approach. Because of the inherent limitation of this method, Baklanov’s arrays are mostly impractical due to the occurrence of small spacings. Such limitations are removed by the author through the use of a modified synthesis procedure. With this new procedure, the author was able to control the pattern over a considerably larger area in space. Thus arrays with average inter-element spacings up to two wavelengths can be synthesised with positive control over all sidelobes in the arrays. The sidelobe levels of the author’s arrays are, as a whole, bery close to the levels of the corresponding theoretically optimal patterns. The element directivities are taken into account in the synthesis process. A total of 30 non-uniformly spaced arrays of varying sizes and sidelobe levels were synthesised using the method developed. Since all these arrays have near to optimal sidelobe characteristics, they provide a basis for a detailed study of the properties of non-uniformly spaced arrays as a whole. A number of interesting points are revealed when pattern parameters like gain, beamwidth, sidelobe level, etc., are studied in relation with the spacing characteristics of the arrays. A better understanding of the properties of non-uniformly spaced arrays is also gained by comparing the pattern characteristics of the synthesised arrays with that of current tapered arrays. The design and testing of a 16-element non-uniformly spaced array of Yagi antennas is described in Chapter 4. This array demonstrates one practical application of the synthesis work reported in this thesis. *The word ‘space’ used throughout the Introduction does not mean the physical space, which is defined, in a 2-D case, by the zenith angle θ. Here, the space is defined by the parameter χ=2πdavsinθ/λ. Thus with an average inter-element spacing dav= λ, the visible space is defined by χ=0 to π.

Apart from recognition accuracy, decoding speed and vocabulary size, another point of consideration when developing a practical ASR application is the adaptability of the system. An ASR system is more useful if it can cope with changes that are introduced by users, for example, new words and new grammar rules. In addition, the system can also automatically update the underlying knowledge sources, such as language model probabilities, for better recognition accuracy. Since the knowledge sources need to be adaptable, it is in°exible to statically combine them. It is because on-line modi¯cation becomes di±cult once all the knowledge sources have been combined into one static search space. The second objective of the thesis is to develop an algorithm which allows dynamic integration of knowledge sources during decoding. In this approach, each knowledge source is represented by a weighted ¯nite state transducer (WFST). The knowledge source that is subject to adaptation is factorized from the entire search space. The adapted knowledge source is then combined with the others during decoding. In this thesis, we propose a generalized dynamic WFST composition algorithm, which avoids the creation of non- coaccessible paths, performs weight look-ahead and does not impose any constraints to the topology of the WFSTs. Experimental results on Wall Street Journal (WSJ1) 20k- word trigram task show that our proposed approach has a better word accuracy versus real-time factor characteristics than other dynamic composition approaches.

The vision of Ambient Intelligence (AmI) can be realised through the integration of embedded technologies, distributed systems, middleware and human machine interfaces and many research efforts have been made to advance these technologies. However, the exclusiveness of these ambient intelligence technologies has reduced their practical values. In this thesis, a novel AmI platform is proposed to facilitate the integration and interoperability of various technologies in the process of developing AmI applications. The platform defines the overall software/hardware architecture and communication interfaces and provides a common base for development, operation and future adaptation of AmI applications. The proposed platform consists of four layers, the physical ubiquitous environment, middleware, multi-agent system (MAS) and application layer. The ubiquitous environment layer accommodates any type of embedded device network for interconnecting different sensors, actuators and computing devices. The middleware layer is built using an IP-based service discovery protocol, Universal Plug and Play (UPnP), which provides a unique communication interface for controlling and monitoring embedded devices. The MAS handles the core distributed and adaptive control functionality and communication with user interfaces. The application layer contains any type of user interface for different AmI applications. An XML-based content language is designed with an XML schema and seven XML messages. The content language standardises the way of interpreting contents of communication between different user interfaces and the MAS. Based on the proposed platform, a complete AmI application prototype called Distributed Embedded Intelligence Room (DEIR) has been implemented. Four different device networks, the SmartHouse network, IP network, Bluetooth and Zigbee network, have been integrated in DEIR to interconnect various embedded sensors and devices. The MAS is implemented using Java Agent DEvelopment framework (JADE). Four application specific agents, known as the UPnP control point agent, IP interface agent, fuzzy inference agent and decision tree agent, are designed and implemented. The UPnP control point agent provides MAS the ability to monitor and to control the underlying hardware devices through the UPnP middleware layer. The IP interface agent handles communication with user interfaces over socket connections. Fuzzy inference and decision tree agents are implemented to provide personalised learning and automated control capabilities. Three user interfaces, including a remote graphical user interface, a mobile PDA interface and a 3D virtual reality interface are implemented. Contents of communication between these user interfaces and the MAS are encoded using the proposed XML content language and transmitted over socket connections. The AmI application prototype, DEIR, has demonstrated the ability of integrating multiple device networks and multiple user interfaces, which is a vital feature for most AmI applications. Two case studies have been carried out to incorporate two adaptive learning and controlling algorithms, known as the adaptive online fuzzy inference system (AOFIS) and ID3 decision tree algorithm, in the MAS of DEIR. The results of case studies show that DEIR has the ability of incorporating multiple adaptive control algorithms as multiple agents. In addition, comparable or better offline learning accuracy and learning speed have been achieved by DEIR compared with other advanced adaptive control algorithms. / Whole document restricted, but available by request, use the feedback form to request access.

Whole document restricted, see Access Instructions file below for details of how to access the print copy. / Future telecommunication networks will employ digital transmission techniques. Such networks will provide a number of benefits including the ability to integrate voice and non-voice messages. The transmission channel of this network can be represented by a cascaded channel composed of a number of elementary channels connected in series. Therefore the modelling of such a channel is of particular interest. The influence of noise and other impairments in the cascaded binary channel cause errors which may be represented by a binary signal called the error sequence. Consequently, an important step in digital channel modelling is estimation of parameters and distribution functions which characterise the statistical properties of error sequences in the channel. Thus, the development of efficient methods for this estimation is a problem of long term interest which should be properly solved. This thesis presents methods and techniques for parameter (primarily the probability of error) and distribution function (primarily the error gap complementary distribution function) estimation using the error sequences obtained by measurement or simulation in elementary or cascaded channels. Theoretical analysis and testing confirm that it is possible to control the accuracy and reliability of estimation. Two principal and practical methods for the probability of error estimation are developed: the modified Monte Carlo method (MMC); and the method based on Chebyshev inequality (MCI). In contrast to the traditional Monte Carlo method based on classical statistics, the methods developed in this thesis aim to specify the sample size required to achieve the desired accuracy. The methods developed are based on the dependence of the sample size on the estimated value of a parameter being estimated. Hence the sample size is a random variable and the confidence limits factor (which specifies the width of confidence interval in respect to the estimated value) is a constant. Based on these methods, this thesis proposes and demonstrates two techniques for parameter estimation. The traditional Monte Carlo method has been primarily used for the probability of error estimation in channels without memory. In this thesis the capabilities of this method are extended to the case of estimating the probability of error in channels with memory and cascaded channels. However, even with this extension, this method is not practical due to its complexity and limitations on the qualification and quantification of the accuracy and reliability of estimation. Also, the extended method is unable to satisfactorily estimate the probability of error in cascaded channels with memory; nor could it improve the speed of the estimation process. Two methods and two techniques for distribution function estimation are developed in this thesis. They are demonstrated by estimating the error gap complementary functions of simulated data. For this purpose, simulators of binary channels with and without memory have been developed. The methods and techniques are characterised by their simplicity in application; ability to quantify the accuracy and reliability; time efficiency; and real time capability. The wider application of the methods and techniques developed in this thesis are demonstrated on three examples: a distribution function estimation using data obtained by indoor wideband radio propagation measurement; BER characteristics measurement; and measurement of the residual probability of error in transmission systems using error correcting codes. From the results obtained in the thesis some recommendations for future work in the field of digital channel modelling and simulation are discussed.

This thesis describes a novel approach for adaptive optimal control and demonstrates its application to a variety of systems, including motion control learning for legged robots. The new controller, called “FOX”, uses a modified form of Albus’s CMAC neural network. It is trained to generate control signals that minimize a system’s performance error. A theoretical consideration of the adaptive control problem is used to show that FOX must assign each CMAC weight an “eligibility” value which controls how that weight is updated. FOX thus implements a kind of reinforcement learning which makes it functionally similar to the cerebellum (a part of the brain that modulates movement). A highly efficient implementation is described which makes FOX suitable for on-line control. FOX requires a small amount of dynamical information about the system being controlled: the system’s impulse response is used to choose the rules that update the eligibility values. A FOX-based controller design methodology is developed, and FOX is tested on four control problems: controlling a simulated linear system, controlling a model gantry crane, balancing an inverted pendulum on a cart, and making a wheeled robot follow a path. In each case FOX is effective: it associates sensor values with (and anticipates) the correct control actions, it compensates for system nonlinearities, and it provides robust control as long as the training is comprehensive enough. FOX is also applied to the control of a simulated hopping monoped, and a walking biped. FOX learns parameters that fine tune the movements of pre-programmed controllers, in a manner analogous to the cerebellar modulation of spinal cord reflexes in human movement. The robots are successfully taught how to move with a steady gait along flat ground, in any direction, and how to climb and descend slopes.

In February 1965, a research programme in radio astronomy was started by the Electrical Engineering Department of the University of Auckland. One of the main purposes of the programme was to provide a unified theme for post-graduate research in the department. The initial aim of the programme was to develop through graduate research, sufficient facilities for radio astronomy observations at frequencies below 100Mhz. Being among the first group of graduate students in the programme, the author was given the problem of studying the antenna requirements for the programme. At the frequencies concerned, the antenna systems are often large and expensive. As the programme is supported at present, only by funds for ordinary graduate research, there is a great need for an antenna array with good performance at minimal cost. This has led the author into his main field of study, viz. the synthesis of arrays with non-uniformly spaced directional elements. The use of directional elements together with non-uniform spacing technique permits larger inter-element spacings in the array without resulting in large sidelobes in the response pattern. Available synthesis methods are inadequate despite the large number of papers published on the subject since its introduction 7 years ago. The synthesis problem involves the determination of a set of element positions to give a desirable response pattern. Because the element position variables lie in the arguments of the cosine terms in the pattern function, the problem becomes highly non-linear. For simplicity, most of the published works have assumed isotropic elements. The methods proposed have been mainly centred on some form of linear approximations to the problem. Consequently, these methods are only effective over a limited region of space*. As a result of this limitation, good pattern characteristics can only be achieved with impractically small spacings. For arrays with less than about 50 elements, the element spacings can best be determined by an optimisation procedure. This method involves the repeated application of small pertubations to the element positions of a starting array until maximum improvement to the sidelobe levels of the array is achieved. An efficient perturbation method has been proposed by Baklanov et al using a matrix approach. Because of the inherent limitation of this method, Baklanov’s arrays are mostly impractical due to the occurrence of small spacings. Such limitations are removed by the author through the use of a modified synthesis procedure. With this new procedure, the author was able to control the pattern over a considerably larger area in space. Thus arrays with average inter-element spacings up to two wavelengths can be synthesised with positive control over all sidelobes in the arrays. The sidelobe levels of the author’s arrays are, as a whole, bery close to the levels of the corresponding theoretically optimal patterns. The element directivities are taken into account in the synthesis process. A total of 30 non-uniformly spaced arrays of varying sizes and sidelobe levels were synthesised using the method developed. Since all these arrays have near to optimal sidelobe characteristics, they provide a basis for a detailed study of the properties of non-uniformly spaced arrays as a whole. A number of interesting points are revealed when pattern parameters like gain, beamwidth, sidelobe level, etc., are studied in relation with the spacing characteristics of the arrays. A better understanding of the properties of non-uniformly spaced arrays is also gained by comparing the pattern characteristics of the synthesised arrays with that of current tapered arrays. The design and testing of a 16-element non-uniformly spaced array of Yagi antennas is described in Chapter 4. This array demonstrates one practical application of the synthesis work reported in this thesis. *The word ‘space’ used throughout the Introduction does not mean the physical space, which is defined, in a 2-D case, by the zenith angle θ. Here, the space is defined by the parameter χ=2πdavsinθ/λ. Thus with an average inter-element spacing dav= λ, the visible space is defined by χ=0 to π.

Apart from recognition accuracy, decoding speed and vocabulary size, another point of consideration when developing a practical ASR application is the adaptability of the system. An ASR system is more useful if it can cope with changes that are introduced by users, for example, new words and new grammar rules. In addition, the system can also automatically update the underlying knowledge sources, such as language model probabilities, for better recognition accuracy. Since the knowledge sources need to be adaptable, it is in°exible to statically combine them. It is because on-line modi¯cation becomes di±cult once all the knowledge sources have been combined into one static search space. The second objective of the thesis is to develop an algorithm which allows dynamic integration of knowledge sources during decoding. In this approach, each knowledge source is represented by a weighted ¯nite state transducer (WFST). The knowledge source that is subject to adaptation is factorized from the entire search space. The adapted knowledge source is then combined with the others during decoding. In this thesis, we propose a generalized dynamic WFST composition algorithm, which avoids the creation of non- coaccessible paths, performs weight look-ahead and does not impose any constraints to the topology of the WFSTs. Experimental results on Wall Street Journal (WSJ1) 20k- word trigram task show that our proposed approach has a better word accuracy versus real-time factor characteristics than other dynamic composition approaches.

The vision of Ambient Intelligence (AmI) can be realised through the integration of embedded technologies, distributed systems, middleware and human machine interfaces and many research efforts have been made to advance these technologies. However, the exclusiveness of these ambient intelligence technologies has reduced their practical values. In this thesis, a novel AmI platform is proposed to facilitate the integration and interoperability of various technologies in the process of developing AmI applications. The platform defines the overall software/hardware architecture and communication interfaces and provides a common base for development, operation and future adaptation of AmI applications. The proposed platform consists of four layers, the physical ubiquitous environment, middleware, multi-agent system (MAS) and application layer. The ubiquitous environment layer accommodates any type of embedded device network for interconnecting different sensors, actuators and computing devices. The middleware layer is built using an IP-based service discovery protocol, Universal Plug and Play (UPnP), which provides a unique communication interface for controlling and monitoring embedded devices. The MAS handles the core distributed and adaptive control functionality and communication with user interfaces. The application layer contains any type of user interface for different AmI applications. An XML-based content language is designed with an XML schema and seven XML messages. The content language standardises the way of interpreting contents of communication between different user interfaces and the MAS. Based on the proposed platform, a complete AmI application prototype called Distributed Embedded Intelligence Room (DEIR) has been implemented. Four different device networks, the SmartHouse network, IP network, Bluetooth and Zigbee network, have been integrated in DEIR to interconnect various embedded sensors and devices. The MAS is implemented using Java Agent DEvelopment framework (JADE). Four application specific agents, known as the UPnP control point agent, IP interface agent, fuzzy inference agent and decision tree agent, are designed and implemented. The UPnP control point agent provides MAS the ability to monitor and to control the underlying hardware devices through the UPnP middleware layer. The IP interface agent handles communication with user interfaces over socket connections. Fuzzy inference and decision tree agents are implemented to provide personalised learning and automated control capabilities. Three user interfaces, including a remote graphical user interface, a mobile PDA interface and a 3D virtual reality interface are implemented. Contents of communication between these user interfaces and the MAS are encoded using the proposed XML content language and transmitted over socket connections. The AmI application prototype, DEIR, has demonstrated the ability of integrating multiple device networks and multiple user interfaces, which is a vital feature for most AmI applications. Two case studies have been carried out to incorporate two adaptive learning and controlling algorithms, known as the adaptive online fuzzy inference system (AOFIS) and ID3 decision tree algorithm, in the MAS of DEIR. The results of case studies show that DEIR has the ability of incorporating multiple adaptive control algorithms as multiple agents. In addition, comparable or better offline learning accuracy and learning speed have been achieved by DEIR compared with other advanced adaptive control algorithms. / Whole document restricted, but available by request, use the feedback form to request access.