<p>We show that in the additive white Gaussian noise channel that a single neuron using the standard sigmodial nonlinearity function is effectively a Bayesian estimator for a binary (±1) level signal and establish a link between the weight(s) of the neuron and the noise variance associated with the signal. This relationship is then extended to neurons where the sigmodial nonlinearity function incorporates a gain term. By extending these results to multi-level signalling we develop a new nonlinearity function that incorporates a gain α which simplifies the structure of the neural network. We show that this gain term in linked to the noise variance much like the gain term in a neuron using sigmodial nonlinearity for binary signalling.</p> <p>In applying neural networks to the channel equalization problem for frequency selective fading we make use of complex neurons in the neutral network. We look at both binary signalling in two dimensions, and higher level signaling as well. Our results show that while neural nets provide a significant performance increase in the case of binary signalling in two dimensions this performance is not reflected in the results for the higher level signalling schemes. In this case the neural net equalizer performance tends to parallel that of the linear transversal equalizer.</p> <p>Under the Rayleigh fading channel, we demonstrate the feasibility of using a simple neural network as a phase predictor to significantly reduce the error rate for quadrature phase shift keying signalling in this channel. With this network it is possible to remove the error floor that is normally encountered in this channel for average signal to noise ratios of less than 60dB.</p> / Doctor of Philosophy (PhD)

<p>This thesis deals with the repetitive learning control of robotic manipulators. The research topic is motivated by the fact that industrial robots usually perform repetitive tasks. Unlike other conventional controllers, the repetitive learning controller can improve the performance of a robot as it repeats the same task. The proposed control strategy has advantages of easy implementation, low cost and better performance. A critical survey of the state of the art in repetitive learning control of robots is presented. This thesis is divided into two major parts: a two operational modes based approach and a neural network based scheme.</p> <p>In the first part of the thesis, a theoretical framework has been developed for designing controllers based on the proposed two operational modes. The repetitive operations of a robot can be divided into two operational modes: a single operational mode and a repetitive operational mode. Based on this, the author proposes repetitive learning schemes for motion control of both rigid and flexible joints robots, as well as for control of constrained robots with rigid joints. The designed controllers converge in both operational modes. In the single operational mode, the controllers behave like adaptive controllers, while similar to a betterment process in the repetitive operational mode. These conclusions are firmly supported by simulation studies.</p> <p>In the second part of this thesis, special attention is paid to schemes based on neural networks. The author's own modification of back-propagation neural networks is first established in the thesis. It is then applied to motion control of both rigid and flexible joints robots, as well as force control of rigid robots. The neural networks play the role of an approximate inverse dynamic model of a robot and are then parallelled with a conventional PD controller to achieve control goals. Very promising results have been reported in this thesis.</p> / Doctor of Philosophy (PhD)

<p>Transient stability analysis, a necessary tool for planning and operating electrical power systems, is most time consuming to use because of its massive computational burden. Therefore, the solution of this problem on multiprocessor systems has attracted a lot of attention and holds the promise of eventual real-time dynamic security assessment.</p> <p>In this thesis, methods to speedup the computation of power systems transient stability simulations using parallel distributed processing are examined. The approach of the thesis is twofold. Firstly, practical implementation must consider the use of current widely available networked computers in the industry. Secondly, the parallel distributed algorithms to be developed in this thesis must as far as possible retain algorithmic advances that have been made on its serial counterpart because of their many advantages.</p> <p>Serial transient stability algorithms are studied and all major algorithmic advances are investigated and implemented. State-of-the-art parallel transient stability calculations are reviewed and for each algorithm the parallelism that has been exploited is identified. The choice of parallel processing hardware and software is of major concern to power utilities. These issues are discussed and justifications for using networked workstation are given.</p> <p>The two approaches mentioned above lead naturally to coarse-grained partitioning of the transient stability problem. A slow coherency network partitioning that divides the network into several coherent areas along naturally occurring weak links is developed in this thesis. Following network partitioning, computational tasks can be distributed to several computers and then executed simultaneously. Solution of large, sparse algebraic linear systems is required in most power system analysis. In transient stability analysis, the solution of the linear transmission network is the major obstacle to real-time simulations. In this thesis, both direct and iterative methods for solving blocked linear algebraic systems associated with the power network are developed and implemented on a cluster of workstations. RPC (Remote Procedure Calls) techniques together with multiprocessing primitives are used for client-server communications. These aspects and techniques are fully discussed in this thesis.</p> <p>Based on the network partitioning and solution methods developed, a transient stability algorithm is parallelized. The process of parallelization is carefully demonstrated. Both shared memory multiprocessor and RPC based--distributed memory--versions are developed. The corresponding speedup due to these methods are analyzed and discussed. It is shown that techniques developed in this thesis can be applied to production grade transient stability programs.</p> / Doctor of Philosophy (PhD)

<p>Theoretical and experimental techniques, which are useful in molecular spectroscopic studies and in the development of Laser Absorption Spectrometers (LAS), are described. Both fixed-frequency and tunable lasers have been employed. All the measurements reported are based on either direct or second harmonic absorption techniques.</p> <p>A commercial CO/CO₂ line-tunable laser has been used to obtain useful criteria for air pollution monitoring via the direct absorption scheme. CO₂ laser absorption measurements on ozone at reduced pressures are reported. Measurements on seventeen NO absorption lines with a CO laser are described. For these, the absolute absorption as a function of pressure has been determined. As a result we have been able to establish accurate values for the absorption of NO at the pertinent CO laser wavelengths. From the best fit between experimental measurements and theoretical calculations we have deduced the separation between the appropriate NO and CO wavelengths, the NO/N₂ pressure broadening, the NO band strength, and the individual NO line strengths.</p> <p>Using a tunable lead-salt semiconductor diode laser, and employing the direct absorption technique, we have developed a simple method for accurate frequency, measurements of ozone absorption lines. This technique is based on employing a tunable semiconductor diode laser, an etalon, and a White cell. We report accurate frequency meaaurements of over 100 absorption lines of ozone in the ν₃ band, which are near the CO₂ laser transitions. Despite the relative simplicity of the technique, we achieved an accuracy of better than 10 MHz in our measurements. This accuracy compares well with that achieved by the more complicated heterodyne techniques of high-resolution spectroscopy. However, these are the first measurements over an entire absorption band using the full resolution of a tunable diode laser.</p> <p>A LAS, which combines the sensitivity of the acousto-optical methods with the convenience of direct, long-path optical detection, is described. The diode laser is wavelength modulated and the second harmonic detection technique is applied. This technique enables us to detect atmospheric pollutant gases with extremely high sensitivity (3 ppb of a weak absorbing pollutant such as SO₂ in its ν₁ band or 3 x 10ˉ³ ppb of a strongly absorbing molecule such as CO). This sensitivity is achieved using a frequency-locking technique for the diode. We report on the detection of O₃, SO₂, NH₃, and N₂O pollutants at their ambient, levels in air. A very high specificity and virtual elimination of interference effects are obtained by sampling the atmospheric air at reduced pressures.</p> <p>A remote LAS station, which is considered to be adjunct to the above LAS, has been built. A remote retroreflector and an off-axis telescope near the laser have been employed to achieve a total pathlength of 1.2 km in the atmosphere. We report theoretical and experimental investigations of SO₂, H₂O, NH₃ and H₂O absorption in the region 1100 - 1200ˉ¹ cm at atmospheric pressure. This enabled us to develop a technique, based on the second harmonic detection scheme, to detect SO₂ in the atmosphere with a sensitivity of 50 ppb employing a diode laser emitting in the ν₁ band of SO₂. We report also on spectroscopic studies and absorption measurements for SO₂ and H₂O in the region 1300 - 1400 cmˉ¹. These studies indicate that it is possible to detect SO₂ with ambient levels in the atmosphere by employing a diode laser radiating in the ν₃ band of SO₂, near 1331.5 cmˉ¹.</p> <p>In all the above mentioned techniques, theoretical calculations of laser transmission through the atmosphere must be predicted prior to applying such techniques. For this reason, a computer package has been developed and tested. This package generates the absorption line parameters of any molecule, even asymmetric-top molecules, in the vibration-rotation infrared region. The absorption (transmission) can be calculated at any pressure.</p> / Doctor of Philosophy (PhD)

<p>This thesis presents a unified treatment of circuit and system design methods embodying centering, tolerancing and tuning. The approach incorporates the nominal parameter values, the corresponding tolerances and tuning variables simultaneously into an optimization procedure designed to obtain the best values for all of them in an effort to reduce cost, or make an otherwise impractically toleranced design more attractive. Intuitively, the aim is to produce the best nominal point to permit the largest tolerances and the smallest tuning ranges (preferably zero) such that one can guarantee, in advance, that in the worst case, the design will meet all the constraints and specifications. Reduced problems are formulated for digital computer implementation. Interpretations are given. Implications of biquadratic functions in the circuit tolerance problems are investigated. Practical implementation in circuit design problems in the frequency domain is treated. The thesis also includes illustrative examples and two realistic problems.</p> / Doctor of Philosophy (PhD)

<p>Wavelength division multiplexing technologies are proving to be an effective method for bring high bandwidth to data communication networks. Such networks utilize some form of wavelength agility at stations in order to access a set of wavelength channels. Since each channel is represented by a unique wavelength, transmissions on one channel are independent from other channels. Unfortunately, this multi-channel nature acts to complicate system operation and often results in increased optical hardware requirements. Recent advances in photonic amplification have re-motivated the use of ring and bus topologies in single-hop wavelength-division-multiplexed (WDM) networks (Muk92). Most often stations on a ring or bus WDM network utilize multiple transmitters and receivers which are coupled to the fiber via passive devices. Compared to networks based upon passive optical stars, ring and bus networks offer much simplified station synchronization requirements. However, when minimal station hardware is desired, network operation can be complicated. In this thesis, four different WDM bus/ring networks based upon passive coupling technology are considered. All four have user station hardware designs with various reductions in the number of tapping points, number of transmitters and receivers and their tunability requirements. With each reduction in hardware, protocol complexities and performance reductions are introduced. In all cases, dynamic packet-switched operation is achieved. The designs thus give an indication of the cost/performance tradeoffs which are possible as the amount of hardware is reduced at the user stations. Common to each of the designs is the use of a headend controller which attaches to each WDM channel. Media access is achieved through information provided by the controller. In the more hardware intensive designs, media access is achieved through a mini-slot contention mechanism (GT94). The remaining designs rely on a hybrid opto-electronic request/allocation protocol motivated by DCCN (JT93). Analytical models are presented which are validated through discrete-event simulations for each design. Although this passive tap arrangement is very simple, synchronizing to upstream transmissions is much more difficult compared to existing single-channel busses and rings. This affects the types of protocols which are practical in passive designs. In this thesis, we also investigate protocols for multichannel photonic bus/ring LANs which use active wavelength-selective taps rather than passive ones. An example of such an implementation might use couplers based on acousto-optic devices (Che90). There are a number of advantages in such a design. Since stations actively tap the wavelength channel that they are accessing, synchronization is identical to existing single-channel rings. This results in much simpler media access protocols than in the passive case. In addition, spatial re-use is possible in the ring designs. In LANs with active taps, when stations re-tune their couplers at an inappropriate time, transit packets may be destroyed. This action is referred to as a "re-tuning collision". The media access protocols must take special actions in order to prevent this from happening. This places restrictions on the physical parameters over which system performance is efficient. Capacity and delay models are also derived, and comparisons are made with the conventional passive designs developed in this thesis.</p> / Doctor of Philosophy (PhD)

<p>This thesis is a report on a Unified Model to predict the performance of Aluminum Conductor Steel Reinforced (ACSR). The Model combines electromagnetic, mechanical, radial conduction, and steady-state thermal models, and uses a probabilistic method to take account of system loads and the stochastic nature of the meteorological parameters. The four models, developed during the past thirty years, have been used by the industry individually to predict the electrical, mechanical and thermal behaviour of stranded overhead conductors. Each aspect was analyzed separately, neglecting dependencies of the involved variables on other aspects of the conductor behaviour. Most results obtained in this way were acceptable while conductors were operating at lower temperatures. However, at high temperatures, the effects that each aspect of the behaviour has on another are no longer negligible. Until now there has not been a feasible mechanism to link the models in order to predict all aspects of the behaviour of the conductor simultaneously, and to take into consideration all the interdependent influences among models. The Unified Model, developed as a result of this work, is a new approach for the simultaneous prediction of all the aspects of conductor behaviour. For the first time all four models: electromagnetic, mechanical, steady-state thermal, and radial conduction models, are coupled such that the conductor is viewed as a whole. In addition, the statistical distribution of system loads, ambient temperature, and rise of conductor temperature above ambient are included in the Unified Model to determine a realistic thermal history of the conductor and to assess annealing in the aluminum layers. This has not been done before. The usefulness of the probability based Unified Model for predicting characteristics of the conductor, such as sag, power loss and loss of tensile strength of nonferrous wires is demonstrated. An example of application of the Model to an ACSR conductor is given. The results are used to study the sensitivity of the electrical, mechanical, and thermal characteristics of the conductor to various parameters and to determine how the electrical, mechanical, and thermal characteristics affect each other. The results are presented in the form of recommendations for the design of the overhead conductors.</p> / Doctor of Philosophy (PhD)

<p>A new, computationally efficient windowing methodology for motion tracking is described. The proposed approach is well suited to real-time focus-of-attention applications in which regions-of-interest, or windows, are used to reduce image data rates. Applications include robot guidance, where high speed image processing is required for real-time position control in operations such as fixtureless assembly for flexible manufacturing. A hierarchy of windowing functions which includes motion detection and target detection and tracking has been developed. This has resulted in a new algorithm for corner detection in image windows, as well as a proposal for measuring the information content of an image based on corner location accuracy. The techniques have been experimentally verified with the implementation of a vision system based on a high speed digital camera, a custom-built video interface board, and a network of digital signal processors. Dynamically positioned at video frame rates, windows within the camera field-of-view are made cooperative by exchanging information among their corresponding processors to allow real-time adaptation to visual motion. A cooperative windowing scheme using two networked target tracking windows is demonstrated. Motion tracking is based on the best-case output of the simultaneous application of a feature-based algorithm applied in the first window and a model-based algorithm running in the second. The experimental results demonstrate the advantages of motion tracking using this cooperative windows approach.</p> / Doctor of Philosophy (PhD)

<p>Information technology is generally believed to enhance learning processes. As a result, researchers are beginning to seek new theories and approaches towards developing information technology enabled learning environments. The focus of the research in this thesis is to provide the structure of an intelligent tutoring system (called MIS-Tutor) that we developed in support of mastery learning for students registered in a university-level management information system course. We adopted formative evaluation, spanning over 4 years and that included the participation of 1,328 students, to evaluate the effectiveness of the MIS-Tutor. An important component of this intelligent tutoring system is the ability to adapt its interaction to individual learning behaviour. This is achieved by means of neural network models to reinforce long-term memory retention and to assess the required time-on-task for each student. The results of our formative evaluation support the effectiveness of the proposed models in support of students' mastery learning.</p> / Doctor of Philosophy (PhD)

<p>The theoretical aspects of coplanar waveguides were studied with special emphasis on the attainment of precise characteristic impedances and low losses in the 1-2 Ghz frequency range. To this end, a literature survey was performed, and characteristic impedance design curves were reproduced exactly by means of a digital computer. An exact tolerance analysis method was then performed in order to determine geometrical tolerances which would enable a characteristic impedance of 50 Ω +/- 10% to be attained for a given relative dielectric constant of 9.2 +/- 10%. A loss theory which neglected dielectric losses and assumed thin films of conducting metal less than 1 skin depth thick was also presented. Two similar fabrication techniques were studied and 43 coplanar waveguide samples were fabricated. Of these 43 samples, 32 were tested by swept frequency techniques and time domain reflectometer techniques.</p> <p>The theory for characteristic impedance was found to be accurate to within +/- 2 Ω providing a few simple design guidelines were followed. The transmission loss theory corresponded very well with experimental measurements, which were estimated to be accurate to within 22%. Furthermore, fixtures for mounting the samples quickly and repetitively were designed and fabricated.</p> / Master of Engineering (ME)