<p>A new adaptive procedure for the detection of a random signal in a background of colored interference plus white noise, all having unknown statistics is derived. The procedure is based on the development of linear models for the two hypotheses H₀ and H₁. Under hypothesis H₀ a wanted signal is absent and under H₁ the wanted signal to present. After parameterizing the input signal under the two hypotheses, the log likelihood ratio is easily established in terms of the innovations processes. The test statistic thus obtained has an intuitive interpretation, which conversely provides a basis for the choice of model candidates for the signal detection problem. Different modelling approaches generally result in different receiver performance. Therefore, choosing appropriate type of model for the two pypotheses is a key step in order to enhance detection. A computationally efficient technique has been developed for this purpose.</p> <p>This adaptive detection method is applied to construct a radar detection scheme for detecting a moving target in a surveillance radar environment. The detector is composed of a pair of adaptive whitening filters followed by a log likelihood ratio calculator. The new scheme is tested by using actual radar data, including weather and ground clutter. Comparison is made with the previously described scheme known as innovations-based detection algorithm (IBDA). The results show that the performance of the new scheme is much better than the IBDA.</p> <p>In a related study, the spatial correlation between radar clutter from adjacent rings is exploited to construct two new adaptive clutter suppression schemes. The schemes are tested by real radar data. Experimental results show that these schemes are around 5 dB and 10 dB better than the adaptive lattice prediction-error filter in terms of improvement factor, respectively.</p> / Doctor of Philosophy (PhD)

<p>The problem of identification (estimation of system parameters and characteristics) is considered. The aspect emphasized is the robustness in identification. By saying robustness, two points are meant. One is the robustness with respect to bad pieces of data and the other is the numerical robustness with respect to truncation and round-off errors in computation.</p> <p>A thorough study has been made on the robust statistical principles and their applicability in system identification is critically evaluated. Off-line and on-line identification algorithms are proposed, which are resilient to undetectable spurious errors in the data and at the same time computationally simple and efficient. The convergence of these algorithms is theoretically established. Besides, a robust recursive algorithms is also proposed which jointly estimates the states and parameters of a linear system in a bootstrap manner. This algorithm is also proven to be converging with probability one. In addition to this, a very general method is developed for evaluating the asymptotic efficiency of robust identification methods. The superiority of the proposed approaches in contrast with the conventional identification methods (least squares and its genaralizations) is illustrated with the help of several simulated as well as real-life examples.</p> <p>The numerical instability caused by improper choice of sampling rates is also subjected to considerable study in the context of identification of continuous-time systems from samples of input-output data. Methods are suggested to overcome this problem. Also, numerically robust schemes are introduced for transforming discrete-time models to their continuous-time equivalents and their performance is compared with other existing methods.</p> / Doctor of Philosophy (PhD)

<p>The adaptive control of linear discrete time multivariable systems is considered. A unifying survey of a number of adaptive control strategies is presented. The various algorithms are shown to be special cases of a more general algorithm. The state space design of self-tuning controllers is considered in detail. Two new algorithms for state space pole assignment self-tuning control are proposed. The first algorithm follows an explicit approach, thus a modification of the bootstrap estimator was used for joint parameter and state estimation of an innovations model. The resulting self-tuning controller is more efficient computationally than the methods based on block canonical forms since a minimal realization can be adopted. The second algorithm may be regarded as an implicit pole assignment controller. The recursive prediction error algorithm is used for joint parameter and state estimation in the controller canonical form. The main contribution of this approach is that on-line computation of transformation matrices is avoided. The subsequent computation of controller parameters is trivial, and the resulting self-tuning controller is robust to over-parameterization. To demonstrate a practical application, the second algorithm was used to design a robust autopilot for a simulated nonlinear model of a Royal Navy frigate subjected to sea disturbances. The autopilot was found to perform well for both the course keeping and course changing modes.</p> / Doctor of Philosophy (PhD)

<p>The problem of identification of linear multivariable continuous-time systems from input-output data is considered. A survey has been made to present the direct and the indirect approaches in identifying continuous-time systems from the samples of the observations. The direct approach with approximate integration seems to be more promising and hence it is adopted in this work. Three direct methods based on the use of block pulse functions, trapezoidal pulse functions and cubic splines have been compared and applied for multivariable systems. A comprehensive study has been conducted to analyze the effect of noise on the identification. The analysis was carried out first for the single-input single-output case and then extended to the multivariable case. A new approach is presented to overcome the combined effect of the errors in the approximation/and additive white noise on the identification of continuous-time systems. The method consists of modelling the combined error term. Extensive simulations are conducted in order to illustrate the merits of the new procedure. The problem of order determination has been considered and three order determination tests have been studied and applied for continuous-time systems, two of them for the first time as far as the author is aware. The problem of the selection of the structure that will give good conditioned parameterization is also considered. A new procedure to identify the structure in the input-output form is presented. This procedure is suitable for both stationary and non-stationary systems when a change in the structure occurs while the order remains constant. It uses the concept of overlapping parameterization to choose a better conditioned parameterization for the multivariable system whenever ill conditioning is detected. A switching criterion is presented based on the complexity principle which provides a good monitor of the conditioning of the parameterization as well as the suitability of the tested structure.</p> / Doctor of Philosophy (PhD)

<p>A procedure for the recording and storing or EMG signals for the automatic extraction of individual motor unit rate coding and morphological information was developed. The extraction of the fate coding information is performed by the analysis of individual motor unit action potential trains (MUAPT'S), obtained by decomposing the composite EMG signal, recorded from a selective surface of a needle electrode. The morphological information is derived from the analysis of macro motor unit potentials (MUP'S). The macro MUP'S are the result of ensemble averaging the cannula recorded signal, using the individual MUAPT'S as sources of synchronizing triggers for the averaging process. A standard single fibre or macro EMG needle electrode can be used.</p> <p>Signals recorded during isometric, constant or slow force varying contractions, up to 50% of the maximum voluntary contraction level, can be successfully analyzed. The number of motor units simultaneously studied, usually ranges from 3 to 5. The processing of the data, for each second of muscle contraction, can be performed by a PDP 11/34 computer in approximately one minute, with 95% accurate, rate coding information obtained. This speed and accuracy may increase the clinical use of rate coding information. The procedure also provides individual motor unit rate coding and morphological information simultaneously, allowing correlation of motor unit sizes with firing rates and identification of macro MUP'S for high threshold motor units.</p> / Doctor of Philosophy (PhD)

<p>Several techniques used by researchers in the area of human locomotion to process and analyse normal and pathological gait electromyographs (EMG) are discussed. Basic elements of neuromuscular organization are described.</p> <p>The thesis reports original work in several topic. The spectral analysis of dynamic EMG acquired during the locomotion of a normal subject was done to confirm the selected sampling frequency, and to determine a suitable low pass filter cutoff for smoothing EMG prior to data analysis.</p> <p>Results of using two filters for smoothing EMG, a second order Butterworth low pass filter, and a mid-point moving window average filter are compared.</p> <p>The cross correlation function is used in analysing EMG, since EMG signals are random. The results of cross correlation are compared with clinical observations in assessing the state of a patient following a stroke. Results for five normal and fourteen hemiplegic subjects are reported.</p> <p>The conclusion is that cross correlation analysis quantifies the state of the patient and assesses post stroke recovery according to the neurological picture of central nervous system control.</p> / Master of Engineering (ME)

<p>This thesis addresses itself to the two principal problems of computer-aided-testing in analog circuits, namely, fault analysis and postproduction tuning.</p> <p>A unified approach to fault location in large analog circuits is introduced. The approach closely meets the goals of practical criteria for fault analysis. Network decomposition and logical analysis are incorporated to identify faulty subnetworks. Necessary and almost sufficient testing conditions for locating fault-free subnetworks are derived. These conditions are based on invoking KCL and topological relations between subnetworks. The application of the approach to practical linear and nonlinear networks is presented. Further fault analysis is carried out to identify faulty elements or regions inside the faulty subnetwork. Deterministic and approximate methods are introduced for that respect. The approximate method utilizes an estimation criterion, namely, the least-one objective function to predict the most likely faulty elements. The deterministic methods verify the existence of faults by examining the consistency of algebraic equations or by matching the subnetwork response using faulty models of the subnetwork elements. A number of network examples are considered to illustrate the application of the introduced methods.</p> <p>The deviation in the response of a manufactured circuit can often be compensated by adjusting specified tunable elements. A number of aspects of the postproduction tuning problem are studied. In particular, the relevant fundamental concepts and definitions are given, the tuning algorithms either functional or deterministic are reviewed and new techniques for choosing tunable parameters and critical response points are introduced. Two new functional tuning techniques are presented. The application of the new techniques in tuning a microwave network example is illustrated. A comparison and evaluation of four different tuning techniques are given by testing them in tuning an active filter example.</p> / Doctor of Philosophy (PhD)

<p>The objective of satellite orbit determination is to accurately estimate a set of orbital elements which describes the orbit of the satellite, using observations of the satellite. The extended Kalman filter has been extensively used for the estimation of the orbital states. The purpose of this work is to find alternative approaches that would reduce the amount of on-line computation required. A nonlinear estimator combining the invariant imbedding concept with stochastic approximation is proposed for this application. A switching criterion utilizing the properties of tile innovations sequence is applied to the combined estimator. Pugacev's estimation theory is also highlighted and the Kalman filter equations are derived as a special case of the general theory. Alternative approaches for forecasting the observables of the satellite via stochastic modeling techniques are proposed. One-step ahead forecasts are obtained using both univariate and multivariate time-series methods. Also, a recursive algorithm for estimating the degree of differencing most suitable for a given time-series is proposed. The results of simulation indicate the efficiency and reliability of the proposed schemes.</p> / Doctor of Philosophy (PhD)

<p>A very rapidly convergent solution (in the form of a likelihood ratio test) for the problem of detecting a discrete-time stochastic process in additive white Gaussian noise is derived.</p> <p>This likelihood ratio test is then applied to the problem of moving-target (aircraft) detection by airport surveillance radar systems. Using real radar data, the receiver operating characteristics are obtained for two different adaptive implementations of this likelihood ratio test, and also for the three versions of the Moving Target Detection algorithms presently in use in modern radar systems.</p> <p>The better of the two adaptive implementations employs Kalman prediction error tapped delay-line filters and attains a minimum of 3 dB average performance improvement relative to the Moving Target Detection algorithms.</p> / Doctor of Philosophy (PhD)

<p>An instrumentation package for sensing rainfall amount and intensity with fine time and space resolution is described. The package comprises a drop counter precipitation sensor, a microcomputer-based data acquisition system, and an intelligent data decoder. The accuracy of the precipitation sensor and the parameters that affect it are discussed. The reliability is reviewed and typical rainfall data are included. A comparison is made between the performance and accuracy of the new precipitation sensor and conventional tipping bucket raingauges. The merits and demerits of the new system are discussed.</p> / Doctor of Philosophy (PhD)