Please use this identifier to cite or link to this item:
|Title:||Optimal Processing of Impulse Radar Signals for Bridge Deck Inspection|
Wong, Kon Max
|Department:||Electrical and Computer Engineering|
|Keywords:||Electrical and Computer Engineering;Electrical and Computer Engineering|
|Abstract:||<p>Impulse radar possess some attractive features that make it particularly useful in probing objects that are buried, encased in other materials or structures. Testing has demonstrated its potential in the detection of deterioration in concrete bridge deck slabs that are covered with bituminous surfacing. In order to benefit fully from using impulse radar in the bridge deck inspection, however, it is necessary to take advantage of the progress in signal processing techniques. This thesis is an attempt to provide a comprehensive treatment of the optimal extraction of information from the reflected radar signals, as to determine the subsurface structure and condition of the bridge decks. Generally, it follows a statistical estimation approach to the problem.</p> <p>In this thesis, a parametric representation is derived ti approximate the radar-generated pulses for probing. The asphalt-covered bridge decks are regarded as a system stratified, lossless, and horizontally layered media with each layer being homogeneous and isotropic. The propagation of electromagnetic waves in such a multi-layered media system can be completely determined by a set of characteristic parameters of the media. Under such assumptions, the reflected radar signals may be well described by a delayed sum model which is specified by the characteristic parameters of the media.</p> <p>Based upon the parametric signal model, a maximum likelihood estimator is formulated to determine the parameters of reflected signals. Computer experiments show that the ML estimation is capable of resolving closely spaced returns in the received signal and producing very accurate parameter estimates. ML estimation of real radar signals reflected from a bridge deck is also performed with success. However, to carry out the ML estimation requires an explicit knowledge of the probability density function of received signal which may not always be available. Moreover, the search for the ML estimates usually involves a global, nonlinar optimization procedure which can be extremely costly in computation time.</p> <p>To overcome the difficulties with the ML method, a new eigenstructure-based (EB) method for parameter estimation is developed in this thesis. The implementation of the new estimation method requires only the autocovariance of the reflected signal, and it is more efficient in computation than the ML method. Computer simulation demonstrates that the EB method results in very satisfactory estimates at high SNR levels, but it becomes inaccurate when the SNR level is low or the radar returns in the recieved signal are very close spaced in time.</p> <p>The error performances of the parameter estimators under varous conditions are evaluated and compared via computer simulation. A detailed analysis of the Cramer-Rao Lower Bound on the estimation error is performed to gain an insight into how various factors affect the estimation performance.</p> <p>An alternative to the parameter estimation approach is predictive deconvolution which is developed on a nonparametric model of the reflected signals. In principle, it is an application of Wiener optimal filtering theory to the deconvolution problem. It is observed in Computer simulation that predicative deconvolution is able to resolve returns closely located in time. Its implementation is carried out by simply solving a set of linear, normal equations, and its operation involves only straightforward linear filtering which demands little computation. However, the performance of predictive deconvolution deteriorates quickly in the presence of even a moderate level of noise in the input signal. This weakness may severely restrict its usefulness in practice.</p>|
|Appears in Collections:||Open Access Dissertations and Theses|
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.