This thesis presents an innovative approach to knowledge
management (KM) from the perspective of embedded system (ES)
development, a form of development that is highly knowledge intensive
and depends on specialised forms of knowledge obtained from a variety
of complex knowledge artefacts. This study follows an experimental
methodology that involves integrating a knowledge management system
(KMS) into ES product prototyping projects, in order to facilitate KM
of a specific form of knowledge, namely embedded system artefact
organisation and adaptation (ESAOA) knowledge. ESAOA knowledge is
produced during ESAOA activities, which concern organising artefacts
that are used to construct an ES and techniques by which engineers
adapt and learn from these artefacts. The focus of this thesis is
narrowed to determining an effective structure for the ESAOA KMS that
facilitates successful completion of ES implementation tasks. This
thesis consequently contributes to KM research at a 'meso level' of
The research methodology involved constructing an experimental KMS, named the ESAOA KMS, which comprises a structured collection of knowledge worker roles, processes, and artefacts together with a collection of support tools. A pilot study was first performed to gain insights into research methods and the KM needs of the users. Findings from this research included the following: defining different forms of ESAOA knowledge; establishing evaluation methods for KM of ESAOA activities; identifying conditions that enable a KMS to facilitate ESAOA activities; assessing the factors that affect ESAOA KM activities; determining different types of KM needs that occurred in projects, and showing that the ESAOA workspace approach was an effective means to integrate the knowledge worker roles, processes, artefacts and support tools of the ESAOA KMS. The conclusion of this thesis identifies situations in which the ESAOA KMS was found to be beneficial, as well as conditions where the KMS was of little use or possibly added to the difficulty of completing ESAOA activities.
Back to list of
Group's PhD Theses
This thesis presents the design and implementation of a signal level simulator supporting a wide variety of radar systems, and focusing on multistatic and netted radars. The simulator places few limits on the simulated system, and supports systems with arbitrary numbers of receivers, transmitters, and scatterers. Similarly, the simulator places no restrictions on the radar waveform to be simulated, and supports pulsed, continuous wave (CW) and carrier-free radar systems.
A flexible model is used to describe the radar system to be simulated, with the parameters of the radar hardware, the properties of scatterers and the layout of objects in the simulated environment specified in XML format. The development of the simulation model focused on balancing the requirements of flexibility and usability, ensuring that the model can be efficiently used to represent any type of radar system.
Oscillator phase noise is a limiting factor on the performance of some types of radar systems. The development of a model for the deterministic and static components of phase noise is presented. Based on this model, an algorithm for the efficient generation of synthetic phase noise sequences was developed, based on a multirate signal processing approach. This thesis presents this algorithm, and results of simulations of the effects of phase noise on synthetic aperture radar (SAR) and pulse-Doppler radar systems.
The FERS simulator, an implementation of the simulation model presented in this thesis, was developed in the C++ and Python programming languages. This simulator is able to perform real-time simulation of some common radar configurations on commodity PC hardware, taking advantage of multicore and multiprocessor machines. FERS has been released as open source software under the GNU general public licence (GPL).
Validation of the simulator output was performed by comparison of simulation results with both theory and measurements. The simulator output was found to be accurate for a wide variety of radar systems, including netted pulse-Doppler, moving target indication (MTI) and synthetic aperture (SAR) radar systems.
Back to list of Group's PhD Theses
Before embarking on any mining operation, it is advantageous to locate the subsurface orebody in three-dimensions with a high resolution (~1m) ahead of actual mining, because this will increase productivity and efficiency. Borehole radar is such an emerging mapping tool in the mining industry, that can be used to image the subsurface orebody with high resolution.
3-D subsurface imaging using two-dimensional aperture synthesis requires many boreholes and is thus not economically feasible in underground orebody imaging. Therefore, the main objective of this thesis is to develop 3-D imaging techniques, using a limited number of boreholes.
An interferometric synthetic aperture radar (InSAR) is a well established space-borne/airborne technique for mapping the Earth’s surface. A borehole InSAR simulation study was thus carried out, using a sidelooking antenna configuration to image the subsurface orebodies, such as potholes and cylindrical Kimberlite structures, in 3-D. A performance analysis of an interferometric experiment is presented.
In general, borehole radars are a wide-band system, often having bandwidths of 75% of the centre frequency. A 3-D image reconstruction technique was developed by means of correlation-type processing, using magnitude images of multiple boreholes coming from different view angles, which is more suitable for wide-band/ultra wide-band borehole radar signals. The technique was tested by using simulated multiple borehole radar magnitude images, as well as real acoustic images that had been captured in air and water media. The 3-D reconstructed grid spacing derivations are presented for a borehole trajectory fanning outward from the borehole centre. In this thesis, a 40kHz air-based sonar system was used in the laboratory environment to emulate inverse synthetic aperture radar (ISAR) data in the context of a real borehole experiment. A deconvolution processing approach was adopted to range-compress the 40kHz sonar data captured in air. A time domain focusing technique was used to focus the simulated borehole radar as well as the real acoustic data.
In the case of homogeneous, isotropic and non-dispersive media, a straight-line wave propagation is considered to process the data in range and azimuth. In a real bore hole environment, however, an accurate electromagnetic (EM) wave propagation model is essential. For the purpose of modelling borehole EM propagation in a conductive medium, 3-D finite difference time domain (FDTD) code was written and implemented in a Cartesian coordinate system by using a uniaxial perfectly matched layer boundary wave absorber. The accuracy of the implemented code was first tested against published results. Thereafter, the code was used to simulate the EM responses from various geological settings, such as cross-well borehole EM wave propagation in a sedimentary layer, reflection from a geological reverse fault, and reflection from a pothole-type orebody structure. Different kinds of imaging modes have been used in the FDTD simulation experiment, such as common offsets, common source and transillumination mode. The radar traces, both transmitted as well as reflected, are affected by the size of the borehole and the electrical properties of borehole mud. It was found that the electrical properties of the borehole mud affect the radar traces more significantly than the size of the borehole. The effect of host rock conductivity on radar traces has also been investigated.
To ensure the accuracy and stability of the FDTD method, the discrete step size of the simulation needs to be set to less than 1 10 th of the minimum significant wavelength. Therefore, for realistic 3-D simulations, there is a requirement for large matrices to be allocated and processed, which easily exceed the limits of a standard desktop PC in terms of memory and speed. In order to overcome these limitations, a parallel version of the 3-D FDTD C code has been implemented using Parallel Virtual Machine (PVM) as middleware running on a Beowulf-type Linux cluster. A speed up of 2.7 was achieved, which corresponds to a 90% efficiency, where a speed of 3 for three slave processors is considered to be 100% efficient.
The signal processing techniques investigated in this thesis have been verified on simulated borehole radar data, as well as on real sonar data. These techniques can therefore also be applied to real borehole data, in order to construct 3-D images of subsurface orebodies.
Back to list of Group's PhD Theses
A Ground Penetrating Radar (GPR) sensor is required to provide information that will allow the user to detect, classify and identify the target. This is an extremely tough requirement, especially when one considers the limited amount of information provided by most GPRs to accomplish this task. One way of increasing this information is to capture the complete scattering matrix of the received radar waveform. The objective of this thesis is to develop a signal processing technique to extract polarimetric feature vectors from Stepped Frequency Continuous Wave (SFCW) GPR data. This was achieved by first developing an algorithm to extract the parameters from single polarization SFCW GPR data and then extending this algorithm to extract target features from fully polarimetic data.
A model is required to enable the extraction of target parameters from raw radar data. A single polarization SFCW GPR model is developed based on the radar geometry and linear approximations to the wavenumber in a lossy medium. Assuming high operating frequencies and/or low conductive losses, the model is shown to be equivalent to the exponential model found in signal processing theory. A number of algorithms exist to extract the required target parameters from the measured data in a least squared sense. In this thesis, the Matrix Pencil-of-Function Method is used. Numerical simulations are presented to show the performance of this algorithm for increasing model error. Simulations are also provided to compare the standard Inverse Discrete Fourier Transform (IDFT) with the algorithm presented in this thesis. The processing is applied to two sets of measured radar data using the radar developed in the thesis. The technique was able to locate the position of the scatterers for both sets of data, thus demonstrating the success of the algorithm on practical measurements.
The single polarization model is extended to a fully polarimetric SFCW GPR model. The model is shown to relate to the multi-dimensional exponential signal proecssing model, given certain assumptions about the target scattering damping factor. The multi-snapshot Matrix Pencil-of-Function Method is used to extract the scattering martix parameters from the raw polarimetric stepped frequency data. Those Huynen target parameters that are independent of the properties of the medium, are extracted from the estimated scattering matrices. Simulations are performed to examine the performance of the algorithm for increasing conductive and dielectric losses. The algorithm is also applied to measured data for a number of targets buried a few centimetres below the ground surface, with promising results.
Finally, the thesis describes the design and development of a low cost, compact and low power SFCW GPR system. It addresses both the philosophy as well as the technology that was used to develop a 200-1600 MHz and a 1-2 GHz system. The system is built around a dual synthesizer heterodyne architecture with a single intermediate frequency stage and a novel coherent demodulator system - with a single reference source. Comparison of the radar system with a commercial impulse system shows that the results are of a similar quality. Further measurements demonstrate the radar performance for different field test cases, including the mapping of the bottom of an outdoor testsite down to 1.6m.
Back to list of Group's PhD Theses
Ultra-wideband synthetic aperture radar (SAR) systems operating in the VHF/UHF region are becoming increasingly popular because of their growing number of applications in the areas of foliage penetration radar (FOPEN) and ground-penetrating radar (GPR). The objective of this thesis is to investigate the following two aspects of low-frequency (VHF/UHF-band) SAR processing:
1. The use of stepped-frequency waveforms to increase the total radar bandwidth, thereby increasing the range resolution, and
2. Radio frequency interference (RFI) suppression.
A stepped-frequency system owes its wide bandwidth to the transmission of a group of narrow-bandwidth pulses, which are then combined using a signal processing technique to achieve the wide bandwidth. Apart from providing an economically viable path for the upgrading of an existing single frequency system, stepped-frequency waveforms also offer opportunities for RFI suppression.
This thesis describes three methods to process stepped-frequency waveforms, namely an IFFTmethod, a time-domainmethod and a frequency-domainmethod. Both the IFFT method and the time-domain method have been found to be unsuitable for SAR processing applications. The IFFT method produces multiple “ghost targets” in the high resolution range profile due to the spill-over effect of energy into consecutive coarse range bins, and the time-domain technique is computationally inefficient on account of the upsampling requirement of the narrow-bandwidth pulses prior to the frequency shift. The frequency-domain technique, however, efficiently uses all the information in the narrowband pulses to obtain high-resolution range profiles which do not contain any “ghost targets”, and is therefore well suited for SAR processing applications. This technique involves the reconstruction of a wider portion of the target’s reflectivity spectrum by combining the individual spectra of the transmitted narrow-bandwidth pulses in the frequency domain. It is shown here how this method may be used to avoid spectral regions that are heavily contaminated with RFI, thereby alleviating the problem of receiver saturation due to RFI. Stepped-frequency waveforms also enable the A/D converter to sample the received narrow-bandwidth waveform with a larger number of bits, which increases the receiver dynamic range, thereby further alleviating the problem of receiver saturation during the presence of RFI.
In addition to using stepped-frequency waveforms for RFI suppression, a number of other techniques have been investigated to suppress RFI. Of these, the notchfilter and the LMS adaptive filter have been implemented and applied on real P-band data obtained from the E-SAR system of the German Aerospace Center (DLR), Oberpfaffenhofen, and on real VHF-band data obtained from the South African SAR (SASAR) system. Both methods significantly suppressed the RFI in the real images investigated.
It was found that the number of range lines upon which the LMS adaptive filter could operate without adaptively changing the filter tap weights was often well above 100. This facilitated the re-writing of the LMS adaptive filter in terms of an equivalent transfer function, which was then integrated with the range-compression stage of the range-Doppler SAR processing algorithm. Since the range-compression and the interference suppression could then be performed simultaneously, large computational savings were achieved.
A technique was derived for suppressing the sidelobes which arise as a result of the interference suppression of the LMS adaptive filter. This method was also integrated with the range-compression stage of the range-Doppler processor, leading to a very efficient implementation of the entire RFI suppression routine.
Back to list of Group's PhD Theses
This thesis focuses on the application of range-Doppler processing at VHF frequencies for wide azimuth beam, airborne, stripmap SAR systems with zero Doppler centroid and modest pulse bandwidths relative to the carrier frequency (simulations less than 30 percent). Such a system is the South African SAR (SASAR) VHF sensor. The theory of such SAR operation is addressed.
In general, closed form analytical expressions for range compressed, range-Doppler domain signals do not exist. Thus, in this work, extensive use is made of simulation. Using simulated SAR signals with severe range curvature, the regions of applicability of standard range-Doppler processing, when applied without Doppler frequency dependent secondary range compression, are investigated for a range of processing parameters in the frequency range of 100 MHz to 200 MHz. The effects on the range impulse response of centre frequency, target closest approach range and nominal range resolution are investigated, each for a range of processed azimuth resolutions. Information is presented in the form of plots showing the degradation in the range resolution and in the form of tabular results, which also include the range peak- and integrated sidelobe levels and the non-linear phase error in the Fourier domain.
An extension to range-Doppler processing, suggested to the candidate by Michael Jin, is demonstrated to provide significantly improved performance over the standard range-Doppler processor for signals with severe range curvature. The basic idea of the extended algorithm, first published by Raney and Vachon in 1989 ad applied in the context of a narrow beam, squinted SAR, is to make an initial correction to a reference range through a multiplication with a reference function in the 2-D frequency domain. Airborne motion compensation strategies for flight path reconstruction are discussed.
In addition, an ERIM-developed approach for efficiently including an azimuth dependence for wide beam motion compensation is discussed in the context of the SASAR system. An overview of the SASAR VHF project, the experience gained, and the encouraging results from the processing of the data from the first radiating flights, is presented. A full Statement of Originality is given.
Back to list of Group's PhD Theses
This page was last updated in January 2007 (RL)