https://www.um.edu.mt/library/oar/handle/123456789/14678 Sun, 12 Apr 2026 03:18:39 GMT 2026-04-12T03:18:39Z https://www.um.edu.mt/library/oar/handle/123456789/101544 Title: Implementation of an automatic phoneme recognition system Abstract: Automatic Speech Recognition (ASR) is becoming increasingly more popularly used in most applications in today's technologies. Over the past years, a lot of research has been carried out, with the current trend being speech recognition in speaker-independent continuous speech environments with large vocabulary. However, most of the research carried out so far focused on software applications, and not much work has been carried out on the design of hardware recognisers, which is also necessary in order to attain further improvement in the field of ASR. Over the past years, ASR was not used widely in potential applications due to a number of limitations, such as processing power and limited hardware resources [1]. However, with today's advances in customised hardware, the design of ASR systems on-chip has become more feasible. In this research, different phoneme recognition systems for multi-speaker continuous speech environments are being proposed. The feature extraction stage was based on the Discrete Wavelet Transform (DWT), and for the classification stage, different Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs) were analysed. From the methods considered, the One-against-one SVM method provided the highest recognition rates. Furthermore, a priorities scheme was also added, so that the three most likely phoneme representations were obtained at the output. The software implementation of this phoneme recognition system has the potential to achieve an accuracy of 75.41%, for the recognition of 42 phoneme classes from the DARPA TIMIT Acoustic-Phonetic Continuous Speech Corpus (TIMIT) corpus. The results obtained were either comparable, or slightly better, than the best results found in the literature [2] - [7], which evaluated their systems on the TIMIT corpus. On average, the potential recognition rate which can be achieved with the proposed phoneme recognition system, results in an increase in accuracy of approximately 6%, when compared to the phoneme recognition systems presented in [2] - [7]. However, the proposed system is more adequate to be implemented on hardware. The phoneme recognition system was then designed on a dedicated chip, in order to evaluate its potential into becoming a portable and efficient system which can be employed in battery-powered devices. The final design can provide a speed which is approximately 4 times faster than the software-based approach, and consumes only 12.Sm W, making it appealing to mobile devices. The performance results obtained from the hardware design demonstrated that this system is a promising basis for future hardware ASR systems. Description: PH.D. Wed, 01 Jan 2014 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/101544 2014-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/101264 Title: A planar circuit model for the analysis of pixel microstrip antennas Abstract: A mobile handset needs to transmit/receive data on multiple frequency bands to access a diverse range of services. This implies both an increase in real estate for mobile terminal antennas as well as ever escalating challenges for the antenna design and manufacturing departments. The pixel patch antenna has been proposed as a possible solution. The pixel patch antenna is made up of a large number of pixels which by themselves do not resonate at any frequency of practical interest. However, when a set of selected pixels are connected/switched together, antennas of different shapes are synthesized to deliver the required electrical characteristics. Ideally these properties are optimized for a given operating environment in real-time. Given the antenna hardware architecture, the problem of synthesizing the various shapes can be casted as a search among many configurations, which is not a trivial problem because of the large search space defined by the pixel array. Early research has focused mostly on the physical construction and architecture of the pixel patch antenna and it is only recently that attention has increased on the search algorithm. A series of experiments using shorting ports, switches and varactor diodes are carried out as part of the literature review, to study tuning techniques. As a result of this study, novel architectures are proposed. The model based search algorithm is reviewed and its performance is compared to the Genetic Algorithm with and without pruning methods. This thesis builds upon the idea of making use of an antenna model in the search and optimization loop. To this end the known planar circuit model, augmented with segmentation and desegmentation methods, is further developed to yield an efficient model that is used to analyse a given arbitrary pixel antenna configuration. In the model's formulation, the Green's function approach is chosen to analyse the planar circuit. The Green's function is available for regular shapes only, and for arbitrary shaped structures, segmentation techniques are used to connect regular shapes together. The Generalized Cavity Model is used to model radiation and the fields beneath the patch. In order to realize the model several studies and novel methods had to be developed. An analysis of the behaviour of the Green's function with respect to the number of modes needed for convergence was carried out. It was determined that that the number of modes needed is a function of frequency and the permittivity of the substrate Er, at which the Green's function is evaluated. It was also found that the maximum number of modes is needed when the value of effective loss tangent is zero. These studies are used to dynamically estimate the number of modes needed per segment. A study is performed on the errors generated by multiple applications of the segmentation technique. The results from this study indicate that the number of segments used to represent the pixel patch antenna should be minimized. This is implemented in the Model Order Reduction algorithm. The relationships between the pixels are represented in a tree and a graph. This tree is traversed in order to build the order of segmentation. The use of these data structures allows the analysis of any arbitrary pixel patch antenna. A novel method to calculate the current passing through a number of n-port networks is developed and described. This method is more computationally efficient than the method used in the original segmentation method. The graph is used to determine the edges of the segments that will be used in the segmentation algorithm. A set of solvers are developed and their application in terms of efficiency are studied. In particular an algorithm that evaluates the input impedance of the antenna only in a set of intervals and interpolates between these points to find the minimum value of the input reflection coefficient is developed. The fields are only evaluated once the resonant frequency is found. The results obtained from this model are compared to CST microwave studio simulations. A set of pixel patch antennas configured in regular shapes, and a set of arbitrary shaped pixel patch configurations are used in the evaluation. The planar circuit model yields accurate results for the resonant frequency while it gives a good estimate of the impedance matching. The different variants of the model are considered to study the trade-off between accuracy and computational efficiency. Finally, developments to the model that potentially improve computational efficiency are suggested and discussed. Description: PH.D. Wed, 01 Jan 2014 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/101264 2014-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/100906 Title: Optimisation of the field description for the Large Hadron Collider during beam commissioning Abstract: The Large Hadron Collider (LHC) at CERN is a 27km circumference particle accelerator. It is made out of 1232 superconducting dipole magnets and 392 superconducting quadrupole magnets, which guide the particles around the LHC. It is designed to have two counter-rotating beams of particles collided at nominal centre of mass energy of 14 TeV to study and explore the fundamental forces and constituents of matter. The operation of the LHC is a challenging task. All the superconducting magnets have to be controlled simultaneously in order to successfully steer the beams around the LHC. Any field variations of the superconducting magnets, which are not immediately corrected, will affect the trajectories of the beam and therefore the LHC performance. Unfortunately, a feedback control system based solely on beam-based measurements is not enough and therefore, a feed-forward control system is also required. The feed-forward control system is used to predict the behaviour of the field variations such that it reduces the load from the feedback system. The Field Description for the Large Hadron Collider (FiDeL) is the feed-forward control system designed to predict the magnetic field variations that result from the inherent properties of the superconducting magnets. This system is based on the analysis of the data obtained from the magnetic measurements of all the superconducting magnets before the magnets were installed in the LHC. This thesis is focused on determining the precision of the FiDeL model and optimising it where necessary. However, in this work, the magnetic behaviour is not studied through magnetic measurements of individual magnets, but through beam-based measurements performed during the operation of the LHC. This is done by measuring the tune and the chromaticity of the LHC, two important beam parameters which ensure beam stability. These two parameters are directly dependent on the magnetic field quality and the control of its harmonic content. Therefore, by analysing such beam-based measurements, the behaviour of the FiDeL model is compared to the behaviour of the whole LHC. Furthermore, this study also gives an outlook of the FiDeL model at the LHC nominal operation of 7 TeV. This consists of studying the main elements of FiDeL that can become critical at such operation in order to anticipate any possible issues and how these can be overcome. The work presented in this thesis has been adopted by CERN during the first LHC running period (2008-2013). It will continue to be used in 2015, following the two- year shutdown after which the LHC will be operating at the nominal design parameters. Description: PH.D. Wed, 01 Jan 2014 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/100906 2014-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/95973 Title: Services and engines for a recommender system Abstract: The number of ecommerce websites has grown rapidly over the past few years. Most of these websites use a back end database and support clients by providing ordinary database query facilities. However, using this approach has the major drawback that searching for a required product becomes difficult: Often users are faced with situations were the database either returns too many results or none at all. Also these websites, regardless of the fact they are used by a large number of users, they rarely take advantage of its population. Recommender systems can help dealing with the mass of products that are available in the databases of online organizations. They take advantage of the users of a web site and exploit personalization technologies to recommend items of interest to what other have liked. To overcome the problems of information overload and help customers receive better suggestions, a Framework is being proposed that identifies a common set of services which software developers should find helpful when implementing a recommender system, and implement these services through engines which are powered by knowledge based technologies. The framework should then be used by developers in order to create recommender systems in a shorter time. Description: B.Sc. IT (Hons)(Melit.) Thu, 01 Jan 2004 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/95973 2004-01-01T00:00:00Z