Search

This thesis presents the work undertaken while carrying out research into acoustic measurement techniques for 3-dimensional acoustic radiation data for musical instruments, specifically when the instruments are being played by musicians. The original contribution to knowledge that is presented includes the development of an algorithm which can be used for post-processing of recorded data to obtain signals from 'virtual' microphones. The project is discussed along with a rationale for the particular test and measurement procedure used in this research and is followed by a literature review outlining both historical and current research and writing relevant to the project. A design for high spatial resolution 3-D acoustic measurement apparatus is proposed, and the design details and construction methods are discussed. The measurement process is described, including the issues surrounding testing and the use of human musicians in the measurement of musical instrument acoustic radiation patterns. A novel algorithm is presented which applies transfer functions derived from interpolated measured data points in order to process recorded audio signals with applications in audio post-production. A prototype implementation of the algorithm is described along with its testing. The conclusion summarises the thesis; contains an evaluation of the work undertaken and the results; and explores potential future work from this project.

Recent advancements in vehicle automation have led to a proliferation of studies in traffic control strategies for the next generation of land vehicles. Current traffic signal based intersection control methods have significant limitations on dealing with rapidly evolving mobility, connectivity and social challenges. Figures for Europe over the period 2007-16 show that 20% of road accidents that have fatalities occur at intersections. Connected and Automated Mobility (CAM) presents a new paradigm for the integration of radically different traffic control methods into cities and towns for increased travel time efficiency and safety. Vehicle-to-Everything (V2X) connectivity between Intelligent Transportation System (ITS) users will make a significant contribution to transforming the current signalised traffic control systems into a more cooperative and reactive control system. This research work proposes a disruptive unsignalised traffic control method using a Reinforcement Learning (RL) algorithm to determine vehicle priorities at intersections and to schedule their crossing with the objectives of reducing congestion and increasing safety. Unlike heuristic rule-based methods, RL agents can learn the complex non-linear relationship between the elements that play a key role in traffic flow, from which an optimal control policy can be obtained. This work also focuses on the data requirements that inform Vehicle-to-Infrastructure (V2I) communication needs of such a system. The proposed traffic control method has been validated on a state-of-the-art simulation tool and a comparison of results with a traditional signalised control method indicated an up to 84% and 41% improvement in terms of reducing vehicle delay times and reducing fuel consumption respectively. In addition to computer simulations, practical experiments have also been conducted on a scaled road network with a single intersection and multiple scaled Connected and Automated Vehicles (CAV) to further validate the proposed control system in a representative but cost-effective setup. A strong correlation has been found between the computer simulation and practical experiment results. The outcome of this research work provides important insights into enabling cooperation between vehicles and traffic infrastructure via V2I communications, and integration of RL algorithms into a safety-critical control system.

The research work described in this thesis was concerned with finding a novel method of minimising motion artefacts in biomedical instrumentation systems. The proposed solution, an Analog Frontend (AFE), was designed to detect any vertical (Y-Plane) or horizontal (X-Plane) movement of the electrode using two strain gauges, which were separated by 90° and fitted onto the electrode. The detected motion was fed back to the system for the removal of any motion artefact. The research started by emphasising the importance of minimising motion artefacts from biomedical signals and explaining how important it is for a clinical misinterpretation of the results. Hence, various motion artefact minimisation techniques undertaken by other researchers in the field were reviewed. This study covered different sources of artefacts, including the 40kHz powerline interference (PLI), 50/60kHz common-mode noise, white noise, and motion artefacts. The system was fully developed and tested and was firstly simulated using MATLAB Simulink tools to prove the effectiveness of the system before starting the implementation and build phase in the lab. The AFE system successfully produced a clean output signal, achieving an average correlation coefficient of 0.995. Also, the system output had a 98% SNR similarity with the clean source signal. Further, the system was then built and tested in the lab and successfully minimised the motion artefacts, achieving an average correlation coefficient of 0.974. Additionally, the final output had a 97.8% SNR similarity with the clean source signal. A novel test rig was developed to test the system with strain gauges. The system was able to remove the detected signal from the test rig and had an average correlation coefficient of 0.957. Lastly, the final output had a 94.2% SNR similarity with the clean source signal.

The research discussed here aims to design a deep learning algorithm based on Convolutional Neural Networks models to detect driver distraction and fatigue using driver facial expressions. The proposed model provides high accuracy during both training and validation. The research was inspired to contribute to transport safety by providing alternative solutions to detect driver habit. First, the thesis discussed Conventional methods, including Haar cascade classifiers and eigenfaces. In 2018 I published a proposal for a blink rate detection system using Haar cascade feature detection. However, due to the advantages of Neural Networks, the research focused on providing a unique solution in that field. An in-depth look at how Neural Networks function, specifically Convolutional Neural Networks (CNNs), was investigated and discussed next. Due to the advantages CNN's have with feature detection in images, the algorithm I proposed in this research uses a CNN architecture. Lastly, I proposed an adaptive approach for deep learning to enhance training, validation and testing accuracies. My original algorithm and subsequent models were trained on two datasets. These were the American University in Cairo (AUC) Distracted Driver Dataset and the UTA Real-Life Drowsiness Dataset (UTA-RLDD). Hence the research proposed two original CNN models that produced high training and validation accuracy. The model designed on the AUC Distracted Driver Dataset achieved good 97% training accuracy and good 96% validation accuracy. Evaluation of this model produced good 99% accuracy. The model designed on UTA-RLDD achieved 100% training accuracy, 69% validation accuracy, and evaluated at 69% accuracy.

This thesis is concerned with the design of a polyhedron enumeration algorithm. The approach taken focuses on specic classes of polyhedra and their graph theoretic properties. This is then compared more broadly to other graph enumeration algorithms that are concerned with the same or a superset which includes these properties. An original and novel algorithm is contributed to this area. The approach taken divides the problem into prescribed vertex and face degree sequences for the graphs. Using a range of existence, ordered enumeration and isomorphism techniques, it finds all unique 4-regular, 3-connected planar graphs. The algorithm is a vertex addition algorithm which means that each result output at a given stage has a new vertex added. Other results from different stages are never required for further computation and comparison, hence the process is embarrassingly parallel. Therefore, the enumeration can be distributed optimally across a cluster of computers. This work has led to a successfully implemented algorithm which takes a different approach to its treatment of the class of 4-regular, 3-connected planar graphs. As such this has led to observations and theory about other classes of graphs and graph embeddings which relate to this research.

Where robots interact directly with humans on a 'one-to-one' basis, it is often quite important for them to be emotionally acceptable, hence the growing interesting in humanoid robots. In some applications it is important that these robots do not just resemble a human being in appearance, but also move like a human being too, to make them emotionally acceptable - hence the interest in biomimetic humanoid robotics. The research described in this thesis is concerned with the design, analysis and evaluation of a biomimetic humanoid robotic head. It is biomimetic in terms of physical design - which is based around a simulated cervical spine, and actuation, which is achieved using pneumatic air muscles (PAMS). The primary purpose of the research, however, and the main original contribution, was to create a humanoid robotic head capable of mimicking complex non-purely rotational human head movements. These include a sliding front-to-back, lateral movement, and a sliding, side-to-side lateral movement. A number of different approaches were considered and evaluated, before finalising the design. As there are no generally accepted metrics in the literature regarding the full range of human head movements, the best benchmarks for comparison are the angular ranges and speeds of humans in terms on pitch (nod), roll (tilt) and yaw (rotate) were used for comparison, and these they were considered desired ranges for the robot. These measured up well in comparison in terms of angular speed and some aspects of range of human necks. Additionally, the lateral movements were measured during the nod, tilt and rotate movements, and established the ability of the robot to perform the complex lateral movements seen in humans, thus proving the benefits of the cervical spine approach. Finally, the emotional acceptance of the robot movements was evaluated against another (commercially made) robot and a human. This was a blind test, in that the (human) evaluators had no way of knowing whether they were evaluation a human or a robot. The tests demonstrated that on scales of Fake/Natural, Machinelike/Humanlike and Unconcsious/Conscious the robot the robot scored similarly to the human.

Electrical Impedance Tomography (EIT) is a non-invasive, portable and low-cost medical imaging technique. Different current patterns are injected to the surface of a conductive body and the corresponding voltages are measured also on the boundary. These mea-surements are the data used to infer the interior conductivity distribution of the object. However, it is well known that the reconstruction process is extremely ill-posed due to the low sensitivity of the boundary voltages to changes in the interior conductivity distribution. The reconstructed images also suffer from poor spatial resolution. In tomographic systems, the spatial resolution is related to the number of applied current patterns and to the number and positions of electrodes which are placed at the surface of the object under examination. Two mammographic sensors were recently developed at the University of Mainz in collaboration with Oxford Brookes University. These prototypes consist of a planar sensing head of circular geometry with twelve large outer (active) electrodes arranged on a ring of radius 4.4cm where the external currents are injected and a set of, respectively thirty six and fifty four point-like high-impedance inner (passive) electrodes arranged in a hexagonal pattern where the induced voltages are measured. Two 2D reconstruction methods were proposed for these devices, one based on resistor network models and another one which uses an integral equation formulation. The novelty of the device and hence of these imaging techniques consists exactly in the distinct use of active and passive electrodes. The 2D images of the conductivity distribution of the interior tissue of the breast provide only information about the existence and location of the tumour. In this thesis different circular designs for the sensing head of this EIT device were analysed. The 2D resistor network approach was adapted to the different data collection geometries and the sensitivity of the reconstructions with respect to errors in the simulate data were investigated before any modifications to the original design were made. A novel 3D reconstruction algorithm was also developed for a simpler geometry of the sensing head which consisted of a rectangular array of thirty six electrodes (twenty active+ sixteen passive). This electrode configuration as well as the proposed imaging technique are intended to be used for breast cancer detection. The algorithm is based on linearizing the conductivity about a constant value and allows real-time reconstructions. The perfor-mance of the algorithm was tested on numerically simulated data and small inclusions with conductivities three or four times the background lying beneath the data collection surface were successfully detected. The results were fairly stable with respect to the noise level in the data and displayed very good spatial resolution in the plane of electrodes.

The goal of this research was to develop an algorithmic system capable of predicting the directional trend of the S&P 500 financial index. The approach I have taken was inspired by the biology of the human retina. Extensive research has been published attempting to predict different financial markets using historical data, testing on an in-sample and trend basis with many employing sophisticated mathematical techniques. In reviewing and evaluating these in-sample methodologies, it became evident that this approach was unable to achieve sufficiently reliable prediction performance for commercial exploitation. For these reasons, I moved to an out-of-sample strategy and am able to predict tomorrow’s (t+1) directional trend of the S&P 500 at 55.1%. The key elements that underpin my bio-inspired out-of-sample system are: Identification of 51 financial market data (FMD) inputs, including other indices, currency pairs, swap rates, that affect the 500 component companies of the S&P 500. The use of an extensive historical data set, comprising the actual daily closing prices of the chosen 51 FMD inputs and S&P 500. The ability to compute this large data set in a time frame of less than 24 hours. The data set was fed into a linear regression algorithm to determine the predicted value of tomorrow’s (t+1) S&P 500 closing price. This process was initially carried out in MatLab which proved the concept of my approach, but (3) above was not met. In order to successfully meet the requirement of handling such a large data set to complete the prediction target on time, I decided to adopt a novel graphics processing unit (GPU) based computational architecture. Through extensive optimisation of my GPU engine, I was able to achieve a sufficient speed up of 150x to meet (3). In achieving my optimum directional trend of 55.1%, an extensive range of tests exploring a number of trade offs were carried out using an 8 year data set. The results I have obtained will form the basis of a commercial investment fund. It should be noted that my algorithm uses financial data of the past 60-days, and as such would not be able to predict rapid market changes such as a stock market crash.

The research described in this thesis is concerned with analysis and design of "HighTransconductance RF MOSFET Voltage-to-Current (V-I) Converters". Various V-I converter circuits published in the past have been reviewed by the author in order to understand the different techniques employed to improve transconductance (Gt), linear operating range and total harmonic distortion (THO). Throughout this research, the emphasis has been to improve the above mentioned parameters. All the V-I converter circuits reported have been simulated using PSPICE and the results compared with the values obtained by theoretical analysis. Some of the results of this work have been already reported by the author in the technical literature. (See Chapter 9, at the end of this thesis, where reference to two publications by the author is given.) It was essential to obtain accurate CMOS device parameters values, such as Early Voltage, transconductance parameter ratios!! (gm/gds), X (gmbl'gm) and inter-electrode capacitances, to facilitate the design the prQcess. This was achieved using an extensive set of simulations for the transistor operating under different bias conditions. Furthermore, a measurement technique, thought to be novel, for the direct determination of the transconductance ratios!! and X is proposed. In the next part of the work several types of current mirror are compared against the standard current mirrors, using analytical and simulation methods. Furthermore several MOSFET V-I converter designs were critically reviewed to understand the various existing techniques and their limitations. Two novel techniques, Drain-Source Feedback Circuits (DSFCs) and Drain-Gate Feedback Circuits (OGFCs) ere implemented with a new temperature-compensation scheme, designed to operate well in an industrial environment (-40°C - +8S°C). It is found that the best types of V -I converters were the DSFCs which, offer a more accurate value of Gt (3.386mS) and the THO less than -S7dB for a differential input operating range SOOm V at 1 GHz with a 3V total rail voltage. The OGFC circuits were also meet the initial design targets, the value of THO is less then -SOdB, and operating in the Giga hertz frequency range is possible. Preliminary investigation on future work shows promising results.

The aim of this research programme was to design and develop a novel CMOS current conveyor, to improve areas such as bandwidth, slew rate, gain, and Powe- Supply Reject Ratio (PSRR). The current conveyor can be used in low frequency applications such as LED drivers for mobile phones and televisions, and high frequency applications such as mixers for up/down converters used in anything from radios to mobile phones. The initial part of the research looked into improving the Power Supply Rejection Ration (PSRR) of the current follower (mirror) by increasing its output impedance. Several types of current mirror were compared using analytical and simulation methods, using a new generic low frequency transistor model which was used to highlight the differences in impedance between BJT and CMOS current mirrors. It was found that the best type of mirror was the regulated cascode current mirror which offered the largest value of output impedance when built from CMOS transistors. Work then moved onto the voltage follower. By initially using a typical CMOS source follower, it was found that the voltage gain suffered from low values transconductance, drain/source resistance, and a larger than expected value of source resistance, which was extracted from simulation and was found to be around 300- 350Q. The best design was a two stage un-buffered amplifier which offered the best Power Supply Rejection (PSRR) voltage gain and bandwidth. Several different types of current conveyor (CCII+) were simulated and the results were compared. It was found that the best types of current conveyor were the cascode type conveyors which offered a voltage gain error of less than 1%. The regulated cascode type current conveyor offered the highest figure of PSRR that of around 60dB. Finally the new cascode type current conveyors were used to build examples of current feedback operational amplifiers (CFOAs), and the cascode type CCIl+ offered a voltage gain error of less than I%, largest bandwidth and best P SRR.

The research described in this thesis is concerned with the analysis, design and development of a novel temperature insensitive Current Controlled Current Source (CCCS), in bipolar technology, in order to provide accurate amplification of a Proportional To Absolute Temperature (PTAT) reference current. The output current of the CCCS is intended for application as the bias current for a bipolar Low Noise Amplifier (LNA) in order to minimise gain variations with temperature across the industrial temperature range (-40·C to 8S·C). The thesis begins with an explanation of key parameters concerned with LNA design and a target specification is defined. In Chapter 2, a conventional LNA, with constant with temperature bias current, is developed following a methodical approach based on conventional techniques. This meets the previously defined specification at room temperature but exhibits large gain variations with changes in temperature. The analysis and simulation results of this conventional LNA serve as a benchmark for comparison with later designs. In order to minimise any gain variations with temperature of a bipolar amplifier it is well known that the applied bias current should be PT AT. Thus, a thorough analysis and comparative review of traditional and novel PTAT reference current generator circuits is conducted in Chapters 3 and 4. Based on these findings the PTAT generator exhibiting best performance in terms of output current accuracy and insensitivity to power supply variations is presented. However, this circuit cannot accurately produce large rnA level currents necessary for LNA bias applications so that sufficient linearity of the LNA is maintained. Thus, a need for some form of accurate CCCS or Voltage Controlled Current Source (VCCS), which should be temperature insensitive in order to preserve the desired temperature coefficient of the reference current/voltage, is highlighted. Traditional VCCS/CCCS designs are investigated in Chapter 5. Limitations of these approaches leads to the design and development ofa novel CCCS with built in PTAT reference. The presented CCCS utilises a new, previously unseen, architecture and has led to a patent application [1]. The author has reported the majority of this work in technical literature [2-4]. In Chapter 6, the output of the novel CCCS is adapted to include the conventional LNA circuit designed previously in Chapter 2. The results of the combined LNA and CCCS are compared with the conventional LNA. The combined LNA and CCCS offers a dramatic reduction in gain variation with temperature.