Theses and Dissertations (Electrical, Electronic and Computer Engineering)

Permanent URI for this collectionhttp://hdl.handle.net/2263/31930

Browse

Now showing 1 - 20 of 455

Current steering with sequential stimulation in cochlear implants
(University of Pretoria, 2017-12) Hanekom, J.J. (Johannes Jurgens); Roux, Johanie
Current steering has been proposed to increase place pitch resolution in cochlear implant (CI) users. Many studies have shown that a current steering effect can be achieved when simultaneous stimulation is used (Koch, Downing, Osberger, Litvak and Greco, 2007, Saoji and Litvak, 2010, Wu and Luo, 2013). Some literature has shown that a current steering effect can also be achieved when sequential stimulation is used (McDermott and McKay, 1994, Kwon and van den Honert, 2006, Swanson, 2008). Literature proposes different features that could underlie place pitch and consequently possibly also current steering effects (McDermott and McKay, 1994, Kwon and van den Honert, 2006, Swanson, 2008, Frijns, Kalkman, Vanpoucke, Bongers and Briaire, 2009, Macherey and Carlyon, 2012, Venter, 2015). The present study confirmed that a current steering effect can be achieved when sequential stimulation is used by using multi-dimensional scaling and statistical analysis in addition to the convention of using cumulative d' values to analyse pitch ranking results of current steering experiments. It was however observed that a current steering effect could only be achieved in listeners who were at least able to pitch rank the pitch of the two individual stimulating electrodes correctly according to expectation. The effect of different stimulation parameters on the pitch ranking ability of CI users during current steering experiments was investigated. Results showed that some parameters only had an effect on the pitch ranking performance of some listeners, while other stimulation parameters affected the results of all the listeners. Wider stimulation pulse widths, for example, led to improved pitch ranking results for some listeners. Most listeners benefited from wider electrode separation distances. Statistical analysis showed that there was a significant improvement in the pitch ranking performance of the listeners during experiments where the stimulation rate was the same as the rate indicated on the clinical MAP of the listener. Person-specific current distribution models were used to predict the cochlear position of different stimuli because of different features that could underlie place pitch, for each of the experiments for four of the listeners who participated in this study. The model predictions were related to the measured pitch ranking results using correlation and mutual information analysis. The results indicated that the current centroid at electrode level, the position of the peak current at the auditory nerve level (because of either an individual stimulating electrode or because of summed currents) and the centroid of neural activation could underlie place pitch. All these features except the position of the peak of the current distribution at the auditory nerve level because of each individual stimulating electrode could underlie current steering effects. Results showed that the centroid of the current distribution at the auditory nerve level probably does not underlie place pitch. Knowledge about the impact that different stimulation parameters have on the ability to achieve a current steering effect could result in more efficient implementation of current steering effects. Proper knowledge of which features underlie place pitch and current steering effects could be used to create models that can be used to predict the results of place pitch experiments.
Viability of cochlear travelling wave signal processing for cochlear implants
(University of Pretoria, 2016-02) Hanekom, J.J. (Johannes Jurgens); Schmidt, Larry
English: The travelling wave encodes acoustic information by stimulating the auditory nerve fibres. Understanding the travelling wave and its process is important for the development of cochlear implants speech processors. The development of a normal hearing auditory model, using a hydrodynamic model of the travelling wave to predict the nerve fibre spiking diagrams, marked the first stage of this study. This study then proceeded to look at the development of a travelling wave speech processing algorithm and model the electrical response due to the stimulation from the vocoder speech processor, and the travelling wave speech processor. The final stage was to predict whether temporal encoding occurred during cochlear implant stimulation for the vocoder speech processor and the travelling wave speech processor. The results showed that the travelling wave normal hearing model was able to predict the nerve fibre characteristics seen in measurements from literature. This showed that the mechanical encoding performed by the travelling wave is vital to the encoding of information in auditory nerve fibres. The travelling wave speech processor was able to encode temporal cues for pitch up to 1060 Hz, where the results for the vocoder speech processor showed the 300 Hz limit seen in other literature of phase locking. Mimicking the travelling wave in cochlear implant speech processors may potentially benefit the delivery of information to the auditory cortex for cochlear implant users. However, these results must be legitimised using animal models and psychoacoustic experiments.
Signal processing by octopus cells for acoustic and electrical hearing : a modelling study
(University of Pretoria, 2016-07) Hanekom, J.J. (Johannes Jurgens); Blignaut, Gertruida Elizabeth
A computational model of a single octopus cell as well as a population of octopus cells was developed. The models were used to investigate the ability of octopus cells to compensate for the travelling wave delay, remove jitter from the neural activity and encode pitch for normal hearing. Furthermore the response of octopus cells to cochlear implant (CI) stimulation with the ACE strategy was investigated to determine whether pitch can be extracted from CI stimulation in the same way as from acoustic stimulation. Their ability to extract the pulse rate from single-electrode stimulation was also investigated. The response of the octopus cells to single-electrode stimulation at different pulse rates was used to predict pulse rate difference limens, which were compared to psychoacoustic measurements found in literature. It was found that octopus cells are sensitive to the delay of synaptic inputs on their dendrites but are broadly tuned to this delay. By evaluating the jitter together with the travelling wave delay present in the activity of auditory nerve fibres (ANFs), it was determined that octopus cells may rather act as coincidence detectors, which extract common interspike intervals (ISIs) from many ANFs. The octopus cell model was found to encode the frequency of pure tones in their ISIs for pure tone acoustic stimulation. They were also found to encode the pitch of a vowel in their ISIs, which was the same as the fundamental frequency extracted from the vowel with a speech processing algorithm. The octopus cell model responded to the pulse rate of the CI stimulation and could therefore not extract the frequency of pure tones from CI stimulation in the same way as from acoustic stimulation. The entrainment of the modelled octopus cell population decreased when the pulse rate of a single electrode increased beyond 300 pps. Pulse rate difference limens were predicted from the standard deviation of the ISIs of the octopus cell population response to single electrode stimulation. The predicted difference limens were in the same range as measured values, which suggests that octopus cells may play a role in the measured perceptual limit at 300 pps. From the findings of this study it is suggested that CI stimulation strategies should be developed to encode pitch in the periodicity of their stimulation to enable octopus cells to extract pitch information from CI stimulation.
The analysis of distributed resources on a load sharing reticulation network
(University of Pretoria, 2023-12-01) Naidoo, Raj; vandermerwe.ca@gmail.com; Van der Merwe, Carel Aron
Traditional reticulation network designs are outdated, based on single value static yearly maximum demands, and do not consider the dynamic nature of load-side DR installations. The increasing presence of privately driven downstream renewable and storage system integration (supported by increasing energy costs, maturing of storage, PV, and inverter technology systems, and an unreliable external network supply) requires time-based analysis to advance beneficial, and mitigate detrimental, shared network parameter changes. Fundamental integration network impacts must be re evaluated for grid integration acceptability and a modernised design approach, dependent on the capacity, capability, implementation, load-to-generation balancing, and power management of symbiotic integrated load-side DR (DG and/or ES) systems. These initial performance factors were analysed by conducting time based impact studies. Key concepts and approaches to the integration of PV DG, BESSs, and the combined DR system were identified and modelled at increasing levels of power penetration and energy arbitrage within the main distinctive reticulation network load profile forms in a visualised time based impact analysis. By identifying individual DR operational parameters and limits, an optimal approach to DR utilisation and power control is defined. Variables include load profiles, load diversity, demands, load factors, PV DG and BESS parameters, system power control, voltage profiles, utilisation factors, reactive power requirements, and fault levels. The maximum levels of DR penetration were defined (creating an upper penetration limit) following the evaluation of DR network parameter impacts and forms the foundation of the power flow control algorithm governing PV DG and BESS operation for equipment synergy and the optimisation of integration advantages. The proposed power control enforces permanent load side maximum demand reductions by up to 32%, with additional energy arbitrage operation enabled during peak period demands. This is achieved by limiting bi-directional power flow internally and maximising the combined DR system capability, utilisation, and operational synergy. Intermittent PV DG is selected for generation support, while more controllable BESS operation is chosen for targeted demand reduction applications in a give-and-take interface across all seasonal changes. The time based analysis, integration methodology, DR penetration limits, and the developed power flow control algorithm provide an expectation baseline for future DR network integration studies, guidance for service agreement inclusions, and the modernisation of traditional network designs without the necessity of an external network smart grid system. This will encourage the integration of higher rated privately driven renewable and energy storage systems to enhance grid advancement for both external and load-side DR integrated networks.
Delay- and disruption-tolerant routing algorithms to support human activity on mars
(University of Pretoria, 2024) Palunčić, Filip; Maharaj, Bodhaswar Tikanath Jugpershad; jason.kamps@tuks.co.za; Kamps, Jason Jack
Deep-space activity is expected to increase rapidly in the coming decades. Most notably, crewed missions to Mars will take place. With humans venturing light minutes away from Earth for the first time, communication becomes challenging. Humans have specific communication needs that become difficult to support in deep space where large propagation delays, high error rates, and intermittent connections are prevalent. Delay- and disruption-tolerant networking (DTN) and the Bundle Protocol provide a reliable communication solution in such challenging environments. The overall performance of DTN protocols is highly dependent on their routing algorithms. With Mars being humanity’s next target in our exploration of the Solar System, this study deals with finding and examining the most suitable routing protocols in the context of Earth-Mars communication. Realistic scenarios of space missions are constructed to enable the comparison of various DTN routing algorithms in simulation. Routing algorithm performance is analysed, and an enhancement to Contact Graph Routing (CGR) is proposed to address a deficiency of the algorithm, improving routing performance in networks featuring parallel channels.
Bi-directional DC-DC converter with energy management and protection capabilities For LVDC grids
(University of Pretoria, 2024-06) Gitau, Michael Njoroge; u15209840@tuks.co.za; Doma, Anesu Emmanuel
The work outlines a framework for enhancing the efficacy of current LVDC microgrid protection techniques. Currently, the two most significant challenges are the detection and interruption of fault currents. The primary aim of a protection strategy is to maintain the dependability of a power system by selectively isolating the components that are responsible for the fault occurrence. Consequently, it is imperative to interrupt the fault current before it reaches the components' maximum ratings. A proposal has been put forth for the implementation of a bidirectional converter to verify the functionality of a "converter cascaded with an Impedance Source Circuit Breaker (ISCB)" system. Contemporary investigations on DC microgrids suggest that the converter and impedance source breaker integration is functional; however, these two pivotal components have been analyzed separately, with the presumption of effortless integration. The combination is expected to exhibit fault current interruption capabilities and function as an energy hub. The analysis and design of a converter operating in Average Current Mode control (ACM) and an ISCB are conducted as separate entities. This work presents a proposed methodology for validating protection features. The obtained simulated results provide confirmation of the successful interruption of the circuit and ripple reduction on the DC branch input current.
Energy-efficient adaptive data rate optimisation scheme for LoRaWAN
(University of Pretoria, 2023-11) Hancke, Gerhard P.; Abu-Mahfouz, A.M.I. (Adnan); rachel.kufakunesu@tuks.co.za; Kufakunesu, Rachel
The Internet of Things (IoT) is a paradigm that has revolutionised wireless network technologies worldwide owing to its low power consumption, low cost of deployment, long-range communication, and its ability to accommodate a substantial number of end devices (EDs). Many industries have adopted the use of IoT to improve decision-making, to function more efficiently, to automate various operations, and to minimise energy usage, thus helping in the reduction of greenhouse gas emissions. Among the IoT technologies, Low Power Wide Area Networks (LPWANs) perform a pivotal function in delivering efficient and scalable connectivity for this ever-expanding wireless communication system. One prominent LPWAN technology, Long Range Wide Area Network (LoRaWAN), has achieved widespread adoption attributable to its operation in the license-free sub-gigahertz frequency band, scalability, low bit rate, and energy efficiency. LoRaWAN networks consist of low-power end devices connected in a star-of-star topology to a network server (NS) through a gateway (GW), enabling seamless exchange of data connected through an IP-based back-haul connection. The EDs are energy-constrained as they are battery operated and can be static or mobile and can sometimes be deployed in difficult-to-reach places, hostile or hazardous environments, necessitating a long battery life lasting several years. Wireless communication is the major source of energy consumption due to interference and packet collisions during packet transmission. It is essential to limit energy utilisation while maintaining the communication between the end devices and the gateway. This study focuses on enhancing energy efficiency in LoRaWAN networks to extend their network lifetime. An essential aspect of LoRaWAN is the Adaptive Data Rate (ADR) scheme, which optimises Quality of Service (QoS) requirements, battery life of EDs, and overall network performance. ADR manages transmission settings such as transmission power (TP), spreading factor (SF), bandwidth (BW), and coding rate (CR) based on the link budget. By optimising ADR, we can reduce airtime, increase network capacity, and improve energy efficiency. Following the introduction of the LoRaWAN protocol in 2015 by the LoRa Alliance, there has been an absence of a standardised ADR implementation in the LoRa specification. It does not define the way the network server controls EDs regarding data rate adaptation. This has led to various ADR schemes being proposed by different vendors to accommodate diverse IoT scenarios and QoS requirements, posing challenges to reliability and suitability. A comprehensive literature review of existing ADR schemes highlighted their focus on scalability, throughput, and energy efficiency. The existing ADR schemes use different algorithms with different computational complexities to optimise the data rate, depending on the different goals such as coverage, received signal strength, congestion, capture effect and channel contention. The issue of energy consumption emerged as a major challenge. To address this challenge, this research study proposed and implemented a novel fuzzy-logic-based adaptive data rate (FL-ADR) scheme for energy-efficient LoRaWAN communication. The impact of multiple GWs on a LoRaWAN network was investigated and the results were incorporated in implementing the LoRaWAN network using a ns-3 based LoRaWAN simulator, a widely used open source simulation platform. The proposed FL-ADR scheme performance was contrasted with other state-of-the-art algorithms. Network performance was monitored by analysing the energy consumption, interference or collision rate, the packets that were lost because the GWs were busy, confirmed packet success rate (CPSR), the uplink packet delivery ratio (UL-PDR) and the energy efficiency of the algorithms. These metrics were evaluated for different data intervals and different network sizes. The simulation results showed that the proposed FL-ADR scheme achieved substantial energy savings of 43% and 14% compared to the standard Semtech-ADR algorithm and the ns-3-ADR algorithm respectively. Although the CPSR and UL-PDR dropped slightly, the FL-ADR algorithm exhibited lower interference/collision rates, confirming its energy efficiency. The FL-ADR managed to efficiently adjust SF and TP despite a trade-off with CPSR and UL-PDR. Additionally, we developed an SNR-based Spreading Factor Interference Rate controlled Adaptive Data Rate Algorithm, SSFIR-ADR, an ADR algorithm that improves packet delivery ratio by reducing packet collision and managing interference. We implemented multiple static end devices connected to a single gateway to resolve the packet delivery ratio challenge without compromising energy consumption. SSFIR-ADR decreases signal interference by managing the SF allocation to reduce the probability of simultaneous transmissions with the same spreading factor in a particular annulus region in the network using a stochastic approach. The performance was evaluated using three state-of-the-art algorithms. The simulation results demonstrated that our proposed approach exhibits an improved packet delivery ratio and interference rate compared to existing solutions. These significant results contribute to the novelty of the proposed adaptive data rate algorithms. In conclusion, this research contributes valuable energy-efficient ADR algorithms for LoRaWAN communication, offering a practical and reliable solution to extend network lifetime and enhance overall network performance in IoT deployments.
Non-line-of-sight identification and mitigation for indoor localization using ultra-wideband sensor networks
(University of Pretoria, 2020) Hancke, Gerhard P.; silva.bjc@gmail.com; De Carvalho e Silva, Bruno Jorge
With the advent of Industry 4.0, indoor localization is central to many applications across multiple domains. Although impulse-radio ultra-wideband (IR-UWB) enables high precision time-of-arrival (TOA) based ranging and localization for wireless sensor networks, there are several challenges, including multi-user interference and non-line-of-sight (NLOS) conditions. NLOS conditions occur when the communication path between receiver and transmitter is obstructed, and these conditions are frequent indoors due to walls and other obstructions. To maintain location accuracy and precision similar to line-of-sight (LOS) conditions, identification and mitigation of these NLOS conditions is crucial. For identification and mitigation methods to be implemented in sensor networks, they must be of low complexity to minimize their influence on localization requirements. This thesis investigates NLOS identification and mitigation for IEEE 802.15.4a IR-UWB sensor networks. The objective of this thesis is to improve location accuracy in NLOS conditions for IR-UWB sensor networks. A comprehensive review of the state-of-the-art in NLOS identification and mitigation is conducted, and limitations of these methods with regards to the use of multiple channels, dependence on training data, mobility and complexity (particularly for applications with time constraints) are highlighted. This thesis proposes identification and mitigation methods that address the limitations found in state-of-the-art methods. A distance residual-based method for NLOS identification is proposed. Compared to conventional NLOS identification which relies on knowledge of LOS and NLOS channel statistics, or analysis of the standard deviation of range measurements over time, this identification method does not rely on these parameters. A NLOS classification method that distinguishes between through-the-wall and around-the-corner conditions using channel statistics extracted from channel impulse responses is proposed. Unlike most methods in literature that focus on distinguishing between LOS and NLOS, this method classifies NLOS conditions into through-the-wall and around-the-corner, therefore providing more context to the location estimate, and consequently enabling mitigation methods to be used for specific types of NLOS conditions. A through-the-wall ranging error mitigation method that relies on floor plans is proposed. A novel model for through-the-wall TOA ranging is proposed and experimentally evaluated. The conventional throughthe- wall TOA ranging model in literature requires many parameters which cannot be calculated in realistic scenarios. Compared to through-the-wall TOA ranging models found in literature, the proposed model relies on information from floor plans to reduce the number of unknown parameters in the model. The results show that NLOS errors caused by through-the-wall propagation are significantly mitigated with the proposed method, resulting in location accuracy which approaches the LOS case. A NLOS mitigation method which corrects location estimates affected by random ranging errors is proposed. This method relies on geometric constraints based on the fact that biases introduced by NLOS conditions in TOA range measurements are positive. The method is evaluated for cases where NLOS ranges are identifiable and cases where they are not identifiable. For the latter case, the results show that the proposed method significantly outperforms state-of-the-art optimization-based mitigation methods in terms of execution time, while retaining similar performance in terms of location accuracy.
Operational cost and emissions minimisation of a multi-microgrid system through energy management
(University of Pretoria, 2019-12) Zhang, Lijun; Ye, Xianming; u12018997@tuks.co.za; Gildenhuys, Tiaan
Existing energy management problems for a multi-microgrid (MMG) system only minimise the operational cost. This study proposes a multi-objective operational cost and emissions minimisation (OCEM) problem for a grid-tied MMG system. The energy management strategy provided by the OCEM problem takes into account the power sharing among all the microgrids (MGs) as well as a price-based demand response programme. The benefits of the OCEM problem are demonstrated by comparing its results with those of a single-objective operational cost minimisation (OCM) problem reported in the literature. These results are obtained by applying both problems to the same MMG case study, which consists of three interconnected MGs and solving these two optimisation problems with a hybrid optimisation algorithm between the genetic algorithm and sequential quadratic programming in MATLAB. When compared to the best solution provided by the OCM problem, the optimal trade-off solution (OTS) provided by the OCEM problem decreases the emissions by 73.07% with a 18.86% increase in the operational cost. The OCEM problem does provide a decision maker with the flexibility to choose the best solution according to the trade-off between the emissions and operational costs, which are competing objectives. The OTS provided by the OCEM problem has several advantages in comparison to the best solution provided by the OCM problem, namely: 1. Increase in the utilisation of the distributed energy resources in particular, the energy generated by the micro-turbines increased by 80.94%. 2. The energy imported from and exported to the main grid decreased by 52.52% and increased by 7.64%, respectively. As a result, the net energy imported from the main grid is negative, which contributes towards emission reduction and main grid support in terms of energy generation. 3. Reduction in the maximum demand of the MMG system from the main grid as the maximum power flow from the main grid to the MMG system decreased by 32.31%. The MMG case study demonstrates the capability of the OCEM problem in designing an energy management strategy, which is cost-effective and minimises the emissions.
Model-plant mismatch diagnosis using plant model ratios for a grinding mill circuit under model predictive control
(University of Pretoria, 2023-11-01) Le Roux, Johan Derik; Craig, Ian K.; heinz.m014@gmail.com; Mittermaier, Heinz Karl
Model-based controllers often extend improved performance to mineral processing plants by leveraging predictive models to account for system dynamics, handling constraints, adapting to changing conditions, and optimizing control inputs. Inaccurate models will cause a deterioration of controller performance, which is often the case for grinding mill circuits. The plant model ratio was developed to diagnose parametric model plant mismatches for first-order plus time delay models. Using a simulation study, the plant model ratio is applied to test the feasibility of using the plant model ratio on a grinding mill circuit. By applying different scenarios of mismatch, some limitations of the plant model ratio are identified and discussed in light of a grinding mill circuit model that is used in model-based controllers. The plant model ratio is capable of identifying parametric model plant mismatches for the model of a grinding mill circuit, specifically changes in the direction of responses. This may occur in cases where disturbances push a grinding mill to operate to the right of the peak of a grind curve.
Dense medium separation : A dynamic model and medium loss analysis
(University of Pretoria, 2024-02-09) Craig, I. K.; le Roux, J. D.; claire.s.lowry@gmail.com; Lowry, Claire Sarah
In this study, a model of a dense medium separation circuit of an iron ore plant in South Africa is developed, focusing on the flow of medium in the circuit. The model is validated using plant data. The model is used to simulate and detect medium losses within the dense medium separation circuit.
Few-shot Learning for Joint Classification of Instrument, Pitch, and Playing Technique of Tones Produced by Bowed String Instruments
(University of Pretoria, 2023-12) Jacobs, J. Pieter; pc.kok@tuks.co.za; Kok, Pieter Cornelius
Instrument playing technique classification is a problem in music information retrieval (MIR) that has only selectively been explored in the context of specific instrumentations or datasets. Classifying playing techniques with pitch is a further challenge that takes a step closer to automatic music transcription (AMT) with playing technique annotation. Traditional deep learning methods have been used for the problems of instrument classification, playing technique classification and multiple-instrument transcription, however, annotated data for the combined problems are scarce, thus it is hard to train a sufficiently complex deep neural network that would be able to generalize to many different instruments, playing styles and recording conditions. This study presents a few-shot learning model for joint instrument, playing technique and pitch classification of single tones using prototypical networks. The few-shot nature of the model allows it to be trained on what data are available and to adapt to new instruments, playing techniques or recording conditions at inference time from a few examples. This model could form part of a tutorial system where a music student would record scales of a given playing technique under the supervision of a music teacher, which would later be used to match and evaluate a performance with the technique. Different deep neural network (DNN) architectures and both log-mel spectrogram and constant-Q transform (CQT) input features are compared. The few-shot models are compared to standard neural network classifier models with transfer learning to show how the few-shot models generalize better to previously unseen playing techniques. Model training is optimized with Bayesian optimization. Prototypical models outperform standard classifier models with transfer learning on all experiments. The 3-shot CQT convolutional neural network (CNN) model performs the best on the joint classification task and achieves a macro F-score of 0.64 on the Studio On Line (OrchideaSOL) string instrument playing technique dataset of previously unseen playing technique classes, which shows an ability for the prototypical model to generalize to a new dataset without much loss of performance compared to evaluation on the training classes. The model also achieves a macro F-score of up to 0.855 on individual instruments, which shows promise for its use in a tutorial set up for any of the string instruments. The models perform just as well when evaluated on extracts from YouTube tutorials and examples of clarinet playing techniques from the Real World Computing (RWC) dataset. The few-shot model also functions as a multitask model, capable of classifying pitch, playing technique or instrument from a recorded sample. The best joint instrument, playing technique and pitch classification prototypical model can accurately classify both playing technique and pitch, and do so just as well or better than models trained more specifically on these problems when compared on the same data. Furthermore, the scenario of instrument, playing technique and pitch classification in the presence of piano accompaniment is investigated, which resulted in some loss of generalization, but still shows promise for the task of main melody extraction, as pitch classification remains high.
The Effect of Temperature Change on the Performance Characteristics of RF PIN Diode Limiters for VHF Applications
(University of Pretoria, 2023-06) Stander, Tinus; u10690922@tuks.co.za; Botha, Cornelius Johannes
This work presents the effect of temperature change on the capacitance of GaAs PIN diodes and the resulting change in performance of RF limiters at VHF. Device temperatures were varied between -25 ºC and 100 ºC, with small-signal parameters (including device capacitance) extracted at regular temperature increments and bias voltages from -20 V to +3 Vdc using a multi-bias parameter extraction method. It was found that the junction capacitance of the four PIN diodes under investigation increases with temperature, as expected from carrier lifetime behavior, whereas the forward-biased series resistance decreases with increasing temperature. Devices were subsequently tested in two different limiter topologies through high-power transient measurements. It was found that the combination of increased capacitance and decreased resistance with increasing temperature increases the transient spike leakage and decreases the flat leakage of a limiter. It was also concluded that, for VHF, an anti-parallel topology provides the best performance over a wide range of temperatures.
End-to-end automated speech recognition using a character based small scale transformer architecture.
(University of Pretoria, 2024-02-12) De Villiers, Pieter; De Freitas, Allan; alex.loubser@gmail.com; Loubser, Alexander
This study explores the feasibility of constructing a small-scale speech recognition system capable of competing with larger, modern automated speech recognition (ASR) systems in both performance and word error rate (WER). Our central hypothesis posits that a compact transformer-based ASR model can yield comparable results, specifically in terms of WER, to traditional ASR models while challenging contemporary ASR systems that boast significantly larger computational sizes. The aim is to extend ASR capabilities to under-resourced languages with limited corpora, catering to scenarios where practitioners face constraints in both data availability and computational resources. The model, comprising a compact convolutional neural network (CNN) and transformer architecture with 2.214 million parameters, challenges the conventional wisdom that large-scale transformer-based ASR systems are essential for achieving high accuracy. In comparison, contemporary ASR systems often deploy over 300 million parameters. Trained on a modest dataset of approximately 3000 hours—significantly less than the 50,000 hours used in larger systems—the proposed model leverages the Common Voice and LibriSpeech datasets. Evaluation on the LibriSpeech test-clean and test-other datasets produced character error rates (CERs) of 6.40% and 16.73% and WERs of 16.03% and 35.51% respectively. Comparisons with existing architectures showcase the efficiency of our model. A gated recurrent unit (GRU) architecture, albeit achieving lower error rates, incurred a computational cost 24 times larger than our proposed model. Large-scale transformer architectures, while achieving marginally lower WERs (2-4% on LibriSpeech test-clean), require 200 times more parameters and 53,000 additional hours of training data. Modern large language models are used to improve the WERs, but require large computational resources. To further enhance performance, a small 4-gram language model was integrated into our end-to-end ASR model, resulting in improved WERs. The overarching goal of this work is to provide a practical solution for practitioners dealing with limited datasets and computational resources, particularly in the context of under-resourced languages.
Specific emitter identification with multiple transmission codes and receivers
(University of Pretoria, 2023) du Plessis, Warren P.; u16076177@tuks.co.za; Diedericks, Lodewicus Johannes
This research introduces an specific emitter identification (SEI) system to enhance secure transmission systems, addressing a critical gap in existing studies. Existing studies lack a comprehensive solution for authenticating transmitters with high accuracy, particularly in scenarios involving multiple digital codes and various receiver configurations. Utilising signal extraction and a random forest classifier, the SEI system achieves a remarkable 98.6% accuracy for single transmission code authentication and sustains 95.4% accuracy when handling multiple digital codes, distinguishing transmitters, and detecting replay attacks. The system’s adaptability to diverse receiver configurations is explored, emphasising the importance of training receiver characteristics for optimal performance. This research provides a robust and adaptable solution for securing communication systems, contributing to increased trust and security in transmission systems.
An energy-efficient sensing matrix for wireless multimedia sensor networks
(University of Pretoria, 2023-08-22) Abu-Mahfouz, Adnan; u04318218@tuks.co.za; Skosana, Vusi Josias
A Wireless Multimedia Sensor Network (WMSN) make possible new surveillance applications in environments that traditional systems would not handle, including search and rescue operations after a disaster. However, WMSNs ought to perform under energy-constrained conditions that insist on novel compression methods to diminish bandwidth usage and extend network lifespan. Compressed Sensing (CS) was presented as a means to achieve overcome the challenges faced by WMSNs. A sensing matrix is crucial to the compressed sensing framework. The sensing matrix can maintain the fidelity of a compressed signal, diminish the sampling rate obligation and improve the strength and performance of the recovery algorithm. A great number of measurement matrices have been proposed to either offer reduced computational complexity or good recovery performance, but only some have managed to accomplish both, and even fewer have been proven in a compelling manner. There are images that do not lend themselves to compression, and to maintain Quality of Service (QoS) expectations, adaptive sampling is essential. Low-performance nodes are essential for making WMSN practical and flexible. Different low-performance nodes have been proposed in the literature, but the Telos Revision B (TelosB) sensor module (mote) can be used as a reference for energy-constrained applications. TelosB is a very low power wireless mote for research and experimentation. The design of sensing matrices has been influenced by practical considerations in WSN. The two major innovations have replaced floating point numbers with bipolar and binary entries and sparse sensing matrices. The Deterministic Partial Canonical Identity (DPCI) matrix was presented to address the needs of an energy-constrained environment for WMSN. The choices of random number generators were discussed, and criteria were developed for selection. Complexity optimisation was undertaken to improve the time complexity of the construction. The DPCI was outperformed by the Deterministic Binary Block Diagonal (DBBD) and Binary Permuted Block Diagonal (BPBD) in terms of recovery performance but gave a substantial computational cost reduction. The DPCI gives a compelling balance between recovery performance and energy efficiency, benefiting energy-sensitive applications. A recovery performance prediction algorithm was also proposed to be used for an adaptive sampling scheme.
Low-cost capacitive CMOS capacitive E. Coli biosensor for point-of-need water quality monitoring
(University of Pretoria, 2023) Joubert, Trudi-Heleen; george.andrews@tuks.co.za; Andrews, George
Exposure to pathogenic Escherichia coli (E. coli) bacteria through contaminated water can cause potentially life-threatening diarrhea and vomiting, and it is a useful water quality indicator. Simulations and experiments are conducted to give guidelines on low-cost capacitive biosensing devices aimed at bacterial sensing, and a custom integrated circuit that can be used in a low-cost capacitive biosensing device is delivered. Finite element modelling was conducted to compare the electric fields (E-fields) and capacitance across different electrode geometries and materials. The size of electrode features had the biggest impact on electric field strength and relative capacitance change in the presence of cell-like structures. The simulation results are used to clarify assumptions on how simulations for the design of capacitive sensing electrodes need to be conducted. Capacitance measurement of low-cost and commercial electrodes was conducted using a benchtop LCR meter and 3 μm microbeads as substitutes for E. coli cells. It was found that the measured capacitance increases as the concentration of microbeads increases, and the low-cost electrodes seem to show a higher-than-expected sensitivity when compared to commercial electrodes with smaller feature sizes. Electric impedance spectroscopy experiments conducted on E. coli cells showed a similar performance as characterisation experiments using microbeads. These insights inform the development of guidelines that may be used to design low-cost electrodes for similar applications. A custom integrated circuit (IC) featuring a capacitive sensing array with sub-surface electrodes was designed and delivered. This IC also includes the custom operational amplifier used in the sensing array, used in a low-cost capacitive biosensor prototype for point-of-need use. The low-cost device was characterised with 3 μm microbeads using a subset of the electrodes used in the LCR experiments, with comparable results achieved using the low-cost device. Lessons learned in the design of the low-cost system guide the development of a design flow for the design of point-of-need water quality monitoring devices and gives guidance on the required hardware to build such devices.
Micro-grid energy optimisation with demand management
(University of Pretoria, 2023) Naidoo, Raj; u15015859@tuks.co.ca; Louw, Jacobus
Pressure on the utility grid and aims to reduce carbon emissions have led to the development of micro-grids. Extensive research has been done to optimally utilise the various energy sources present therein. Demand management strategies have also been developed to minimize strain on the national utility grid. However, not much research has been done to consider demand management in a microgrid environment. This article proposes the development of a microgrid for a small residential complex in South Africa. Heuristic optimisation techniques are investigated to ensure optimal energy source utilisation and load scheduling, while minimising inconvenience caused for consumers. It is expected that the system will offer significant long-term savings as well as a stable supply of electricity.
Optimal sizing and control of supercapacitors for cost-effective hybridization of battery-alone energy storage systems
(University of Pretoria, 2022) Xia, Xiaohua; Zhang, Lijun; u15394027@tuks.co.za; Masaki, Mukalu Sandro
With growing effects of global warming, batteries have a crucial role to play in supporting global effort to reach carbon neutrality. However, batteries remain prone to early failure. Operational causes of premature deterioration are high current intensity, abrupt current fluctuations, repetitive charge-discharge cycles, deep discharge, and thermal stress. An effective way to mitigate stress sustained by the battery, and therefore increase its service life, is association of batteries and supercapacitors (SC) into battery-SC hybrid energy storage systems (HESS). Despite considerable interest enjoyed among the research community, the adoption of battery-SC HESS is still marginal in practice due to the relatively high cost of SC. As a result, most batteries are still implemented as stand-alone energy storage systems, or battery-alone energy storage systems (BESS). Given the continuous advancement in the manufacturing economics of SC, in one hand, and the market domination of BESS, in the other hand, close attention should be paid to the upgrade of BESS into battery-SC HESS. The study of contemporary literature shows that such a conversion has been scarcely investigated before. Most control strategies reported in the past were also found unsuitable for retrofit applications, due to significant modifications required on the existing control infrastructures. In view of technical, economic, and environmental implications of such modifications, this may raise concerns for stakeholders. Regarding sizing of battery-SC HESS, it was found that previous investigations focusing on their control used arbitrary SC size to test the performances of controllers. On the other hand, studies interested in the economic sizing of this HESS usually failed to fully assess the costs and benefits over the entire lifespan of the energy system. Moreover, other shortcomings such as a fixed service life for each ES equipment regardless of the operating conditions, overlooking of certain cost components and economic parameters (inflation and escalation rate of electricity price) were also found in some studies. This usually results in misleading assessments of costs and benefits associated with battery-SC HESS. In an attempt to address the above gaps in the current literature, this thesis presents two control strategies aimed at achieving trouble-free retrofit of BESS with SC, a preliminary investigation on the impact of spatial arrangement on the thermal stress sustained by the battery and SC cells, and finally an SC sizing model for cost-effective hybridization of BESS. Primarily designed for hybrid renewable systems (HRS) originally equipped with BESS and controlled by a receding horizon control (RHC), the first control scheme consists of a hierarchical RHC of the SC-retrofitted HRS. Depending on the characteristics of the power management unit (PMU) running the previous RHC scheme, no or little modifications are required to integrate the SC into the existing infrastructure. Besides the reduction in electrical stress sustained by the battery, the proposed control framework also increases the amount of energy supplied by intermittent renewable resources and the power stability at the point of common coupling (PCC). The second control model is built around a fuzzy logic controller unit to assist in retrofitting any BESS with SC. Thermoelectric management of batteries is realized through sharing of low-frequency current components between the two ES devices, besides the SC’s full supply of high-frequency current components. Improvement in battery service life is demonstrated. Compared to the previous controller, specially designed for retrofitting of BESS-equipped power plants controlled by RHC, this thermoelectric controller is intended for implementation on any existing BESS. The preliminary study conducted on the influence of spatial arrangement of battery-SC HEES cells on the thermal management of batteries demonstrates that the proximity between a battery cell and an SC cell can effectively contribute to the cooling process of batteries, thanks to thermal interaction between them. Accordingly, adequate arrangement of ES cells can offer a passive assistance to the above controller in achieving thermoelectric management of batteries. Finally, the thesis introduces a life cycle cost (LCC)-based optimization model that assists in properly sizing SC intended for retrofit in existing BESS. The possibility offered to stakeholders to take informed decisions about the economic opportunity of such an upgrade is also demonstrated.
Evaluation and expansion of observable dynamic froth flotation models for control
(University of Pretoria, 2023) Le Roux, Johan Derik; Craig, Ian K.; jacolouis.venter@gmail.com; Venter, Jaco-Louis
This work builds on existing observable dynamic models of froth flotation circuits, aimed at on-line parameter estimation and model-based control. The models are analysed and two main limitations are identified and addressed: the lack of explicit modelling of reagent effects and the need for dynamic validation on large-scale industrial plant data. The feasibility of expanding a froth flotation model to include reagent effects is investigated. A Sobol sensitivity analysis is used to identify the crucial parameters. The model is expanded with two different reagent effect models. Both expansions include mass balance models of the frother concentration in each cell. The first model expands an empirical parameter in the air recovery model, related to the froth height at which peak air recovery (PAR) is achieved, as a linear function of frother concentration. The second model adds a linear frother concentration term to the existing air recovery model to modify the steady-state air recovery directly. Observability analyses of the expanded models show that all states and the important time-varying model parameters are observable (and identifiable) from the available on-line measurements. Most importantly, the frother concentrations are shown to be observable without concentration measurements. Simulations of the model expansions show that the second model can qualitatively predict the impact of increased frother dosage on air recovery, grade and recovery, while the first model can only predict the correct effect under certain conditions. The implementation of a Moving Horizon Estimator (MHE) based on the model (excluding reagent effects) on data from an industrial rougher bank is investigated with the aim of validating the model and parameter estimation approach. The available plant data and its limitations are discussed and additional model analysis is conducted. An expanded observability analysis of the model identifies groups of parameters for which identifiability is linked. It is shown that without on-line compositional measurements only a reduced model that lumps all recovery mechanisms into a single empirical equation is observable. The reduced model is used to develop the MHE which is implemented on data from the Mogalakwena North Concentrator (MNC) historian. The state and parameter estimates are then used to evaluate the model prediction accuracy over a shifting control horizon, as would be done in model predictive control (MPC). Estimation results show that there are substantial amounts of unmodelled dynamics and/or disturbances. Parameter estimates compensate somewhat, but the model predictions are only accurate over some sections of the data. The lack of on-line compositional measurements as well as uncertainty regarding the validity of calculated measurements and assumptions prevented a fair evaluation of the full potential of the model, but served to highlight drawbacks and challenges that will need to be addressed in future work.

Browse

Recent Submissions