[diss] Sähkötekniikan korkeakoulu / ELEC

Permanent URI for this collectionhttps://aaltodoc.aalto.fi/handle/123456789/53

Browse

Recent Submissions

Now showing 1 - 20 of 1056
  • Algorithms for robust human-machine interfacing via surface electromyography
    (2024) Yeung, Dennis
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Surface electromyographic (sEMG) signals are an information-rich medium from which motor intention can be decoded. This property has led to their use as the primary interface between amputees and robotic prostheses for decades. Since the inception of such prosthetic systems, their mechatronics design and onboard processing power have advanced significantly. Similarly, the algorithms that facilitate the estimation of user commands have evolved from simple heuristics to data-driven models. However, the increased complexity of the interfacing algorithms has also incurred greater sensitivity to signal non-stationarities that can manifest during device operation and deteriorate control performance. Perspiration, fatigue, and electrode displacement are common sources of such signal perturbations and these issues extend to any wearable human-machine interface (HMI) that utilizes sEMG. To address this, adaptive control models that compensate for such signal perturbations and restore control performance have been proposed. Moreover, the emergence of surface decomposition algorithms that extract motor unit (MU) firing times from high-density sEMG recordings has also spurred the investigation of HMI driven by MU spike trains. This alternate means of extracting neural information from surface recordings offers certain advantages over traditional features derived from the global sEMG signal. Since MU spike times constitute the most basic bit of neural information responsible for force generation, more intuitive and dexterous interfacing may be derived. In this doctoral thesis comprising four journal articles and two conference articles, contributions in the domain of non-adaptive, supervised adaptive, and unsupervised adaptive control models are made. Furthermore, methods that advance the integration of MU decomposition techniques to HMI applications are proposed. Publication I proposes a directional forgetting extension to existing supervised adaptation schemes for improved co-adaptive stability. Publication II investigates the effects of imposing sparsity constraints to a muscle synergy-inspired control model in terms of robustness against electrode shifting. Publication III presents an unsupervised learning scheme to facilitate co-adaptive control using the muscle synergy-inspired model. Publication IV investigates the incorporation of feature selection methods to the initialization of MU-driven HMI. Publication V presents a semi-automatic method for generating benchmark decompositions referenced against intramuscular EMG recordings. Finally, Publication VI presents an adaptive algorithm for online decomposition capable of compensating for changes in joint angle and contraction intensities.
  • Automatic Classification of Voice Disorders and Phonation Types from Speech Signals
    (2024) Javanmardi, Farhad
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Speech is the most natural form of human communication that not only conveys a message in the form of language, but also transfers information about speaker attributes such as emotions, age, and state of health. Systems that automatically classify this para- and extra-linguistic information have received considerable attention in speech technology. One area of this research is the study of automatic classification of voice disorders from speech. This area has potential applications in voice care as it provides objective and cost-effective analysis tools compared to subjective and time- consuming auditory-perceptual assessments currently performed by clinicians. This thesis investigates the effectiveness of various feature extraction methods and supervised machine learning (ML) and deep learning (DL) techniques in the development of automatic speech-based classification systems for two tasks: (1) classification of voice disorders and (2) classification of phonation types. Multiple classification systems are built, each designed to address a specific topic. The first topic is the study of systems to detect voice disorders in various laryngeal diseases from sustained phonation of vowels. Moreover, the application of data augmentation (DA) is studied in this topic. As the second topic, the thesis investigates the detection of dysarthria and the multi-class classification of the severity level of dysarthria from speech. As the third topic, the thesis investigates the classification of three phonation types (breathy, modal, and pressed). In the classification of voice disorders, the use of convolutional neural networks (CNNs) with 2- dimensional spectral feature representations achieved better performance than 1-dimensional features. Moreover, the use of DA methods in the system training showed absolute accuracy improvements of up to about 4%. In the classification of dysarthria, the use of pre-trained model- based features showed in the best cases absolute accuracy improvements of about 10% compared to conventional features. Furthermore, fine-tuning the pre-trained models resulted in features with better generalization capabilities in dysarthria detection. In the classification of phonation types, the use of the neck surface accelerometer (NSA) signal showed better classification performance compared to the speech signal. In addition, pre-trained model-based features outperformed the conventional features for both speech and NSA signals. In conclusion, this thesis resulted in improvements in automatic, speech-based classification of voice disorders by combining ML and DL classifiers with spectral and pre-trained model-based features and by taking advantage of NSA in the classification of phonation types.
  • Towards free-positioning self-adaptive wireless power transfer systems
    (2024) Al Mahmud, Shamsul Arefeen
     School of Electrical Engineering | Doctoral dissertation (article-based)
    The quest for achieving wireless power transfer (WPT) over a large area to facilitate the charging of multiple devices has given rise to numerous challenges, particularly concerning the free positioning charging freedom of the receiver anywhere in a large area. Addressing these challenges, this doctoral dissertation proposes and validates several innovative solutions and methods for enhancing the performance and robustness of large-area WPT systems. One of the primary obstacles in the WPT system is the deterioration of performance as the receiver's position and characteristics change. This necessitates the observation of load resistance and receiver position to enable effective power transfer. This dissertation focuses on a data-driven approach to estimate WPT system characteristics to ensure effective power transfer. The machine learning assisted approach presented in the dissertation only relies on the transmitter-side measurements. This method differentiates between low and high system efficiency which assists in deciding turn-on and -off transmitters in multi-transmitter WPT systems. Additionally, the dissertation presents a method aimed at efficient positioning-free wireless power transfer over a considerably large area. This is achieved via a novel transmitter arrangement within a pad-like area, thereby assuring unbroken paths for magnetic flux for effective power transfer. The dissertation also introduces an innovative technique for the activation and deactivation of the transmitters based on the position and rotation of the receiver, using only the measurements taken from the transmitter side. This is a crucial element for the future implementation of high-efficiency and cost-effective dynamic wireless power transfer devices. This strategy eliminates the need for any additional sensors or data communication, leading to potential applications in the wireless charging of industrial robots and drones. In this dissertation, a planar receiver structure comprising three coils is presented, which enables unimpeded positioning within the designated transmitting zone, thereby sidelining the necessity for intricate control mechanisms or the tracking of receiver positioning. This revolutionary design for the receiver ensures a consistent rate of efficiency across the entirety of the WPT area. In addition, the dissertation discusses the coil design while also demonstrating how to optimize the volume and shape of the magnetic materials employed in WPT transmitters. Finally, this dissertation puts forth a novel methodology for coil activation in multi-transmitter wireless power transfer systems, thus providing complete flexibility in receiver positioning over a large area. This non-coherent power combining method eliminates power transfer blind spots and exhibits a steady DC-DC power transfer efficiency of approximately 90%, irrespective of arbitrary receiver movements, and dispenses with the necessity for any additional dynamic controls.
  • Modeling and Control of Bearingless Linear Motors
    (2024) Hosseinzadeh, Reza
     School of Electrical Engineering | Doctoral dissertation (article-based)
    This dissertation proposes dynamic modeling and model-based control design for bearingless linear motor systems. A bearingless motor produces both the traction and levitation forces using the same iron core. In bearingless linear systems, the magnetic levitation of the moving part (mover) renders the linear mechanical bearings redundant. However, controlling bearingless systems is more challenging than those with mechanical bearings. Conventional control methods, such as Proportional-Integral-Derivative (PID) controllers, can be employed, but their tuning method is often heuristic. Therefore, this dissertation considers model-based control methods. The prerequisite to this control design methodology is a sufficiently accurate dynamic model of the system. Thus, this dissertation considers two dynamic models for bearingless systems. A dynamic model is presented for a bearingless Linear Flux-Switching Permanent-Magnet (LFSPM) motor. This dynamic model is developed based on magnetic equivalent circuits, including the effects, such as magnetic saturation and airgap variation. For a double-sided bearingless LFSPM motor system, a state-feedback control method is presented. An approximate feedback linearization method is proposed to account for any nonlinear airgap dependency and the cross-coupling between the attraction and thrust forces. For energy efficient control, a resistive-loss minimization method is proposed. The minimization algorithm calculates the minimum reference currents for the given force references and the position of the mover. The results from the optimization method are used in the form of lookup tables and artificial neural networks. A comparison is provided between the two implementation methods. A dynamic model is developed for six-degree-of-freedom bearingless systems. The example system is a quadruple-sided bearingless linear motor system comprising eight three-phase motor units. To model the unbalanced magnetic torque, each motor unit is spatially divided into several submotors. The submotors of a motor unit have the same current, while the flux linkages and forces are different in a tilted position. The developed models can be utilized in time-domain simulations for system analysis and real-time control system development. The methods presented in this dissertation are evaluated using double-sided and quadruple-sided bearingless flux-switching permanent-magnet linear motors.
  • Resource optimization for massive MIMO systems
    (2024) Kocharlakota, Kameswara Atchutaram
     School of Electrical Engineering | Doctoral dissertation (article-based)
    The thesis delves into the intricacies of resource optimization in both massive Multiple-Input Multiple-Output (MIMO) and cell-free massive MIMO (CFmMIMO) systems, which are pivotal for the advancement of 5G and beyond wireless networks. The research primarily addresses the challenges of pilot contamination and power allocation, which significantly impact the spectral efficiency (SE) and overall performance of these systems. Initially, the thesis explores the massive MIMO systems, focusing on the impact of pilot overhead and the accuracy of channel estimation on the SE. Closed-form expressions for the uplink (UL) and downlink (DL) SEs under conditions of imperfect channel state information (CSI) are derived. These expressions are crucial in understanding the trade-offs involved in pilot resource allocation, emphasizing that efficient pilot management is essential for maintaining high system performance. The analysis provides the closed-form expressions as vital tools for selecting optimal pilot overhead parameters. In the latter part, the thesis shifts focus to CFmMIMO systems, which distribute antennas across a large area to provide uniform coverage and enhance the performance of cell-edge users. Here, the primary challenge addressed is the downlink power control. Traditional methods for power control are computationally intensive and often inadequate for the centralized nature of CFmMIMO systems. To overcome these limitations, the research introduces advanced deep learning techniques, specifically Attention Neural Networks (ANN) and Pilot contamination-Aware Power Control (PAPC) transformer neural network, for power control. These models leverage the capabilities of masked multi-head attention networks, enabling efficient power allocation even in the presence of pilot contamination. The ANN-based approach initially transforms the constrained optimization problem into an unconstrained one, optimized through unsupervised learning. Subsequently, PAPC further refines this approach by incorporating additional architectural enhancements, such as pre-processing and post-processing stages, which improve performance and scalability while reducing computational complexity. Extensive simulations validate the effectiveness of these proposed solutions, demonstrating their potential to significantly reduce the computational complexity while providing state-of-the-art performance in CFmMIMO systems. In conclusion, this thesis makes significant contributions to the field of wireless communications by providing innovative solutions and comprehensive analytical tools for resource optimization in massive MIMO and CFmMIMO systems. The findings and methodologies presented are expected to pave the way for more efficient and reliable next-generation wireless communication technologies, addressing critical challenges in pilot resource allocation and power control.
  • Statistically Robust and Sparsity Promoting Inference and Estimation for Large-Scale Data
    (2024) Mozafari Majd, Emadaldin
     School of Electrical Engineering | Doctoral dissertation (article-based)
    The proliferation of data at a massive rate has led to the generation of massive heterogeneous datasets that are now being collected and analyzed in almost every business, industry, and science area. With datasets expanding in volume and dimensionality, many traditional procedures in signal processing, statistics, and machine learning face difficulties such as high computational complexity, lack of robustness, and reduced statistical performance. Furthermore, the probability of having missing and outlying data points grows when observing large-scale data, which further complicates estimation and inference tasks. The realization of sophisticated inference and estimation procedures that are compatible with distributed computing architectures, robust against outlying data points, and maintain reliable or even optimal statistical performance when sample size, dimensionality, or both grow is of paramount importance. This dissertation addresses two commonly arising problems in the analysis of large-scale data, which are possibly high-dimensional. It introduces new contributions in terms of theory and algorithms to the fields of inference and estimation. First, we develop two-stage robust and distributed statistical inference procedures for sparse linear regression models. These procedures combine a collaborative variable selection process, which identifies relevant variables, with computationally low-complexity and robust adjusted one-step bootstrap methods applied to data with only selected relevant variables. This approach allows for leveraging distributed computation and storage architectures while remaining robust to outlying data points and heavy-tailed distributions. The inferential tasks considered in this dissertation include the construction of confidence intervals for estimated parameter values within the identified support set (the subset of parameters associated with relevant variables) and estimating standard errors for the underlying estimator in the bootstrap method. This dissertation also provides theoretical results by establishing the bootstrap consistency and robustness properties of the proposed inference procedures. Second, we introduce a new regularization-based estimator for high-dimensional sparse linear models in the face of heavy-tailed distributions and outlying data points. The proposed adaptive τ-Lasso estimator possesses desirable oracle properties. This estimator is asymptotically normal when applied to data with its design matrix restricted to truly relevant variables. Moreover, it achieves variable selection consistency in the asymptotic sense. Furthermore, we establish the finite-sample breakdown point and characterize the influence function. Simulation results are presented to verify the analytical robustness results and demonstrate the competitive performance of the class of τ-Lasso and adaptive τ-Lasso in prediction RMSE and variable selection criteria, particularly in scenarios involving adversarial contamination.
  • Manipulation of nonmagnetic particles and liquids on a programmable air-ferrofluid interface
    (2024) Harischandra, P. A. Diluka
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Magnetic manipulation, a technique that manipulates magnetic objects using magnetic fields has emerged providing significant advantages in a multitude of fields, such as robotics, materials science, and biomedical engineering. Although there have been remarkable advances, traditional magnetic manipulation methods require the manipulated objects to be magnetic. However, most materials in nature are nonmagnetic, and therefore the current magnetic manipulation methods for nonmagnetic materials are insufficient. This thesis addresses the challenges in manipulation of nonmagnetic particles and presents novel methods to enhance the manipulation capabilities for liquid droplets on an air-ferrofluid interface. First, this thesis introduces a novel approach for closed-loop magnetic manipulation of non-magnetic particles on a programmable air-ferrofluid interface. The method uses magnetic fields for the programmable deformation of the air-ferrofluid interface, enabling the movement of particles. The proposed automatic manipulation method can perform path following of nonmagnetic particles in predefined trajectories. This approach overcomes the limitations of state-of-the-art methods that are restricted to closed-loop manipulation of magnetic particles and it broadens the manipulation capabilities at the air-liquid interface for living and non-living matter. Secondly, this thesis presents a novel approach to shape nonmagnetic liquids on a programmable air-ferrofluid interface. Current methods for shaping liquid droplets using magnetic fields require the manipulated droplet to possess magnetic properties. This thesis demonstrates that the deformations created at an air-ferrofluid interface can produce diverse convex and concave shapes of a nonmagnetic droplet at the same interface. Additionally, droplets can also be rotated or stirred. The methods presented in this thesis have significantly enhanced the manipulation capabilities for liquids at the air-liquid interface. Thirdly, this thesis proposes a novel method to predict droplet shape evolution resulting from sequences of actuations and the required actuations for a given shape sequence using a Long Short-Term Memory (LSTM) network. The method can be used to learn sequences of convex or concave shapes of the droplet and allows capturing and retaining the sequence of droplet morphology changes due to actuations. Lastly, this thesis presents a breakthrough in automatic shaping of liquids. Existing methods to automatically shape a droplet are limited to elliptical shape and the droplet must be magnetic. This research introduces a data-efficient online learning method using Bayesian optimization to morph liquid droplets into desired target shapes on the programmable air-ferrofluid interface. Using this method, desired convex or concave target shapes of the droplet can be achieved in a few iterations.
  • Critical Technologies and Architectural Research in the Context of Lunar Habitats
    (2024) Nyman, Leo
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Space habitats are currently a topic of interest for space agencies around the globe. As of now, such habitats are operational only in low Earth orbit, but this is expected to change in the near future. An international consortium is in the process of constructing hardware for the planned Lunar Gateway, which will be the first space habitat situated beyond the protective cover of Earth's magnetosphere. Additionally, several nations have plans to establish outposts in the Moon's south pole region. These deep-space environments challenge technology in specific areas, including additive manufacturing, thermal design, sensors, and specialised thin-film coatings. This dissertation presents research findings in the aforementioned areas. Of particular significance is additive manufacturing, given that logistical support for the initial lunar outposts will be limited, making the in situ production of spare parts crucial. To address this issue, this work outlines a path towards additive manufacturing of space-grade parts and evaluates their performance. Importantly, the development of additive manufacturing for use in space will enable the use of locally resourced materials on the Moon. With relevance to thermal systems, this work showcases the performance of a low-emittance copper-coating system enhanced with atomic-layer-deposited aluminium oxide. Such coating systems are vital to provide the required thermo-optical properties necessary for the external surfaces of spacecraft. These surfaces can also fulfill aesthetic roles and thus be useful to architects designing future habitat systems for the Moon. Despite robust thermal systems, spacecraft components are commonly exposed to low and elevated temperatures. Therefore, parts that have been additively manufactured must survive thermal cycling tests designed to screen spacecraft parts. Furthemore, parts made using additive manufacturing can be coated with area-selective atomic layer deposition, and this work explores some of the synergies of this technique. This dissertation represents a multidisciplinary approach – merging technology and architecture – in the context of lunar habitats.
  • Contact Metallization Design for Low-Temperature Interconnects in MEMS Integration
    (2024) Emadi, Fahimeh
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Advanced electronic interconnects must meet various criteria, including low-temperature (LT) processing, miniaturization, and a stable microstructure with optimal electrical, mechanical, and thermomechanical properties. Specific application requirements, such as those for micro-electromechanical systems (MEMS) combining mechanical and electrical parts on a micrometer scale, must also be considered. Cu-Sn solid-liquid interdiffusion (SLID) interconnects show potential for utilization in MEMS integration. However, challenges persist, such as high processing temperatures and the complexity of wet chemical electroplating materials (e.g., Cu and Sn) on wafers containing fragile movable MEMS devices. Addressing these challenges involves employing LT Sn-In instead of Sn, isolating the electroplating processes to passive structures, and using a thin film contact metallization deposited via physical vapor deposition (PVD) on the device's wafer side, providing an alternative to wet chemical processes. Therefore, this thesis aimed to identify suitable contact metallizations for Cu-Sn-based SLID systems, design the metallization stack, accordingly, investigate the microstructural evolution and mechanical properties of the interconnects, assess their reliability, and demonstrate the utilization of LT-SLID, connected with TSVs, to create three-dimensional (3D) interconnects with a specific focus on 3D MEMS integration. Cobalt emerged as the most viable contact metallization option to interact with Cu-Sn-based SLID interconnects. The results showed that the microstructure of the Cu-Sn/Co bond line evolves over time at elevated temperatures or longer bonding times; hence, the Co-to-Sn thickness ratio must be controlled to prevent bond failures. No such concerns were observed with the Cu-Sn-In/Co joints. Moreover, the Cu-Sn-In/Co system exhibited promising results that met the interconnects' requirements, such as the tensile strength requirements of MIL-STD-883 method 2027.2. The bonding process was demonstrated at temperatures as low as 200 °C, resulting in a void-free stable microstructure even after a high-temperature storage (HTS) test. Furthermore, the tensile strength of the bonds improved after the HTS test. The compatibility of the developed interconnects with TSVs was confirmed, enabling the fabrication of 3D SLID-TSV interconnects for the advanced integration of MEMS devices. These interconnects demonstrate better performance than Cu-Sn SLID-TSV interconnects, which have faced challenges such as silicon cracking and void formation. This finding highlights the effectiveness of the 3D LT-interconnects.
  • Design and optimization of a decentralized multi-robot exploration behavior taking into account energy constraints
    (2024) Leal Martínez, David
     School of Electrical Engineering | Doctoral dissertation (monograph)
    The robot revolution is just around the corner. Robots have already started appearing in public spaces and are becoming available to everyone. However, most robots today still need a human in the loop supervising, recharging batteries, or making decisions, especially in the face of uncertainty. There are still some pieces of the robot autonomy puzzle that are still either missing or that need to be refined to enable robots to work completely on their own. One such piece is the robot exploration problem: how can robots explore very large spaces without human intervention? This piece is crucial for tasks, such as space exploration or disaster relief. This thesis work focuses on designing and optimizing a distributed exploration strategy called Decentralized Frontier-based Exploration (DFBS). This strategy aims to allow every robot to make its own decisions based on their perception of the world by using a shared map created by all the robots in the group. This work builds on top of the Frontier-based exploration strategy, that defines a frontier as the borderline between explored and unexplored space, and extends it by using of a fully decentralized approach that tackles fault tolerance, and also considers robots with limited energy reserves and their replenishment. Field and service robotics has been one of the main foci of the Center of Excellence in Generic Intelligent Machines, part of the Automation and Systems Technology department of Aalto University. In this laboratory, the MarsuBot robot fleet was created as a test bed for multi-robot exploration. This work aims to contribute to the efforts of the laboratory by creating a simulation environment called MarSim, in which the MarsuBot fleet can be simulated, and the DFBS strategy can be implemented and optimized. Experiments were performed using MarSim, simulating scenarios of varying difficulty to evaluate the performance of the proposed exploration strategy in comparison with a purely stochastic and reactive strategy. Over six million simulations were performed in Triton, Aalto University's supercomputer, to search for the best parameter combination that minimized the energy spent by the whole fleet while exploring the different scenarios. These results were compared to the ones obtained by an off-the-shelf Bayesian Optimization tool running on a single computer. The results offer a complete analysis of the performance of the Decentralized Frontier-based Exploration in comparison to a basic reactive behavior, along with the optimization of its parameters using Bayesian Optimization and verification of the results by comparing them to the experimental results obtained using Grid Search. This work also analyzes the performance of Bayesian Optimization as a tool to optimize the starting parameters of a robot in a very small amount of experiments to evaluate the possibility of using it to optimize the real MarsuBot fleet.
  • Improved Calibration and Uncertainty Estimation Methods for Optical Radiometry
    (2024) Maham, Kinza
     School of Electrical Engineering | Doctoral dissertation (article-based)
    This thesis focuses on developing and characterizing detectors, measurement setups, and methods tailored for optical radiometry. Additionally, emphasis is placed on the importance and methods of uncertainty estimation in metrology, aiming for reliable measurements. The predictable quantum efficient detector (PQED) ensures traceability of optical power to the SI system, making it a promising primary standard for detector calibration. The PQED is used to calibrate standard trap detectors, achieving an expanded uncertainty of 0.05% in the visible region. A comparative analysis with a cryogenic radiometer validates the PQED as a primary calibration standard. To extend calibration capabilities from the visible region upto the wavelength of 2000 nm, a portable tunable laser line setup is built. As a reference, the setup uses InGaAs and Ge detectors that are characterized against a pyroelectric radiometer. The tunable laser's increased spectral power reduces the expanded uncertainty from 4% to 2.2% - 2.6% across 820 nm - 1600 nm compared to older calibration methods at Metrology Research Institute (MRI). This thesis also discusses estimation of uncertainties and their correlations within the spectral mismatch factor of solar cells using Monte Carlo analysis. Our study assesses the uncertainties linked with the spectral mismatch factor (SMM) across various scenarios. We examine three scenarios: the worst-case scenario, where components are assumed to have severe correlation; the average scenario, where partial correlation is presumed; and the best-case scenario, where components are considered uncorrelated. The resulting expanded uncertainties (with a coverage factor k = 2) for these scenarios are 1.26%, 0.44%, and 0.06%, respectively. These figures represent the spectrum of potential uncertainties based on the assumed correlation conditions among the components involved in the SMM correction factor. The new method for calculating uncertainties benefits from prior knowledge of correlations. This knowledge is sought by analysing Consultative committee for photometry and radiometry (CCPR) key comparisons of radiometric quantities spectrally. This analysis provides insights into spectral correlations, aiding in the quantitative assessment of uncertainties in spectral integrals. Leveraging these insights, the uncertainty estimation methods in optical radiometry, including SMM and color-correlated temperature (CCT), are improved.
  • The Gopher Antenna: A New Type of Ground Penetrating Radar Antenna
    (2024) Voipio, Veli
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Currently, ground penetrating radar (GPR) antennas for ground-coupled setups are inefficient. They use resistors and/or dissipative material to improve the impedance bandwidth and thus achieve optimal pulse shape. These antennas are mostly bowtie dipoles and have an omnidirectional pattern, although the casing and the electrical properties of the ground modify the pattern. The center frequency of the antennas in free space is often 500 MHz, and the required bandwidth is wide: the spectrum ratio is from 1:2 to 1:10 at a −10 dB limit. The phase center must remain stable over the whole bandwidth. The main objective of this dissertation is to develop an antenna that is more efficient and has good directivity. The concept of a patch antenna with a dual resonant structure was used as a basis. To achieve this goal, a feed antenna with three resonant frequencies was used. The novel point in the design is the second resonant frequency (not the first, as is typical) of the feed that is matched to the line impedance by a coupled dual resonant parasitic patch structure. Around the center frequency, the matching is good, and there are no dissipative materials. Therefore, the antenna is efficient. The radiated spectrum expands well beyond the matched impedance area and thus is not efficient in that part of the radiated spectrum, but it provides a Gaussian spectrum which is preferred by GPR users. The patch antenna design provides an inherent directivity, often 9 dBi, and combining the feed improves the directivity further. For practical reasons, this antenna type was given the name the "Gopher antenna". A reasonably priced GPR without an antenna was provided by one manufacturer. This was partly modified by the author to fit better the Gopher antenna so that the tests could be performed. Measurements were done in settings where there were known objects underground, and on a lake where the lake bottom was visible. The received data needed extensive processing. The sophisticated free software for the processing did not read the files from the system, thus a simple processing software was designed by the author. This had the advantage of enabling the testing of various pre- and postprocessing methods, and the trace integration method with optional deconvolution and cross-correlation methods were found to be useful with this data. The antenna was implemented, and the measured results validate the concept. The lake profiles are quite clear, and the ground profiles show reflections from known objects in the expected size range. As there was no standard way of describing the antenna radiation in the ground, I propose that the highest electric field magnitude is stored in each FDTD pixel during the simulation. This provides a useful graphic with which to compare antenna patterns. The pattern was also tested in a case study by tilting the antenna. With this efficient and directive antenna, the GPR is expected to see deeper.
  • Towards Smart Cities: Antenna-Embedded Walls and Antenna-User Interaction Modeling for Enhanced Urban Connectivity
    (2024) Vähä-Savo, Lauri
     School of Electrical Engineering | Doctoral dissertation (article-based)
    With the rapid evolution of mobile networks and the ongoing deployment of 5G, there is a pressing need for innovative solutions to accommodate the growing number of users and higher data rates facilitated by higher carrier frequency bands. However, as the carrier frequency increases, communication networks face heightened sensitivity to blockages, impeding signal propagation and reducing reliability. This thesis investigates novel electromagnetic modeling techniques of antenna-embedded building walls and antenna-human interaction to enhance 5G and future mobile communication systems, providing tools to develop and test improvements that mitigate their adverse effects on communication quality. The first part of this thesis explores the concept of signal-transmissive walls, where two planar antenna elements are connected back-to-back by a coaxial cable embedded into a load-bearing wall to guide signals from outside into a building. A load-bearing wall, with and without the embedded antenna system, is studied in terms of electromagnetic transmission, thermal insulation, and mechanical stress distribution. Numerical simulations of electromagnetic transmission coefficient, up to 8 GHz, demonstrate that the signal-transmissive wall significantly improves transmission coefficients for carrier frequencies above 2.6 GHz. The full-wave simulation models are later validated by measurements. An analytical electromagnetic transmission model of the wall is developed to optimize signal transmission up to millimeter-wave (mmWave) frequencies, achieving a more than 70 dB improvement over a bare load-bearing wall at 30 GHz. The second part of the thesis focuses on the modeling of antenna-user interactions for mmWave mobile phones. Empirical studies of antenna radiation with a mobile user at mmWave frequencies have largely relied on real humans, lacking repeatability. To address this, a physical human body phantom for mmWave frequencies is presented for the first time, validated by comparing measured spherical coverage of a reference antenna array on a mobile phone-sized chassis against simulations with an accurate numerical human body model. Three different mobile antenna arrays operating at 28 GHz are evaluated using the developed full-body human phantoms to study various antenna array configurations. Finally, photogrammetry is applied to antenna-hand interaction studies for the first time, using 3D hand models of individual users to assess the effects of hand palms and natural grips on the radiation characteristics of 28 GHz mobile phone antennas. The results reveal significant variations in realized gains and radiation efficiency across different user hand models.
  • Curved boundary integral method and its application to Mie theory: Electromagnetic beam synthesis and scattering analysis
    (2024) Lamberg, Joel
     School of Electrical Engineering | Doctoral dissertation (article-based)
    This doctoral thesis presents the development and application of the curved boundary integral method (CBIM) in conjunction with Mie Theory to enhance electromagnetic beam synthesis and scattering analysis, focusing on terahertz (THz) corneal imaging. This research adapts the proposed CBIM to model the interactions of electromagnetic beams with the human eye, aiming to advance non-invasive imaging techniques for the early detection of ocular diseases. The presented theories are scalable to any classical electromagnetism frequency range. The thesis introduces the CBIM, a sophisticated method and computational tool for synthesizing electromagnetic fields from arbitrary source field distributions on compact and regular surfaces. This method approximates beam synthesis using only electric field distributions, neglecting magnetic ones, which is accurate for surfaces with radii of curvature larger than a few wavelengths. The presented method allows precisely manipulating beam properties such as wavefront, amplitude, phase, and polarization directly from the source surface. Subsequently, Mie scattering theory is integrated into the analysis by extending CBIM into a source-free, basis-function based 3D angular spectrum method, enabling synthesized beams to be expanded into vector spherical harmonics. These theoretical advancements enhance electromagnetic field applications in biomedical contexts, particularly within the 0.1-1 THz range, which is well-suited for penetrating 0.5 mm into the human cornea. Simulations and theoretical analyses demonstrate the high accuracy and effectiveness of the CBIM, its extension to the 3D angular spectrum method, and its applications in Mie scattering theory. These methods show potential in biomedical applications and optical engineering. This thesis further explores the application of this methodology in THz corneal spectroscopy, illustrating how wavefront-modified and polarization-optimized vector beams can significantly reduce errors associated with traditional Gaussian beam analysis. Findings could improve the diagnostic capabilities of THz imaging technologies in clinical settings. This work significantly advances the theoretical framework of electromagnetic beam synthesis using CBIM and its modification to the 3D angular spectrum method. It allows for free manipulation of the incident field by its wavefront, amplitude, phase, and polarization distribution, showcasing the practical implications of these methods in enhancing the resolution and diagnostic accuracy of THz corneal spectroscopy and contributing significantly to the early detection and monitoring of ocular diseases.
  • Enabling sustainable and cost-efficient semi-autonomous forest machine chain - Modeling, estimation and control for autonomous driving in terrain
    (2024) Badar, Tabish
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Traditionally, two humans operate the existing cut-to-length (CTL) forest machine chain, which includes a harvester and a forwarder. The harvester fells and cuts the trees into logs, whereas the forwarder carries the CTL logs to transportation sites. A fully loaded forwarder risks damaging the soft forest terrain. In addition, the rollover of forwarders is a real risk. The motivation for the Autologger project was to introduce a novel forest machine chain concept to raise its productivity while minimizing terrain damage. This thesis aimed to study and develop smart harvester and autonomous forwarder functions. The purpose of the smart harvester is to build an initial three-dimensional (3D) model of the driving path. The autonomous forwarder, in turn, tracks the shown path, utilizing a 3D terrain model while avoiding vehicle rollover. Two articles focus on estimating the 3D form of the solid path. The ground height was estimated without relying on a camera or LiDAR. The four papers focus on building vehicle models incorporating a 3D terrain model for autonomous driving in terrain. The vehicle model was suitable for exact non-linear vehicle simulations, state estimation, and nonlinear model predictive control (NMPC)-based 3D motion control with rollover avoidance.The solution to the smart harvester problem was to measure the wheel heights. The height-odometry algorithm measures the height profile of the path using wheel height measurements, the vehicle's attitude data, and its geometry. The aided height-odometry method filters the biases and errors from the height-odometry output using a priori 3D terrain map. The solution to the autonomous forwarder problem was to utilize a six-degrees-of-freedom (6-DOF) vehicle model to simulate the dynamics of the off-road vehicles, as it has all the necessary components, i.e., forces and moments. A linear tire force model was adapted in the 6-DOF vehicle simulations, assuming the vehicle operates in the primary handling regime. The constituent force models were modified to include the 3D map information. The 6-DOF dynamical model for car-like vehicles was extended to center-articulated vehicles with 1-DOF articulation using a combined center of gravity (CG) approach. The vehicle simulator contributed to devising system calibration procedures, identifying actuator dynamics, and quantifying sensor delays. The simulations facilitated the development of a continuous-discrete extended Kalman filter (CDEKF) for state estimation, designing NMPC for 3D motion control, and studying rollover avoidance. Polaris (a terrain car) was used as a case study to validate the (aided) height-odometry method(s) and augmented 6-DOF vehicle model through various experiments. The estimated wheel heights followed the ground truth within a few centimeters. Stable state estimates were obtained even with erroneous satellite navigation data in the forest. The real-time NMPC-based 3D motion control was ultimately demonstrated on the university's campus.
  • A multifrequency view on the characteristics and evolution of narrow-line Seyfert 1 galaxies
    (2024) Varglund, Irene
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Narrow-line Seyfert 1 (NLS1) galaxies are a peculiar type of active galactic nuclei (AGN). These sources are identified by their narrow emission lines and are believed to be in an early evolutionary stage, perhaps in their first activity cycle. Originally these sources were thought to have no significant radio emission, but this hypothesis has been proven wrong by the presence of powerful, relativistic jets. The discovery of jets in these sources contradicts both the traditional classification system and the jet paradigm. In this thesis, large samples of NLS1 galaxies have been examined at both radio and optical frequencies by using archival data and data obtained through recent observing time proposals. Due to the difficulty in accurately classifying these sources, the traditional classification system of these sources states that they are incapable of significant radio emission, as well as of hosting powerful relativistic jets, most large samples of NLS1 galaxies are contaminated with other AGN, such as broad-line Seyfert 1 galaxies. Due to this, a study in this thesis focused on obtaining the cleanest large sample of NLS1 galaxies currently available by studying the optical spectrum of 11 001 sources, resulting in roughly 4000 sources deemed as most likely genuine NLS1 galaxies. The host galaxy morphologies of both jetted and non-jetted NLS1 galaxies of both northern and southern NLS1 galaxies were investigated in two separate studies. The aim was to try and decipher whether or not there are any clear differences between these two types of AGN. The results indicate that the predominant host of NLS1 galaxies is disk-like with the jetted and non-jetted sources sharing similar host galaxies. Furthermore, based on these results, major mergers do not seem to correlate with jettedness. In radio, an extensive analysis of the cleanest NLS1 galaxy sample was performed at three different frequencies: 144 MHz, 1.4 GHz, and 3 GHz. Nearly half of the sources were detected in at least one of these frequencies, with the majority of the detections at 144 MHz. Many of these sources present clear AGN activity, with over half of the detections at 3 GHz having a radio luminosity higher than what is typically found in star formation processes. Several compact steep-spectrum sources were also identified. The variability seen in some NLS1 galaxies is unique, with very large flux density changes occurring on shorter-than-expected timescales. Various explanations for the variability have been discussed and deemed as impossible, improbable, and possible. The extraordinary behavior seen in these sources can provide clues on the evolution of them and other AGN.
  • Inductive wireless power transfer systems with high positional freedom
    (2024) Liu, Yining
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Wireless power transfer (WPT) techniques provide opportunities for convenient and safe power delivery and battery charging over a wide variety of applications, ranging from low-power consumer electronics to industrial applications with kilo-Watt power levels. Currently, research in WPT field is very active but still at its early stage of development, where challenges arise at developments of every stage and component of power transfer systems. In addition to the main wireless link with inductively coupled coils, power converters and power switching components are also important topics since they are critical to the overall system efficiency. Development of WPT systems requires interdisciplinary studies including power electronics, electromagnetics, as well as control and optimization.  This dissertation provides an overview of challenges that have appeared during recent developments of WPT techniques and proposes solutions to some of the unsolved ones. In the first half of the dissertation, the typical structure of a single-channel WPT system is introduced stage-by-stage, from which we bring up discussions of challenges in terms of efficiency and losses, position freedom for power transfer, as well as design accuracy and simplicity. Known state-of-the-art research works have provided novel designs of coil structures and converter topologies as solutions to some of these challenges, while new challenges always come along which requires further settlements.  Novel solutions in this thesis are presented in the second half with respect to the three main aspects of challenges: positional freedom, power transfer efficiency, and implementation accuracy. New coil structures and a non-coherent power combining method are proposed to improve the degree of position freedom for wireless power transfer. In terms of challenges brought by increasing operating frequencies, new converter topologies and parameter design approaches are introduced to secure soft switching operations in both the design and implementation phases. Finally, the implementation is discussed at the system level, taking into consideration parasitic effects and mismatches between multiple power stages, and solutions are provided case-by-case. The parasitic effects in the system are either compensated or carefully avoided based on the proposed design guidelines.
  • A Parametric Spatial Audio Compression Codec for Higher-Order Ambisonics
    (2024) Hold, Christoph
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Spatial audio has the potential to revolutionize how we consume music and other audio content by enabling an immersive audio experience. Therefore, the technologyand entertainment industry recently adapted their services and began delivering spatial audio formats. Higher-order Ambisonics (HOA), representing the audio scene in the spherical harmonic domain (SHD), offers various benefits as a spatial audio format, notably the independence of the recording and reproduction setup. However, a critical challenge remains: high-quality spatial audio content is largely inaccessible due to the required number of audio channels and data. Audio codecs can successfully reduce the technical challenges originating from distribution and storage. Despite the demand for high channel-count spatial audio continuing to rise, traditional multichannel codecs fall short of delivering the required performance for HOA. Akin to parametric audio coding, model-based parametric spatial audio techniques can be adapted for perceptual spatial audio coding. Model-based spatial audio techniques may parameterize the input scene in a perceptually meaningful and compact way. The input scene parameterization allows signal-dependent processing such as directional optimizations and informed upmixing, overcoming typical challenges of signal-independent processing. This work proposes a spatial audio codec for HOA using parametric Directional Audio Coding (DirAC). First, a modified spherical harmonic transform strategy is developed that enables analysis, modification, and reconstruction of HOA signals. The following study explores a compression strategy achieving perfect reconstruction of low-order SHD components and parameterized resynthesis of higher-order SHD components, establishing the perceptual effectiveness of this duality. Furthermore, SHD post-processing is derived that leverages the input parameterization to improve the codec output by matching to target signal properties. Finally, this work introduces a HOA audio codec based on the aforementioned theoretical foundations. The experimental results demonstrate significant improvements over traditional multi-channel audio codecs, highlighting the potential of the proposed codec to deliver high-quality spatial audio, advocating for including input parameterization side-information in order to avoid coding excessive channel-counts. The implemented codec achieves excellent perceptual quality ratings while reducing the transport data to only a few percent of the input audio data.In conclusion, this research advances the state of the art in spatial audio coding and yields further development in spatial audio codecs for delivering HOA, making the HOA format and its benefits more accessible, thus enabling wider adoption in various media applications.
  • Improved transcranial magnetic stimulation protocols to locate brain activations
    (2024) Matilainen, Noora
     School of Electrical Engineering | Doctoral dissertation (article-based)
    Transcranial magnetic stimulation (TMS) is a non-invasive neurostimulation technique used in clinical treatment and research. It is a technique that provides essential information about brain activity and function, as well as effective treatment for certain neurological disorders. The use of TMS is however still limited by several fundamental uncertainties. For example, it remains uncertain which forms of stimulation are required to elicit specific responses. In addition, the TMS procedure itself can be time consuming and is prone to errors. This summary offers new knowledge of how TMS parameters affect neurostimulation and what they stimulate. Publication I examines the effect of the TMS inter-pulse interval (IPI) on motor evoked potential (MEP) amplitude in active and resting muscles. Previous research has shown that MEP amplitudes are significantly influenced by IPI in resting muscles, with shorter intervals generally leading to decreased amplitudes. This study, however, reveals that active muscle contraction during TMS eliminates the modulating effect of IPI, allowing the use of shorter IPIs which speeds up TMS procedures. Publication II investigates the accuracy of a three-point navigated TMS, still a commonly used approach for neuronavigation. The findings reveal that errors in landmark pointing can significantly impact the accuracy of coil positioning and the induced electric field, highlighting the importance of minimizing such errors in TMS research. Publication III explores the use of computational dosimetry to predict the optimal coil positioning and to estimate motor threshold values in TMS. While the study shows promising results in predicting optimal coil locations, the accuracy of predicting hotspots is slightly less than the hypothetical target of 1 cm. Nevertheless, the method is possibly useful in clinical practise, offering potential improvements in the speed and reliability of TMS hotspot-finding procedures. Publication IV contributes to TMS localization and investigates the differences between posteroanterior (PA) and anteroposterior (AP) coil current directions. The study suggests that PA-TMS primarily activates the precentral gyrus, while AP-TMS is more likely to activate the postcentral gyrus, with both directions showing a higher likelihood of white matter activation. Together, these four studies contribute to a deeper understanding of TMS mechanisms, the optimization of stimulation protocols, and improved accuracy in TMS procedures, with implications for both research and clinical applications.
  • Fabrication and characterization of two-dimensional material based devices for photonics and electronics
    (2024) Uddin, MD Gius
     School of Electrical Engineering | Doctoral dissertation (article-based)
    This thesis explores the potential of two-dimensional (2D) materials in different practical applications and presents the results divided in three parts. The first part focuses on the miniaturized spectrometers. Unlike conventional tabletop spectrometers, we demonstrate miniaturized (~22×8 μm2) computational spectrometers that rely on the electrically tunable spectral response of 2D materials-based single-junction for spectral reconstruction. We achieve high peak wavelength accuracy (~3 nm) and a broad operation window covering the visible and the near-infrared regions, indicating the great potential of the spectrometers to enable numerous portable applications.     The second part of this thesis examines different strategies for tuning the optical and electrical properties of 2D materials. We demonstrate that morphological manipulation of 2D indium selenide (InSe) facilitates enhanced light-matter interaction in InSe. Our 2D InSe/1D nanowire heterostructures, exhibit more than 5 times enhanced optical responses compared to that from bare InSe. Moreover, significant optical anisotropy is observed that makes our mixed-dimensional heterostructures a good candidate for diverse polarization-dependent optoelectronic applications such as photodetectors. Further, in this thesis, we explore a strain engineering approach to increase the charge carrier mobility of molybdenum ditelluride (MoTe₂) field-effect transistors (FETs). It involves the creation of hole arrays in the substrate, transfer of MoTe2 flakes on the hole arrays, and subsequent deposition of ALD Al2O3 passivation layer on top of the MoTe2 flakes. We achieve ~6 times higher charge carrier mobility in the strained MoTe2 FETs than those MoTe2 FETs without strain. The results offer a bright prospect to realize 2D materials-based high-performance devices for future electronics.   In the final part of this thesis, we experimentally demonstrate a novel concept for the miniaturization of broadband light sources. Coherent broadband light is generated (via difference-frequency generation) for the first time with gallium selenide and niobium oxide diiodide crystals at the deep-subwavelength thickness (<100 nm). The broadband spectrum spans more than an octave (from ~565 to 1906 nm) without the need for dispersion engineering. Compared with conventional methods, our demonstration is ~5 orders of magnitude thinner and requires ~3 orders of magnitude lower excitation power. The results open a new path to create ultra-compact, on-chip broadband light sources.