Piezoelectric Energy Harvesting (PEH) offers a viable and sustainable solution for powering low-power electronic systems in hard-to-reach or maintenance-intensive environments, such as those found in aviation. The focus of this thesis lies on the circuit design aspect of the system, encompassing harvesting efficiency, regulation, and adaptive switching logic. The results validate that the proposed design fulfills the energy demands of a low-power wireless sensor, even under irregular excitation patterns. A range of advanced rectifierless interface topologies are investigated, including Rectifierless Synchronous Electric Charge Extraction (ReL-SECE), Rectifierless Synchronized Switch Harvesting on Inductor (ReL-SSHI), and Parallel-SSHI (P-SSHI). These non-linear techniques are selected for their ability to maximize power output, eliminate diode thresholds, and dynamically adapt to varying load and vibration conditions, significantly improving power transfer compared to traditional full-bridge rectifiers. Energy storage is realized through the use capacitors due to their rapid charge/discharge characteristics and long operational life. To ensure circuit safety and longevity, an over-voltage protection mechanism is implemented using a shunt regulator in combination with a resistive voltage divider. Additionally, a dedicated voltage regulation stage is incorporated to maintain a stable 3.3 V DC output, necessary for reliable operation of downstream wireless transmitters and onboard sensor systems. The proposed system is designed to be energy-autonomous, capable of cold-start operation, and optimized for long-term deployment without the need for external maintenance or battery replacement. The approach aligns with current trends in micro-energy harvesting (MEH) for IoT applications, emphasizing compactness, adaptability, and robustness under real-world conditions. Overall, the system demonstrates the practical potential of intelligent, self-powered harvesting circuits in enabling sustainable sensor networks and instrumentation in modern aircraft environments.

In the last decade, the new functional neuroimaging technique functional ultrasound (fUS) has emerged as a potential new tool for clinical and neuroscientific applications. Unlike several conventional methods for functional brain imaging, fUS offers an unparalleled combination of submillimeter-subsecond spatiotemporal resolution and the ability to penetrate deep into brain tissue and capture large areas of interest. This makes fUS an exceptionally valuable tool for investigating brain function. The principal physiological mechanism utilized by fUS imaging is neurovascular coupling, which connects neuronal activity with local hemodynamic changes. Within the brain, increased neuronal activity results in the dilation of nearby blood vessels, subsequently leading to an increased local blood flow and volume.
In fUS, the increase in blood volume is measured by the intensity of the Doppler signal caused by the Doppler effect. To obtain relevant information from functional neuroimaging data, blind source separation (BSS) is often used. BSS is the separation of a set of source signals from a set of mixed signals, without the aid of information about the source signals or the mixing process. A commonly used BSS method for identifying brain networks and artifacts in functional neuroimaging techniques like fMRI is independent component analysis, which is a low-rank matrix decomposition. This work explores the use of canonical polyadic decomposition (CPD), a low-rank tensor decomposition, for BSS of fUS data. The CPD approximates a 3rd-order tensor, consisting of a space, frequency, and time dimension, by a sum of rank-1 tensors. The proposed method offers three key advantages that make it highly relevant to the field of fUS signal processing: 1) Tensor-based BSS by CPD allows for frequency as a third dimension to complement the spatial and temporal dimensions; 2) Tensor-based BSS by CPD allows for the use of all raw fUS data (compound images) rather than only power doppler images, exploiting more available information; and 3) Tensor-based BSS by CPD allows for a BSS method without strong constraints such as the case with matrix-based methods like independent component analysis. The full processing pipeline developed in this work consists of four distinct stages. In stage one, pre-processing is carried out through SVD filtering, temporal demeaning, and pixel-based normalization. During the SVD filtering, normalization of the singular values is performed, which is novel. In stage two, compression is performed using truncated multilinear singular value decomposition. In stage three, BSS is carried out using CPD on the compressed data. In stage four, stable components are extracted by running the CPD 100 times and clustering the resulting components. These clusters were then averaged to create mean components. The CPD rank was estimated based on two quantifiable aspects: 1) Correlation between mean components and 2) Frequency of occurrence of mean components. The method is entirely data-driven, can be applied to entire raw fUS datasets, and can run on a laptop with standard RAM. Application to task experiment data of a mouse verified that activity that is temporally correlated with the stimulus can be extracted in expected regions. The components also indicate expected functional connectivity, which is often not the case for matrix-based methods. Moreover, frequency spectra showed different characteristics for different types of components, which displays the relevance of this dimension for BSS, especially for nuisance components. In conclusion, the proposed tensor-based blind source separation pipeline for raw fUS data was able to identify artifacts and meaningful neurological components based on distinctive characteristics in the temporal, spatial, and spectral domains. This work serves as a starting point for more advanced denoising, the identification of novel brain networks, and the comparison of brain networks across healthy and pathological conditions. Nevertheless, further research is necessary to ensure the utility of the method.

A deeper understanding of Multiple Sclerosis (MS) symptom progression is required for diagnostic accuracy and patient care. Remote monitoring through smartphones can provide continuous insights in the well-being of MS patients. This research aims to explore differences between MS and healthy typing behavior through collected typing behavior data (keystroke dynamics) on smartphones. The main contribution of this work was finding patterns within typing data that distinguish healthy typing behavior from MS with unlabeled time series data, except the diagnosis. A dataset of three combined studies was provided, which included 182 patients suffering from MS and 335 healthy subjects. An autoencoder was fitted on healthy samples to learn the temporal patterns in healthy typing behavior. Then, an anomaly detection method leveraged both the reconstruction loss and embeddings of these samples to determine whether MS samples deviated from healthy behavior. MS subjects were evaluated against healthy subjects through the ROC-AUC on both sample level and subject level. An initial hold out dataset of 20\% of the subjects was left out for validation. The overall performance was poor (AUC 0.6-0.8), because we aimed to find one general method for all MS subjects. Key limitations were the large variance in healthy typing behavior, relatively short window lengths, and difficulty in quantification of the training step of the autoencoder. Despite the difficulty of evaluating the performance of our suggested method, the described methods provide interesting monitoring possibilities for neurologists to enhance MS diagnosis. These monitoring possibilities could extend towards monitoring the effect of treatment, which is the first step towards personalised healthcare in MS.

Autonomic imbalance, characterized by suppressed vagal activity and increased sympathetic activity significantly contribute to the development and progression of cardiovascular diseases. A non- invasive neuromodulation technique that may influence the cardiac autonomic nervous system (CANS) and restore autonomic imbalance is transcutaneous vagus nerve stimulation (tVNS). This thesis focuses on a novel cervical tVNS device (Pulsetto) which targets the vagus nerve through the neck. The aim of this research is to investigate the efficacy of this new device and provide insights into how cervical tVNS influences the CANS.

Two experiments were conducted: the first explored cervical tVNS in 8 atrial fibrillation (AF) patients, while the second involved 40 healthy participants, randomly assigned to either a stimulation group (n=30) or a sham group (n=10). Participants in the stimulation group received 10 minutes of stimulation. Heart rate variability (HRV) and cardiac conduction were measured via a 3-lead ECG, with data analysis focusing on HRV parameters, conduction intervals, and wave amplitude detection.

Significant HRV changes were observed during stimulation compared to pre-stimulation. Cervical tVNS significantly decreased mean HR (P<0.001) and LF/HF (P=0.038), while significantly increasing RMSSD (P=0.001), PNN50 (P=0.001) and HF power (P=0.003). Additionally, the QT interval and T-wave amplitude significantly increased (P=0.001 and P=0.030 respectively) in the stimulation group. None of these parameters changed in the sham group.

This thesis provides evidence that cervical tVNS can modulate cardiovascular autonomic control in healthy participants by increasing parasympathetic activity. Additionally, it is the first study to observe an increased T-wave amplitude during cervical tVNS, suggesting a novel effect on ventricular conduction. These insights indicate that cervical tVNS holds great potential for treating arrhythmias and other cardiovascular diseases.

This Bachelor of Science thesis presents the development of a portable readout system for a graphenebased gas sensor array, aiming to bring advanced gas sensing technology from a controlled laboratory environment to practical field applications, such as greenhouses and vineyards. The project focuses on integrating various hardware components into a single printed circuit board (PCB) equipped with a microcontroller, ensuring accurate resistance measurement of graphene strips, generating programmable waveforms for micro hotplate control, and facilitating data communication via USB and Modbus interfaces. The design includes high-resolution analog-to-digital converters (ADCs) for precise current and voltage measurements, a digital-to-analog converter (DAC) for waveform generation, and a robust power management system supporting multiple power inputs. The system supports dual operation modes: interfacing with a desktop for laboratory use and functioning autonomously as an integrated sensor unit. Key challenges addressed include ensuring measurement accuracy, reliable sensor array connectivity, and user-friendly interfacing for both lab and field scenarios. Comprehensive test plans are outlined to verify the system’s performance, focusing on critical requirements such as resistance measurement accuracy, waveform generation, and communication interfaces. The successful implementation of a large part of this readout system demonstrates its potential for enhancing gas detection and monitoring in agricultural applications, providing a scalable solution for future deployment and development.

Climate change poses a serious threat to ecosystems and increases the need for accurate and rigorous monitoring of ecosystems. Current monitoring solutions are often bulky, expensive, and lack critical functionalities such as on-board inference capabilities, robust wireless connections, and a diverse sensor suite. Ecological monitoring projects often suffer from inefficiencies caused by the large time delays between collecting data and analyzing said data, as well as having to spend large amounts of time in the field setting up the sensors manually. This thesis addresses many of these issues by designing a sensor with an extensive sensor suite, robust wireless capabilities and an on-board audio classifier able to perform real-time inference. Furthermore, attention is paid to making the system extendable in the future and allow for potentially integrating the sensors with a drone delivery- and retrieval system. The system tests performed indicate that the system has great potential given more time to tweak some of its identified shortcomings.

Introduction The Acoustic Change Complex (ACC) is a cortical auditory evoked potential elicited by a change in an ongoing sound that consists of a P1-N1-P2 complex. The ACC holds promise as a non-invasive, passive, and objective measure to monitor speech discrimination. This can improve the treatment of hearing loss for young children and adults that are not able to execute gold standard speech perception tests. However, the reported measurement times of the ACC are lengthy, which might impede the clinical feasibility of measuring young children and adults with behavioural issues that are difficult to test. Measurement time can be reduced by enhancing the signal to noise ratio (SNR). Furthermore, enhancement of the SNR might increase the sensitivity and specificity of the ACC as a measure of speech discrimination. Two stimulus parameters that might increase the SNR are an increase in the pre-transition duration (PTD) and stimulus complexity. The PTD is the duration between the onset of the stimulus and the change in the acoustic characteristics of the stimulus. Besides increasing the SNR, measurement time directly increases with an increase in PTD. Therefore, the optimal PTD regarding efficiency should be studied. Objectives The aims are to study the effect of 1) the PTD and 2) tonal complexity on the N1-P2 peak-to-peak amplitude, baseline noise and SNR of the ACC. To evaluate the optimal PTD, efficiency was included as an outcome measure for the first objective. Methods The ACC was measured in eighteen normal hearing adults to pure tone stimuli with a frequency change from 1kHz to 1.1kHz. The studied PTDs were 0.25, 0.5, 1, 2 and 3s. Furthermore, a complex tone with the same center frequencies and a PTD of 1s was presented. Efficiency was measured as the SNR divided by the total measurement time. Main results An increase in PTD significantly increased the N1-P2 peak-to-peak amplitude up to a PTD of 2s. PTD 0.25s was excluded from all analyses due to overlap with the onset response. The SNR of PTD 0.5s was significantly smaller than all other measured PTDs. The PTD of 1s was significantly more efficient than a PTD of 0.5s, had a measurement time of 6.67min and the ACC was present in 58.8% of the participants. ACC presence increased (to 100% for PTD 3s) with an increase in PTD. There was no significant difference in the N1-P2 peak-to-peak amplitude or SNR between the complex and pure tone. However, ACC presence increased with 22.5 percentage points for the complex tone compared to the pure tone. Conclusion The ACC is affected by the PTD. The PTDs of 2 and 3s generated the significantly largest responses. We recommend utilizing a PTD of 1s instead of 0.5s, as it resulted in a significantly higher N1-P2 peak-to-peak amplitude, SNR and efficiency. Furthermore, increasing tonal complexity or the PTD beyond 1s seems promising to increase presence of the ACC without significantly decreasing the SNR or efficiency.

Brain disorders in children pose significant challenges to their development, impacting cognition, speech, movement, and behavior. The uncertainty surrounding prognostic information at the time of diagnosis leaves families with numerous questions about the future. The Child Brain Lab at Erasmus MC Sophia Children's Hospital conducts IQ, electroencephalogram (EEG), speech, and movement tests in playful environments, enhancing scientific research and healthcare practices for a better understanding of disease progression.

The Otolaryngology department at the Child Brain Lab focuses on auditory-related potentials (ERPs) obtained from EEG measurements to predict the future development of children with brain disorders. Analyzing ERP data from experiments like Mismatch Negativity (MMN) and Acoustic Change Complex (ACC) yields insights into developmental trajectories and connections between hearing, language, and brain development.

This thesis aims to explore alternative methodologies for extracting comprehensive information from ERPs, overcoming limitations of the commonly used peak amplitude and latency analysis. Tensor decompositions are employed to exploit structural information present in the data, using data fusion methods to combine multiple datasets for improved classification and deeper insights into group differences.

Simulations on artificial ERP data demonstrate that data fusion methods perform better on two ERP tensors compared to single tensor decomposition when group differences are shared between datasets. On a real dataset, tensor decompositions show promise for classifying subjects based on auditory event-related potentials while giving more insights into the neurological sources.

This report proposes an alternative method for analyzing ERP data, highlighting the potential of tensor decompositions and data fusion techniques.

Rank detection is crucial in array processing applications, as many algorithms rely on accurately estimating the rank of the data matrix to ensure optimal performance. Under Gaussian white noise, rank can be detected through eigenvalue analysis. However, in arbitrary noise, prewhitening the data matrix with the noise covariance matrix is necessary, and rank detection is achieved by examining the generalized eigenvalues. Existing methods often assume the noise covariance structure or require a large number of noise samples. This thesis focuses on addressing the rank detection problem in scenarios with limited noise samples and arbitrary noise environments.

Firstly, we investigate the largest generalized eigenvalue threshold for the prewhitened data sample covariance matrix according to the random matrix theory. We develop a rank detection algorithm based on the threshold via a sequential test, and provide the performance analysis. A series of simulations demonstrate its superiority over conventional methods such as Minimum Description Length (MDL) and Akaike's Information Criterion (AIC).

Secondly, since the Short-time Fourier Transform (STFT) is commonly used for non-stationary signal analysis, we extend our rank detection method to the STFT domain. The correlations introduced by the STFT have a significant impact on the distribution of the noise. Therefore, we develop a technique to remove correlations among time-frequency bins based on exact expressions of these correlations. After successfully eliminating these correlations, our proposed rank detection method achieves enhanced reliability and performance in the STFT domain.

Lastly, we evaluate the effectiveness of our rank detection method in speech enhancement applications. Simulations confirm that utilizing the estimated rank improves speech quality compared to using the known number of sources.

The work presented in this thesis investigates the creation of virtual sound sources in a room equipped with a limited number of loudspeakers. This limited number of loudspeakers is typical for consumer loudspeaker systems. Ideally, these systems can provide a listening experience in which localisation cues are completely present. However, in current systems, this is not the case. The limited number of loudspeakers make it impractical to produce a physically accurate sound field. A possible solution is to create a perceptually accurate sound field instead. In this work, a step towards an algorithm which can do so is presented. The developed algorithm requires knowledge of the listener placement, the room, and the loudspeaker placement. Spatial weighting is used to construct a region in which the acoustic en-ergy should be large and a region in which the acoustic energy should be limited. The loudspeaker playback signals are obtained by maximising the energy ratio between these regions, while at the same time ensuring that the perceptual difference between the received audio and reference audio remains limited. The algorithm employs a convex optimisation problem to facilitate efficiently solving for the playback signals. Concretely, six convex optimisation problems are proposed with somewhat increasing complexity and different weighting matrices. For each of the optimisation problems, the proposed algorithm is compared against a simple amplitude panning algorithm and a nearest neighbour algorithm. It is found that none of the considered algorithms is clearly preferred over the others in terms of the considered evaluation metrics. A major limitation of the presented work is that the evaluation metrics do not explicitly test for localisation accuracy. In future work, this should be investigated by including subjective tests.

When a person suffers from severe to profound hearing loss, a cochlear implant (CI) can aid in restoring auditory perception and speech comprehension. To obtain good speech comprehension, fitting of a CI to the user’s specific characteristics is crucial. Fitting can be a time-consuming process which requires the cooperation of a CI user and is dependent on the used methods (e.g., T-level measurements). Besides transmitting electrical stimuli, a CI can also record the response of the auditory nerve fibres to a stimulus. This response is called the electrically evoked compound action potential (eCAP). eCAP responses can be measured objectively without a user’s active cooperation and could potentially aid in the fitting of a CI. For this purpose, the eCAP thresholds are of main interest. eCAP thresholds can be determined manually by a clinical specialist, or automatically by an automatic threshold detection method. Automatic eCAP threshold detection can therefore be of aid in a completely objective and uniform CI fitting method. The goal of this study was to compare different automatic eCAP threshold detection methods (in combination with different averaging methods and different artefact reduction methods) based on the precision and accuracy of these methods.
Five different automatic eCAP threshold detection methods have been examined in this study: sigmoid amplitude growth function (AGF), linear AGF, signal-to-noise ratio (SNR), cross-covariance between adjacent levels and cross-covariance with maximum level. The two different averaging methods that have been examined are standard averaging (SA) and FineGrain averaging (FG), the two different artefact reduction methods are alternating polarity (AP) and forward masking (FM). In total, 20 different combinations have been examined. The success rates of these 20 combinations have been determined, threshold confidence intervals (TCIs) were calculated as a measure of precision and the correlations between eCAP thresholds and T-levels were determined as a measure of accuracy of the different (combinations of) methods.
The combination of FG FM resulted in the highest success rates for different threshold detection methods, and the threshold detection method SNR had the overall highest success rates. A two-way ANOVA revealed that both artefact reduction/averaging method and threshold detection method have a significant effect on the TCIs. The combination of FG FM had the best resultsregarding the TCIs, and the sigmoid AGF threshold detection method was the threshold detection method with the lowest mean TCI. A similar two-way ANOVA was performed for the correlation between eCAP thresholds and T-levels, revealing the same results as for the TCIs that both artefact reduction/averaging method and threshold detection method have a significant effect on the correlation coefficients. FG FM was again the best performing combination, and the sigmoid AGF threshold detection method resulted in the highest correlation coefficients.
Based on these results, it can be stated that the FG FM combination for averaging and artefact reduction was the overall best combination. When comparing the different automatic threshold detection methods, the sigmoid AGF method resulted in eCAP thresholds with the highest precision and accuracy. Future research should focus on obtaining more data, further refinements of the different automatic eCAP threshold detection methods and the use of the determined eCAP thresholds in the clinical fitting of a CI.

Introduction: Currently, there is no method available for intra-operative evaluation of the completeness (transmurality and continuity) of surgical ablation lesions. This study aimed to investigate the changes in electrogram characteristics and activation patterns caused by different (non)transmural and (dis)continuous ablation lesions in ex vivo perfused porcine hearts.

Methods: Donation after circulatory death porcine hearts ex vivo perfused in Langendorff mode were used to perform ablation experiments in a controlled setting. Three subsequent radiofrequency ablation lesions – with different degrees of transmurality and continuity - were created on the right ventricle using AtriCure’s Isolator Synergy Bipolar clamp. Electrograms of the lesion and surrounding tissue were recorded by unipolar high-resolution mapping. These measurements were executed during pacing perpendicular to the ablation lesion from two sides and during intrinsic cardiac rhythm. Electrograms were processed using custom-made software. The inter-electrode conduction time, potential voltage, potential slope, and R-to-S-amplitude ratio were analyzed.

Results: The first radiofrequency application significantly affected all parameters in the lesion area. Conduction times increased, the potential voltage and slope decreased, and there was a loss of S-wave amplitude. The increase in conduction time and the decrease in voltage were less steep when there was a conduction gap in the ablation line. However, conduction time was less sensitive to lesion transmurality because it remained stable even when the lesion became more transmural. The potential voltage on the other hand, became significantly lower in transmural lesions, showing an overall decrease of 84% from baseline to the third (complete) lesion. The potential slope showed similar trends as the voltage, although it was less discriminative for (non)transmurality and (dis)continuity. The loss of the S-wave became significantly more pronounced with more radiofrequency delivery.

Conclusions: Complete ablation lesions are characterized by a stable conduction time when applying subsequent ablation, a decrease in potential voltage of 84% on the lesion and its border zone, and loss of the contribution of S-wave amplitude. The combination of these parameters in one tool could help to detect incomplete surgical ablation lesions in the individual patient during Maze surgery. This could potentially reduce post-maze gap-related atrial tachyarrhythmias and thus improve long-term success rates.

The main purpose of a radar is to detect, recognize, and track objects of interest. When it is known that only a single target is present, the matched filter is proven to be optimal detector. However, in practice, a radar scene often consists of multiple targets. For example, in air surveillance and monitoring applications, multiple aircrafts might be in the airspace. When multiple targets are to be detected the matched filter is not guaranteed to give the best results. This can happen when a strong reflector masks the signals reflected from weak reflectors, thereby resulting in missed detections. Furthermore, when the sensor resolution is low, targets that are spaced closely together may only result in a single target actually being detected. This research explores how the Relevance Vector Machine (RVM) framework may be used to achieve a better multi-target detector than the commonly used basic matched filter approach. RVM was selected to resolve the multi-target detection problem as it estimates the target locations iteratively. In this research it was shown how the RVM framework can be used to model the fluctuation of swerling I/II targets. Additionally, the RVM algorithm was modified to incorporate a notion of statistical thresholding. Simulations show that using RVM the false alarm rate can be reduced and target locations can be more accurately recovered compared to other existing methods in case of multiple swerling I/II fluctuating targets. Furthermore, the proposed approach is shown to have a much lower convergence time compared to a similar expectation-maximization based method, namely Enhanced Sparse Bayesian Learning.

To better understand how brain signals are processed and even how the human mind works, analyzing the hemodynamic signal model is one of the most essential steps. In the CUBE group of Erasmus MC, functional ultrasound (fUS) data of a mouse’s brain is recorded. By using this fUS dataset, this thesis will solve the problem regarding the joint estimation of hemodynamic response function (HRF) and the underlying stimulus. Usually, hemodynamic responses are investigated in the time domain, while this thesis provides another perspective
from frequency domain signal processing.

We consider the hemodynamic response as a convolutive signal mixture, then try to transform it into an instantaneous mixing model by converting the context into the frequency domain. By applying independent vector analysis (IVA), this estimation problem can be solved without facing permutation ambiguity which is a well-unknown problem regarding independent component analysis (ICA). Additional steps before and after IVA are also discussed so that a whole estimation road map is formed.

Both simulation and experimental analysis are provided to validate this estimation algorithm. Results show that by using this method, both stimulus and HRF estimation can be achieved satisfyingly in a suitable experimental setting. This thesis provides insights and future potentials for IVA to be further investigated in neural signal processing problems.

In recent years, the large increase in connected devices and the data that is collected by these devices has caused a heightened interest in distributed processing. Many practical distributed networks are of heterogeneous nature. Because of this, algorithms operating within these networks need to be simple, robust against network dynamics and resource efficient. Additionally, if privacy preservation methods are properly implemented, they add to the power of distributed processing by making it possible to leverage the data of many different users, without infringing the privacy of the individuals involved.
In this study we focus on the primal-dual method of multipliers (PDMM), which is a promising dis- tributed optimisation algorithm that seems to be suitable for distributed optimisation in heterogeneous networks. Most theoretical work that can be found in existing literature focuses on synchronous ver- sions of PDMM. However, in heterogeneous networks, asynchronous algorithms are favourable over synchronous algorithms. So far, simulation results have indicated that asynchronous PDMM converges and can even converge in the presence of transmission losses.
In this work we analyse the properties of stochastic PDMM, which is a general framework that can model variations of PDMM such as asynchronous PDMM and PDMM with transmission losses. We build upon previous empirical results of PDMM and formulate theoretical proofs to substantiate these results. After defining stochastic PDMM and proving its convergence, we compare a number of PDMM variations that have been mentioned throughout the literature. Lastly, we derive a lower bound for the variance of the auxiliary variable in the context of stochastic PDMM, assuming uniform updating probabilities. This lower bound indicates that subspace based privacy preservation is applicable to certain instances of stochastic PDMM, like asynchronous PDMM.
The main result of this work is a theoretical proof that shows that stochastic PDMM converges almost surely if the updating probabilities of each auxiliary variable are nonzero. Two important conclusions that follow from this proof are the almost sure convergence of asynchronous PDMM and unicast PDMM with transmission losses. Another useful result is the fact that subspace based privacy preservation is effective when using asynchronous PDMM.

Outlier detection in time series has important applications in a wide variety of fields, such as patient health, weather forecasting, and cyber security. Unfortunately, outlier detection in time series data poses many challenges, making it difficult to establish an accurate and efficient detection method. In this thesis, we propose the Random Projection Outlier Ensemble (RPOE) method. The ensemble combines the results of a diverse set of components allowing it to accurately detect the different types of outliers. The components are based on the most efficient reconstruction-based method, the Random Projection (RP) method, which we enhance with a sliding window to better capture the dependencies of the time series and highlight context-dependent outliers. As a result, the RPOE method performs competitively when tested on different datasets. Moreover, the efficiency of the base detectors allows the RPOE method to have a remarkable run-time. Thus, with the RPOE method, we believe to have taken a step forward in the search for an effective outlier detection method.

Currently, trained machine learning models are readily available, but their training data might not be (for example due to privacy reasons). This thesis investigates how pre-trained models can be combined for performance on all their source domains, without access to data. This problem is formulated as a Multiple-Source Domain Adaptation (MSA) problem setting, where models trained on source domains are combined so that the combiner is robust to application on any unknown target domain. This thesis explores the MSA setting and presents a perspective on MSA theory from literature. The issue in the MSA setting is that target models are not robust in general, leading to negative transfer. Firstly, this issue is illustrated by example of the source models and linearly weighted combinations of the source models. Next, existing theory that guarantees the existence of a robust model is investigated. It is argued that a performance bound has the potential to be extended from the perspective of additional knowledge: in addition to the in the MSA setting available source models, some additional knowledge of the source domains might be used by the model. Existing MSA theory’s assumptions are clarified and the theory is split in two. One half is inherent to the MSA setting and guarantees a model with as robustness property the loss on a matching mixture. The other half assumes a combiner that uses additional knowledge of the source domains, for which the robustness is proven satisfactory. Finally, it is investigated what makes additional knowledge in the MSA setting useful. Current literature assumes a specific target model--the distribution-weighted (DW) combiner--that is viewed as using the sources' joint distributions as additional knowledge. It is argued that knowledge of the training process of the source models can also be used as additional knowledge. In conclusion, this thesis discusses how robustness in the MSA setting can be improved from that of the source models by basing the combiner on additional knowledge of the source domains.

This project develops and tests algorithms for joint signal processing of data from two radars located on the rooftop of EWI (PARSAX and MESEWI). The particular tasks consist of automatic alignment of radar data in space (2D map) and time by observing moving targets of opportunity in the high-resolution mode. After the data alignment procedure, the detection algorithms for optimal fusing of dual-polarization and multi-frequency data are proposed. The detection results are considered the input for moving target (an auto) tracking and its signature extraction. The developed techniques were tested on the data records in experimental scenarios.

Introduction: Potentials measured at the epicardial surface contain information regarding the conductive properties of the atrial tissue. The current lack of morphological categorization during atrial fibrillation (AF) provokes the usage of unsupervised learning methods to evaluate time series electrograms across the atrial surface. Analysis of regional potential morphology increases insight in individual tissue characteristics, possibly aiding in future selection of patients with AF benefitting from additional treatment with.

Methods: single potentials (SP) are extracted from four defined regions: right atrium (RA), left atrium (LA), pulmonary veins (PV) and Bachmann’s bundle (BB). The methods are divided in two parts. In part one, the clusters are determined. Choice of algorithm, evaluation of distance metrics and evaluation in reduced dimensionality are performed sequentially to determine the eventual clustering setup. The clustering setup is repeated for different types of input data, originating from different regional structures (all four regions, solely LA and solely the RA) to avoid deterioration of cluster formation from single region(s). In the second part, the clusters are adapted to evaluate morphological characteristics in the included regions.

Results: Data from a total of 23 patients was used, containing 128 files and 468588 SPs. K-means using the first 8 principal components was deemed the most suitable clustering method with the current data, identifying a total of five clusters. The clusters showed evident different morphology, mostly distinguishable by RS-ratio and potential duration. RA showed a predominant S to an RS-wave pattern, comparable to PV. Both showed similar contribution of fast and slow S-wave morphology. LA showed a RS to R-wave morphology. The cluster showing a slow S-wave morphology was substantially increased across BB compared to other regions but also when compared to other clusters in BB.

Conclusion: Looking at spatial cluster occurrence through an RS-ratio perspective, the beforehand expected deviation from AF compared to sinus rhythm (SR) was not observed. The observed pattern during AF suggests more elaborate contribution of atrial tissue than often targeted in current daily practice. The methodology and defined groups of potentials facilitate evaluation of potential morphology at an individual level. Translation to an endocardial perspective could assist in selecting patients eligible for additional treatment when technologically feasible. Future research should be focused on including potentials containing multiple deflections and estimation of progression state of the atrial tissue.

In recent decades, the field of autonomous driving has witnessed rapid development, benefiting from the development of artificial intelligence-related technologies such as machine learning. Autonomous perception in driving is a key challenge, in which multi-sensor fusion is a common feature. Due to the high resolution and rich information, the camera is one of the core perceptual sensor in autonomous systems. However, the camera provides no knowledge on distance (or depth), which is insufficient for the requirements of autonomous driving. On the other hand, LiDAR provides accurate distance measurements, however the information is sparse. The complementary characteristics of cameras and LiDAR have been exploited over the past decade for autonomous navigation. In order to be able to fuse the camera and LiDAR sensor system jointly, an efficient and accurate calibration process between sensors is essential. Conventional methods for calibrating the camera and LIDAR rely on deploying artificial objects, e.g., checkerboard, on the field. Given the impracticality of such solutions, targetless calibration solutions have been proposed over the past years, which require no human intervention and are readily applicable for various autonomous systems, e.g., automotive, drones, rovers, and robots.

In this thesis, we review and analyze several classic targetless calibration schemes. Based on some of their shortcomings, a new multi-feature workflow called MulFEA (Multi-Feature Edge Alignment) is proposed. MulFEA uses the cylindrical projection method to transform the 3D-2D calibration problem into a 2D-2D calibration problem and exploits a variety of LiDAR feature information to supplement the scarce LiDAR point cloud boundaries to achieve higher features similarity compared to camera images. In addition, a feature matching function with a precision factor is designed to improve the smoothness of the objective function solution space and reduce local optima. Our results are validated using the open-source KITTI dataset, and we compare our results with several existing targetless calibration methods. In many different types of roadway environments, our algorithm provides more reliable results regarding the shape of the objective function in the 6-DOF space, which is more conducive for the optimization algorithms to solve. In the end, we also analyze the shortcomings of our proposed solutions and put forward a prospect for future research in the field of joint camera-Lidar calibration algorithms.