The purpose of this thesis is to emphasize the lacking areas in the field of direction of arrival estimation and to propose building blocks for continued solution development in the area. A review of current methods are discussed and their pitfalls are emphasized. DOA estimators are compared to each other for usage on a conformal microphone array which receives impulsive, wideband signals. Further, many DOA estimators rely on the number of source signals prior to DOA estimation. Though techniques exist to achieve this, they lack robustness to estimate for certain signal types, particularly in the case where multiple radar targets exist in the same range bin. A deep neural network approach is proposed and evaluated for this particular case. The studies detailed in this thesis are specific to acoustic and radar applications for DOA estimation.

abstract: Research on developing new algorithms to improve information on brain functionality and structure is ongoing. Studying neural activity through dipole source localization with electroencephalography (EEG) and magnetoencephalography (MEG) sensor measurements can lead to diagnosis and treatment of a brain disorder and can also identify the area of the brain from where the disorder has originated. Designing advanced localization algorithms that can adapt to environmental changes is considered a significant shift from manual diagnosis which is based on the knowledge and observation of the doctor, to an adaptive and improved brain disorder diagnosis as these algorithms can track activities that might not be noticed by the human eye. An important consideration of these localization algorithms, however, is to try and minimize the overall power consumption in order to improve the study and treatment of brain disorders. This thesis considers the problem of estimating dynamic parameters of neural dipole sources while minimizing the system's overall power consumption; this is achieved by minimizing the number of EEG/MEG measurements sensors without a loss in estimation performance accuracy. As the EEG/MEG measurements models are related non-linearity to the dipole source locations and moments, these dynamic parameters can be estimated using sequential Monte Carlo methods such as particle filtering. Due to the large number of sensors required to record EEG/MEG Measurements for use in the particle filter, over long period recordings, a large amounts of power is required for storage and transmission. In order to reduce the overall power consumption, two methods are proposed. The first method used the predicted mean square estimation error as the performance metric under the constraint of a maximum power consumption. The performance metric of the second method uses the distance between the location of the sensors and the location estimate of the dipole source at the previous time step; this sensor scheduling scheme results in maximizing the overall signal-to-noise ratio. The performance of both methods is demonstrated using simulated data, and both methods show that they can provide good estimation results with significant reduction in the number of activated sensors at each time step. / Dissertation/Thesis / M.S. Electrical Engineering 2012

Nesta dissertaÃÃo sÃo propostos mÃtodos baseados em processamento tensorial de sinais para a estimaÃÃo de parÃmetros em arranjos de antenas vetoriais elÃtricas (Tripolo), considerando diferentes estruturas de arranjos (ULA, L-shape e UPA). Inicialmente, à utilizado um arranjo em L-shape,resultando em um modelo tensorial de terceira ordem (3-D) que, junto ao algoritmo de estimaÃÃo T-ALS (do inglÃs, Trilinear Alternating Least Squares), possibilita a identificaÃÃo cega de pelo menos o dobro de fontes estimadas pelos modelos tradicionais. Em seguida, sob transmissÃo supervisionada, à proposto um mÃtodo alternativo, utilizando a decomposiÃÃo SVD, o qual à comparado ao mÃtodo tensorial com uso do algoritmo T-ALS. Uma segunda abordagem à proposta utilizando-se uma estrutura de arranjo planar de antenas (UPA), a qual faz uso de um modelo tensorial de quarta ordem (4-D) junto ao algoritmo de estimaÃÃo Q-ALS (do inglÃs, Quadrilinear Alternating Least Squares). Neste caso, um mÃtodo alternativo à proposto usando a fatoraÃÃo do produto de Khatri-Rao e uma anÃlise comparativa destes mÃtodos à realizada. Considerando-se o caso supervisionado, à feito ainda um estudo comparativo dos algoritmos Q-ALS, T-ALS e SVD, e um novo algoritmo, chamado Nested-SVD à proposto. Por fim, foi realizada a modelagem computacional do tripolo elÃtrico com uso de software de simulaÃÃo de alta frequÃncia (HFSS), possibilitando a extraÃÃo do parÃmetro de ganho espacial dos arranjos L-shape e UPA. Em seguida, à feita a avaliaÃÃo do desempenho dos mÃtodos tensoriais propostos usando este parÃmetro em uma situaÃÃo mais realista, e comparado ao desempenho usando modelos idealizados de arranjos de antenas com ganho unitÃrio e omnidirecional. O desempenho dos mÃtodos propostos à avaliado atravÃs de simulaÃÃes de Monte Carlo em diferentes cenÃrios e configuraÃÃes de arranjo / In this dissertation, we propose methods based on tensor signal processing for the parameter estimation in electric vector (Tripole) antenna arrays, considering different structures of arrays (ULA, L-shape and UPA). Initially, using a L-shape array, we develop a third order (3-D) tensor model for the received data. Based on this model, a trilinear alternating least squares (T-ALS) algorithm is used for the blind estimation of the sourceâs parameters. Then, under supervised transmission an alternative method is proposed by resorting to the SVD decomposition, which is compared to the T-ALS algorithm. A second approach is proposed, which is based on a uniform planar array antenna (UPA). In this case a fourth-order (4-D) tensor model is obtained, and the Q-ALS (Quadrilinear Alternating Least Squares) algorithm is used for parameter estimation. An alternative method is also proposed, which exploits the factorization of the Khatri-Rao product. Considering the supervised case, a new algorithm called Nested-SVD is proposed and a comparative study with Q-ALS, T-ALS and SVD algorithms is carried out. The performance of the proposed methods is evaluated through Monte Carlo simulations in different scenarios and array settings. Finally, computational modeling of electric tripole using the high frequency simulation software (HFSS) was performed, enabling the extraction of the L-shape and UPA spatial array gain. Then, the performance of the proposed tensor methods is evaluated in a more realistic scenario, and compared to idealized omnidirectional and unitary gain antenna array models

This thesis considers source localization and tracking when both the signal propagation model and the source motion dynamics are unknown. Algorithms are developed for different scenarios. The algorithms are discussed when a source is stationary or mobile, under the condition when sensors are fixed or mobile. These algorithms exploit the strictly decreasing properties of the model in terms of distance, but do not depend on the form and the values of the models. Therefore, these algorithms could be applied when the signal propagation models and the source motion are unknown. The only assumption made is that the signal propagation strength decreases in distance. For a given performance specification, the optimal number and placement of the sensors is also discussed. Convergence and other properties of the algorithms are established under various noise assumptions.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Techniques for estimating unknown terms - such as eddy diffusivity and counter-gradient - in atmospheric flow are presented in this study. The methos is also used to identify the source term in atmospheric pollution. The scheme adopted is based on inverse problem methodology. The inverse problem is formulated as a non-linear optimization problem, where the objective function is defined as the square difference between observational data and data from a transport mathematical model. For estimating the properties of the atmospheric flux, an implicit strategy was used, and an Eulerian model was used as forward model. The estimation of the pollutant source term was tested employing a source-repector technique. In the pollutant dispersion simulation, a Lagrangian model was applied. For some inversions, regularized solutions should be searched. The Tikhonov and entropy regularizations were considered. Three different optimization methods were used: Levenberg-Marquardt, quasi-Newton (deterministic) e simulated annealing (stochastic). The results show a good performance of the proposed methodology in many tested situations. / Neste estudo apresentam-se técnicas para estimar termos desconhecidos em fluxos atmosféricos, tais como coeficiente de difusão turbulento e termo de contra-gradiente. O método é também usado para a estimação de termos de fonte em poluição atmosférica. O esquema adotado é baseado na metodologia de problema inverso. O problema inverso é formulado como um problema de otimização não linear, onde a função objetivo é definida como a diferença quadrática entre dados experimentais e dados obtidos através de um modelo matemático de transporte. Para a estimação de propriedades de fluxo atmosféricos, uma estratégia de inversão implícita foi utilizada, onde um modelo eureliano foi empregado como modelo matemático. Para a estimação do termo de fonte de poluição, um procedimento de inversão empregando a técnica fonte-receptor foi testado em diversos cenários físicos. Para simulação da dispersão de poluentes na atmosfera foi empregado um modelo lagrangeano. Em alguns casos tornou-se necessário aplicar técnicas de regularização na obtenção da solução inversa. Regularização de Tikhonov e em entrópicas foram empregadas, quando necessário. Três diferentes métodos de otimização são utilizados: Levenberg-Marquardt e quase-Newton (determinísticos) e recozimento simulado (estocástico). Os resultados comprovam a robustez do método de inversão nas diversas situações testadas.

The current thesis addresses some of the unresolved predictions of recent models of the semantic brain system, such as the hub-and-spokes model. In particular, we tackle different aspects of the hypothesis that a widespread network of interacting heteromodal (hub(s)) and unimodal (spokes) cortices underlie semantic cognition. For this purpose, we use connectivity analyses, measures of graph theory and permutation-based statistics with source reconstructed Electro-/MagnetoEncephaloGraphy (EEG/MEG) data in order to track dynamic modulations of activity and connectivity within the semantic networks while a concept unfolds in the brain. Moreover, in order to obtain more accurate connectivity estimates of the semantic networks, we propose novel methods for some of the challenges associated with EEG/MEG connectivity analysis in source space. We utilised data-driven analyses of EEG/MEG recordings of visual word recognition paradigms and found that: 1) Bilateral Anterior Temporal Lobes (ATLs) acted as potential processor hubs for higher-level abstract representation of concepts. This was reflected in modulations of activity by multiple contrasts of semantic variables; 2) ATL and Angular Gyrus (AG) acted as potential integrator hubs for integration of information produced in distributed semantic areas. This was observed using Dynamic Causal Modelling of connectivity among the main left-hemispheric candidate hubs and modulations of functional connectivity of ATL and AG to semantic spokes by word concreteness. Furthermore, examining whole-brain connectomes using measures of graph theory revealed modules in the right ATL and parietal cortex as global hubs; 3) Brain oscillations associated with perception and action in low-level cortices, in particular Alpha and Gamma rhythms, were modulated in response to words with those sensory-motor attributes in the corresponding spokes, shedding light on the mechanism of semantic representations in spokes; 4) Three types of hub-hub, hub-spoke and spoke-spoke connectivity were found to underlie dynamic semantic graphs. Importantly, these results were obtained using novel approaches proposed to address two challenges associated with EEG/MEG connectivity. Firstly, in order to find the most suitable of several connectivity metrics, we utilised principal component analysis (PCA) to find commonalities and differences of those methods when applied to a dataset and identified the most suitable metric based on the maximum explained variance. Secondly, reconstruction of EEG/MEG connectomes using anatomical or fMRI-based parcellations can be significantly contaminated by spurious leakage-induced connections in source space. We, therefore, utilised cross-talk functions in order to optimise the number, size and locations of cortical parcels, obtaining EEG/MEG-adaptive parcellations. In summary, this thesis proposes approaches for optimising EEG/MEG connectivity analyses and applies them to provide the first empirical evidence regarding some of the core predictions of the hub-and-spokes model. The key findings support the general framework of the hub(s)-and-spokes, but also suggest modifications to the model, particularly regarding the definition of semantic hub(s).

Cette thèse est dédiée à l'estimation de l'intensité d'une source de polluant de type particulaire par inversion de signaux de concentration mesurés avec un nombre fini de capteurs placés loin de la source. Cette méthode d'estimation inclut deux étapes distinctes. La première étape consiste à déterminer les paramètres du modèle d'inversion en utilisant une source d'aérosol connue et les mesures de concentration en particules correspondantes. Dans une seconde étape, une source d'aérosol inconnue est reconstruite à partir de l'inversion du modèle et des mesures de la concentration. Ce manuscrit traite dans un premier temps du cas stationnaire. L'approche théorique exposée permet de proposer un placement optimal des capteurs en plus de la méthode d'estimation de la source. Dans un second temps, on considère le cas où la source inconnue d'aérosol est instationnaire. La méthode d'estimation repose sur une approche convolutive du système, en introduisant la notion d'impédance source/capteur. Après une présentation de la technique d'inversion propre à la méthode d'estimation, elle est appliquée expérimentalement au cas des machines à bois éléctroportatives, dans le but de les discriminer en fonction de leur caractère émissif / This thesis is dedicated to the determination of unknown aerosol sources emission profiles from aerosol concentration measurements in the far-field. This procedure includes two distinct steps. The first step consists in determining the model linking the aerosol source and the concentration measurements using a known source of aerosols and the corresponding dust measurements. In a second step, the unknown source of aerosols is reconstructed by inverting the model for the measured aerosol concentrations. This manuscript deals in a first time with the stationary case. The exposed theoretical approach allows to suggest an optimal sensors placement in addition to the source estimation method. In a second time, we consider the case where the unknown aerosol source is unsteady. The estimation method is then based on a convolutive system approach, introducing the concept of source/sensor impedance. After a presentation of the numerical inversion technique, the method is applied experimentally to the real case of hand-held wood working machines so as to classify the machines with respect to their emission rate

The data traffic has been growing drastic over the past few years due to digitization and new technologies that are introduced to the market, such as autonomous cars. In order to meet this demand, the MIMO-OFDM system is used in the fifth generation wireless network, 5G. Designing the optimal wireless network is currently the main research within the area of telecommunication. In order to achieve such a system, multiple factors has to be taken into account, such as the suppression of interference from other users. A traditional method called linear minimum mean square error filter is currently used to suppress the interferences. To derive such a filter, a selection of parameters has to be estimated. One of these parameters is the ideal interference plus noise covariance matrix. By gathering prior knowledge of the underlying structure of the interference channel matrices in terms of the number of interferers and their corresponding bandwidths, the estimation of the ideal covariance matrix could be facilitated. As for this thesis, machine learning algorithms were used to extract these prior knowledge. More specifically, a two or three hidden layer feedforward neural network and a support vector machine with a linear kernel was used. The empirical findings implies promising results with accuracies above 95% for each model. / Under de senaste åren har dataanvändningen ökat drastiskt på grund av digitaliseringen och allteftersom nya teknologier introduceras på marknaden, exempelvis självkörande bilar. För att bemöta denna efterfrågan används ett s.k. MIMO-OFDM system i den femte generationens trådlösa nätverk, 5G. Att designa det optimala trådlösa nätverket är för närvarande huvudforskningen inom telekommunikation och för att uppnå ett sådant system måste flera faktorer beaktas, bland annat störningar från andra användare. En traditionell metod som används för att dämpa störningarna kallas för linjära minsta medelkvadratfelsfilter. För att hitta ett sådant filter måste flera olika parametrar estimeras, en av dessa är den ideala störning samt bruskovariansmatrisen. Genom att ta reda på den underliggande strukturen i störningsmatriserna i termer av antal störningar samt deras motsvarande bandbredd, är något som underlättar uppskattningen av den ideala kovariansmatrisen. I följande avhandling har olika maskininlärningsalgoritmer applicerats för att extrahera dessa informationer. Mer specifikt, ett neuralt nätverk med två eller tre gömda lager samt stödvektormaskin med en linjär kärna har använts. De slutliga resultaten är lovande med en noggrannhet på minst 95% för respektive modell.