Rich probabilistic representations for bearing only decentralised data fusion

The aim of this paper is to demonstrate the validity of using Gaussian mixture models (GMM) for representing probabilistic distributions in a decentralised data fusion (DDF) framework. GMMs are a powerful and compact stochastic representation allowing efficient communication of feature properties in large scale decentralised sensor networks. It will be shown that GMMs provide a basis for analytical solutions to the update and prediction operations for general Bayesian filtering. Furthermore, a variant on the Covariance Intersect algorithm for Gaussian mixtures will be presented ensuring a conservative update for the fusion of correlated information between two nodes in the network. In addition, purely visual sensory data will be used to show that decentralised data fusion and tracking of non-Gaussian states observed by multiple autonomous vehicles is feasible.

Simulating dynamic covariance structures for testing the adaptive behavior of variable selection algorithms (Invited Paper)

Variable selection for regression is a classical statistical problem, motivated by concerns that too large a number of covariates may bring about overfitting and unnecessarily high measurement costs. Novel difficulties arise in streaming contexts, where the correlation structure of the process may be drifting, in which case it must be constantly tracked so that selections may be revised accordingly. A particularly interesting phenomenon is that non-selected covariates become missing variables, inducing bias on subsequent decisions. This raises an intricate exploration-exploitation tradeoff, whose dependence on the covariance tracking algorithm and the choice of variable selection scheme is too complex to be dealt with analytically. We hence capitalise on the strength of simulations to explore this problem, taking the opportunity to tackle the difficult task of simulating dynamic correlation structures. © 2008 IEEE.

Population Monte Carlo Algorithm in High Dimensions

The population Monte Carlo algorithm is an iterative importance sampling scheme for solving static problems. We examine the population Monte Carlo algorithm in a simplified setting, a single step of the general algorithm, and study a fundamental problem that occurs in applying importance sampling to high-dimensional problem. The precision of the computed estimate from the simplified setting is measured by the asymptotic variance of estimate under conditions on the importance function. We demonstrate the exponential growth of the asymptotic variance with the dimension and show that the optimal covariance matrix for the importance function can be estimated in special cases.

Statistically assisted fluid registration algorithm - SAFIRA

In this paper, we develop and validate a new Statistically Assisted Fluid Registration Algorithm (SAFIRA) for brain images. A non-statistical version of this algorithm was first implemented in [2] and re-formulated using Lagrangian mechanics in [3]. Here we extend this algorithm to 3D: given 3D brain images from a population, vector fields and their corresponding deformation matrices are computed in a first round of registrations using the non-statistical implementation. Covariance matrices for both the deformation matrices and the vector fields are then obtained and incorporated (separately or jointly) in the regularizing (i.e., the non-conservative Lagrangian) terms, creating four versions of the algorithm. We evaluated the accuracy of each algorithm variant using the manually labeled LPBA40 dataset, which provides us with ground truth anatomical segmentations. We also compared the power of the different algorithms using tensor-based morphometry -a technique to analyze local volumetric differences in brain structure- applied to 46 3D brain scans from healthy monozygotic twins.

Disjunct eddy covariance measurements of volatile organic compound fluxes using proton transfer reaction mass spectrometry

Volatile organic compounds (VOCs) are emitted into the atmosphere from natural and anthropogenic sources, vegetation being the dominant source on a global scale. Some of these reactive compounds are deemed major contributors or inhibitors to aerosol particle formation and growth, thus making VOC measurements essential for current climate change research. This thesis discusses ecosystem scale VOC fluxes measured above a boreal Scots pine dominated forest in southern Finland. The flux measurements were performed using the micrometeorological disjunct eddy covariance (DEC) method combined with proton transfer reaction mass spectrometry (PTR-MS), which is an online technique for measuring VOC concentrations. The measurement, calibration, and calculation procedures developed in this work proved to be well suited to long-term VOC concentration and flux measurements with PTR-MS. A new averaging approach based on running averaged covariance functions improved the determination of the lag time between wind and concentration measurements, which is a common challenge in DEC when measuring fluxes near the detection limit. The ecosystem scale emissions of methanol, acetaldehyde, and acetone were substantial. These three oxygenated VOCs made up about half of the total emissions, with the rest comprised of monoterpenes. Contrary to the traditional assumption that monoterpene emissions from Scots pine originate mainly as evaporation from specialized storage pools, the DEC measurements indicated a significant contribution from de novo biosynthesis to the ecosystem scale monoterpene emissions. This thesis offers practical guidelines for long-term DEC measurements with PTR-MS. In particular, the new averaging approach to the lag time determination seems useful in the automation of DEC flux calculations. Seasonal variation in the monoterpene biosynthesis and the detailed structure of a revised hybrid algorithm, describing both de novo and pool emissions, should be determined in further studies to improve biological realism in the modelling of monoterpene emissions from Scots pine forests. The increasing number of DEC measurements of oxygenated VOCs will probably enable better estimates of the role of these compounds in plant physiology and tropospheric chemistry. Keywords: disjunct eddy covariance, lag time determination, long-term flux measurements, proton transfer reaction mass spectrometry, Scots pine forests, volatile organic compounds

Monaural sound source separation using covariance profile of partials

This paper addresses the problem of separation of pitched sounds in monaural recordings. We present a novel feature for the estimation of parameters of overlapping harmonics which considers the covariance of partials of pitched sounds. Sound templates are formed from the monophonic parts of the mixture recording. A match for every note is found among these templates on the basis of covariance profile of their harmonics. The matching template for the note provides the second order characteristics for the overlapped harmonics of the note. The algorithm is tested on the RWC music database instrument sounds. The results clearly show that the covariance characteristics can be used to reconstruct overlapping harmonics effectively.

Streaming covariance selection with applications to adaptive querying in sensor networks

Sensor networks can be naturally represented as graphical models, where the edge set encodes the presence of sparsity in the correlation structure between sensors. Such graphical representations can be valuable for information mining purposes as well as for optimizing bandwidth and battery usage with minimal loss of estimation accuracy. We use a computationally efficient technique for estimating sparse graphical models which fits a sparse linear regression locally at each node of the graph via the Lasso estimator. Using a recently suggested online, temporally adaptive implementation of the Lasso, we propose an algorithm for streaming graphical model selection over sensor networks. With battery consumption minimization applications in mind, we use this algorithm as the basis of an adaptive querying scheme. We discuss implementation issues in the context of environmental monitoring using sensor networks, where the objective is short-term forecasting of local wind direction. The algorithm is tested against real UK weather data and conclusions are drawn about certain tradeoffs inherent in decentralized sensor networks data analysis. © 2010 The Author. Published by Oxford University Press on behalf of The British Computer Society. All rights reserved.

The REFLEX project: Comparing different algorithms and implementations for the inversion of a terrestrial ecosystem model against eddy covariance data

We describe a model-data fusion (MDF) inter-comparison project (REFLEX), which compared various algorithms for estimating carbon (C) model parameters consistent with both measured carbon fluxes and states and a simple C model. Participants were provided with the model and with both synthetic net ecosystem exchange (NEE) of CO2 and leaf area index (LAI) data, generated from the model with added noise, and observed NEE and LAI data from two eddy covariance sites. Participants endeavoured to estimate model parameters and states consistent with the model for all cases over the two years for which data were provided, and generate predictions for one additional year without observations. Nine participants contributed results using Metropolis algorithms, Kalman filters and a genetic algorithm. For the synthetic data case, parameter estimates compared well with the true values. The results of the analyses indicated that parameters linked directly to gross primary production (GPP) and ecosystem respiration, such as those related to foliage allocation and turnover, or temperature sensitivity of heterotrophic respiration, were best constrained and characterised. Poorly estimated parameters were those related to the allocation to and turnover of fine root/wood pools. Estimates of confidence intervals varied among algorithms, but several algorithms successfully located the true values of annual fluxes from synthetic experiments within relatively narrow 90% confidence intervals, achieving >80% success rate and mean NEE confidence intervals <110 gC m−2 year−1 for the synthetic case. Annual C flux estimates generated by participants generally agreed with gap-filling approaches using half-hourly data. The estimation of ecosystem respiration and GPP through MDF agreed well with outputs from partitioning studies using half-hourly data. Confidence limits on annual NEE increased by an average of 88% in the prediction year compared to the previous year, when data were available. Confidence intervals on annual NEE increased by 30% when observed data were used instead of synthetic data, reflecting and quantifying the addition of model error. Finally, our analyses indicated that incorporating additional constraints, using data on C pools (wood, soil and fine roots) would help to reduce uncertainties for model parameters poorly served by eddy covariance data.

Identification of micro-scale anthropogenic CO2, heat and moisture sources – processing eddy covariance fluxes for a dense urban environment

Anthropogenic emissions of heat and exhaust gases play an important role in the atmospheric boundary layer, altering air quality, greenhouse gas concentrations and the transport of heat and moisture at various scales. This is particularly evident in urban areas where emission sources are integrated in the highly heterogeneous urban canopy layer and directly linked to human activities which exhibit significant temporal variability. It is common practice to use eddy covariance observations to estimate turbulent surface fluxes of latent heat, sensible heat and carbon dioxide, which can be attributed to a local scale source area. This study provides a method to assess the influence of micro-scale anthropogenic emissions on heat, moisture and carbon dioxide exchange in a highly urbanized environment for two sites in central London, UK. A new algorithm for the Identification of Micro-scale Anthropogenic Sources (IMAS) is presented, with two aims. Firstly, IMAS filters out the influence of micro-scale emissions and allows for the analysis of the turbulent fluxes representative of the local scale source area. Secondly, it is used to give a first order estimate of anthropogenic heat flux and carbon dioxide flux representative of the building scale. The algorithm is evaluated using directional and temporal analysis. The algorithm is then used at a second site which was not incorporated in its development. The spatial and temporal local scale patterns, as well as micro-scale fluxes, appear physically reasonable and can be incorporated in the analysis of long-term eddy covariance measurements at the sites in central London. In addition to the new IMAS-technique, further steps in quality control and quality assurance used for the flux processing are presented. The methods and results have implications for urban flux measurements in dense urbanised settings with significant sources of heat and greenhouse gases.

Fully adaptive gaussian mixture metropolis-hastings algorithm

Markov Chain Monte Carlo methods are widely used in signal processing and communications for statistical inference and stochastic optimization. In this work, we introduce an efficient adaptive Metropolis-Hastings algorithm to draw samples from generic multimodal and multidimensional target distributions. The proposal density is a mixture of Gaussian densities with all parameters (weights, mean vectors and covariance matrices) updated using all the previously generated samples applying simple recursive rules. Numerical results for the one and two-dimensional cases are provided.

Methodologies for Collision Risk Computation = Metodologías para Cálculo de Riesgo de Colisión

Esta tesis aborda metodologías para el cálculo de riesgo de colisión de satélites. La minimización del riesgo de colisión se debe abordar desde dos puntos de vista distintos. Desde el punto de vista operacional, es necesario filtrar los objetos que pueden presentar un encuentro entre todos los objetos que comparten el espacio con un satélite operacional. Puesto que las órbitas, del objeto operacional y del objeto envuelto en la colisión, no se conocen perfectamente, la geometría del encuentro y el riesgo de colisión deben ser evaluados. De acuerdo con dicha geometría o riesgo, una maniobra evasiva puede ser necesaria para evitar la colisión. Dichas maniobras implican un consumo de combustible que impacta en la capacidad de mantenimiento orbital y por tanto de la visa útil del satélite. Por tanto, el combustible necesario a lo largo de la vida útil de un satélite debe ser estimado en fase de diseño de la misión para una correcta definición de su vida útil, especialmente para satélites orbitando en regímenes orbitales muy poblados. Los dos aspectos, diseño de misión y aspectos operacionales en relación con el riesgo de colisión están abordados en esta tesis y se resumen en la Figura 3. En relación con los aspectos relacionados con el diseño de misión (parte inferior de la figura), es necesario evaluar estadísticamente las características de de la población espacial y las teorías que permiten calcular el número medio de eventos encontrados por una misión y su capacidad de reducir riesgo de colisión. Estos dos aspectos definen los procedimientos más apropiados para reducir el riesgo de colisión en fase operacional. Este aspecto es abordado, comenzando por la teoría descrita en [Sánchez-Ortiz, 2006]T.14 e implementada por el autor de esta tesis en la herramienta ARES [Sánchez-Ortiz, 2004b]T.15 proporcionada por ESA para la evaluación de estrategias de evitación de colisión. Esta teoría es extendida en esta tesis para considerar las características de los datos orbitales disponibles en las fases operacionales de un satélite (sección 4.3.3). Además, esta teoría se ha extendido para considerar riesgo máximo de colisión cuando la incertidumbre de las órbitas de objetos catalogados no es conocida (como se da el caso para los TLE), y en el caso de querer sólo considerar riesgo de colisión catastrófico (sección 4.3.2.3). Dichas mejoras se han incluido en la nueva versión de ARES [Domínguez-González and Sánchez-Ortiz, 2012b]T.12 puesta a disposición a través de [SDUP,2014]R.60. En fase operacional, los catálogos que proporcionan datos orbitales de los objetos espaciales, son procesados rutinariamente, para identificar posibles encuentros que se analizan en base a algoritmos de cálculo de riesgo de colisión para proponer maniobras de evasión. Actualmente existe una única fuente de datos públicos, el catálogo TLE (de sus siglas en inglés, Two Line Elements). Además, el Joint Space Operation Center (JSpOC) Americano proporciona mensajes con alertas de colisión (CSM) cuando el sistema de vigilancia americano identifica un posible encuentro. En función de los datos usados en fase operacional (TLE o CSM), la estrategia de evitación puede ser diferente debido a las características de dicha información. Es preciso conocer las principales características de los datos disponibles (respecto a la precisión de los datos orbitales) para estimar los posibles eventos de colisión encontrados por un satélite a lo largo de su vida útil. En caso de los TLE, cuya precisión orbital no es proporcionada, la información de precisión orbital derivada de un análisis estadístico se puede usar también en el proceso operacional así como en el diseño de la misión. En caso de utilizar CSM como base de las operaciones de evitación de colisiones, se conoce la precisión orbital de los dos objetos involucrados. Estas características se han analizado en detalle, evaluando estadísticamente las características de ambos tipos de datos. Una vez concluido dicho análisis, se ha analizado el impacto de utilizar TLE o CSM en las operaciones del satélite (sección 5.1). Este análisis se ha publicado en una revista especializada [Sánchez-Ortiz, 2015b]T.3. En dicho análisis, se proporcionan recomendaciones para distintas misiones (tamaño del satélite y régimen orbital) en relación con las estrategias de evitación de colisión para reducir el riesgo de colisión de manera significativa. Por ejemplo, en el caso de un satélite en órbita heliosíncrona en régimen orbital LEO, el valor típico del ACPL que se usa de manera extendida es 10-4. Este valor no es adecuado cuando los esquemas de evitación de colisión se realizan sobre datos TLE. En este caso, la capacidad de reducción de riesgo es prácticamente nula (debido a las grandes incertidumbres de los datos TLE) incluso para tiempos cortos de predicción. Para conseguir una reducción significativa del riesgo, sería necesario usar un ACPL en torno a 10-6 o inferior, produciendo unas 10 alarmas al año por satélite (considerando predicciones a un día) o 100 alarmas al año (con predicciones a tres días). Por tanto, la principal conclusión es la falta de idoneidad de los datos TLE para el cálculo de eventos de colisión. Al contrario, usando los datos CSM, debido a su mejor precisión orbital, se puede obtener una reducción significativa del riesgo con ACPL en torno a 10-4 (considerando 3 días de predicción). Incluso 5 días de predicción pueden ser considerados con ACPL en torno a 10-5. Incluso tiempos de predicción más largos se pueden usar (7 días) con reducción del 90% del riesgo y unas 5 alarmas al año (en caso de predicciones de 5 días, el número de maniobras se mantiene en unas 2 al año). La dinámica en GEO es diferente al caso LEO y hace que el crecimiento de las incertidumbres orbitales con el tiempo de propagación sea menor. Por el contrario, las incertidumbres derivadas de la determinación orbital son peores que en LEO por las diferencias en las capacidades de observación de uno y otro régimen orbital. Además, se debe considerar que los tiempos de predicción considerados para LEO pueden no ser apropiados para el caso de un satélite GEO (puesto que tiene un periodo orbital mayor). En este caso usando datos TLE, una reducción significativa del riesgo sólo se consigue con valores pequeños de ACPL, produciendo una alarma por año cuando los eventos de colisión se predicen a un día vista (tiempo muy corto para implementar maniobras de evitación de colisión).Valores más adecuados de ACPL se encuentran entre 5•10-8 y 10-7, muy por debajo de los valores usados en las operaciones actuales de la mayoría de las misiones GEO (de nuevo, no se recomienda en este régimen orbital basar las estrategias de evitación de colisión en TLE). Los datos CSM permiten una reducción de riesgo apropiada con ACPL entre 10-5 y 10-4 con tiempos de predicción cortos y medios (10-5 se recomienda para predicciones a 5 o 7 días). El número de maniobras realizadas sería una en 10 años de misión. Se debe notar que estos cálculos están realizados para un satélite de unos 2 metros de radio. En el futuro, otros sistemas de vigilancia espacial (como el programa SSA de la ESA), proporcionarán catálogos adicionales de objetos espaciales con el objetivo de reducir el riesgo de colisión de los satélites. Para definir dichos sistemas de vigilancia, es necesario identificar las prestaciones del catalogo en función de la reducción de riesgo que se pretende conseguir. Las características del catálogo que afectan principalmente a dicha capacidad son la cobertura (número de objetos incluidos en el catalogo, limitado principalmente por el tamaño mínimo de los objetos en función de las limitaciones de los sensores utilizados) y la precisión de los datos orbitales (derivada de las prestaciones de los sensores en relación con la precisión de las medidas y la capacidad de re-observación de los objetos). El resultado de dicho análisis (sección 5.2) se ha publicado en una revista especializada [Sánchez-Ortiz, 2015a]T.2. Este análisis no estaba inicialmente previsto durante la tesis, y permite mostrar como la teoría descrita en esta tesis, inicialmente definida para facilitar el diseño de misiones (parte superior de la figura 1) se ha extendido y se puede aplicar para otros propósitos como el dimensionado de un sistema de vigilancia espacial (parte inferior de la figura 1). La principal diferencia de los dos análisis se basa en considerar las capacidades de catalogación (precisión y tamaño de objetos observados) como una variable a modificar en el caso de un diseño de un sistema de vigilancia), siendo fijas en el caso de un diseño de misión. En el caso de las salidas generadas en el análisis, todos los aspectos calculados en un análisis estadístico de riesgo de colisión son importantes para diseño de misión (con el objetivo de calcular la estrategia de evitación y la cantidad de combustible a utilizar), mientras que en el caso de un diseño de un sistema de vigilancia, los aspectos más importantes son el número de maniobras y falsas alarmas (fiabilidad del sistema) y la capacidad de reducción de riesgo (efectividad del sistema). Adicionalmente, un sistema de vigilancia espacial debe ser caracterizado por su capacidad de evitar colisiones catastróficas (evitando así in incremento dramático de la población de basura espacial), mientras que el diseño de una misión debe considerar todo tipo de encuentros, puesto que un operador está interesado en evitar tanto las colisiones catastróficas como las letales. Del análisis de las prestaciones (tamaño de objetos a catalogar y precisión orbital) requeridas a un sistema de vigilancia espacial se concluye que ambos aspectos han de ser fijados de manera diferente para los distintos regímenes orbitales. En el caso de LEO se hace necesario observar objetos de hasta 5cm de radio, mientras que en GEO se rebaja este requisito hasta los 100 cm para cubrir las colisiones catastróficas. La razón principal para esta diferencia viene de las diferentes velocidades relativas entre los objetos en ambos regímenes orbitales. En relación con la precisión orbital, ésta ha de ser muy buena en LEO para poder reducir el número de falsas alarmas, mientras que en regímenes orbitales más altos se pueden considerar precisiones medias. En relación con los aspectos operaciones de la determinación de riesgo de colisión, existen varios algoritmos de cálculo de riesgo entre dos objetos espaciales. La Figura 2 proporciona un resumen de los casos en cuanto a algoritmos de cálculo de riesgo de colisión y como se abordan en esta tesis. Normalmente se consideran objetos esféricos para simplificar el cálculo de riesgo (caso A). Este caso está ampliamente abordado en la literatura y no se analiza en detalle en esta tesis. Un caso de ejemplo se proporciona en la sección 4.2. Considerar la forma real de los objetos (caso B) permite calcular el riesgo de una manera más precisa. Un nuevo algoritmo es definido en esta tesis para calcular el riesgo de colisión cuando al menos uno de los objetos se considera complejo (sección 4.4.2). Dicho algoritmo permite calcular el riesgo de colisión para objetos formados por un conjunto de cajas, y se ha presentado en varias conferencias internacionales. Para evaluar las prestaciones de dicho algoritmo, sus resultados se han comparado con un análisis de Monte Carlo que se ha definido para considerar colisiones entre cajas de manera adecuada (sección 4.1.2.3), pues la búsqueda de colisiones simples aplicables para objetos esféricos no es aplicable a este caso. Este análisis de Monte Carlo se considera la verdad a la hora de calcular los resultados del algoritmos, dicha comparativa se presenta en la sección 4.4.4. En el caso de satélites que no se pueden considerar esféricos, el uso de un modelo de la geometría del satélite permite descartar eventos que no son colisiones reales o estimar con mayor precisión el riesgo asociado a un evento. El uso de estos algoritmos con geometrías complejas es más relevante para objetos de dimensiones grandes debido a las prestaciones de precisión orbital actuales. En el futuro, si los sistemas de vigilancia mejoran y las órbitas son conocidas con mayor precisión, la importancia de considerar la geometría real de los satélites será cada vez más relevante. La sección 5.4 presenta un ejemplo para un sistema de grandes dimensiones (satélite con un tether). Adicionalmente, si los dos objetos involucrados en la colisión tienen velocidad relativa baja (y geometría simple, Caso C en la Figura 2), la mayor parte de los algoritmos no son aplicables requiriendo implementaciones dedicadas para este caso particular. En esta tesis, uno de estos algoritmos presentado en la literatura [Patera, 2001]R.26 se ha analizado para determinar su idoneidad en distintos tipos de eventos (sección 4.5). La evaluación frete a un análisis de Monte Carlo se proporciona en la sección 4.5.2. Tras este análisis, se ha considerado adecuado para abordar las colisiones de baja velocidad. En particular, se ha concluido que el uso de algoritmos dedicados para baja velocidad son necesarios en función del tamaño del volumen de colisión proyectado en el plano de encuentro (B-plane) y del tamaño de la incertidumbre asociada al vector posición entre los dos objetos. Para incertidumbres grandes, estos algoritmos se hacen más necesarios pues la duración del intervalo en que los elipsoides de error de los dos objetos pueden intersecar es mayor. Dicho algoritmo se ha probado integrando el algoritmo de colisión para objetos con geometrías complejas. El resultado de dicho análisis muestra que este algoritmo puede ser extendido fácilmente para considerar diferentes tipos de algoritmos de cálculo de riesgo de colisión (sección 4.5.3). Ambos algoritmos, junto con el método Monte Carlo para geometrías complejas, se han implementado en la herramienta operacional de la ESA CORAM, que es utilizada para evaluar el riesgo de colisión en las actividades rutinarias de los satélites operados por ESA [Sánchez-Ortiz, 2013a]T.11. Este hecho muestra el interés y relevancia de los algoritmos desarrollados para la mejora de las operaciones de los satélites. Dichos algoritmos han sido presentados en varias conferencias internacionales [Sánchez-Ortiz, 2013b]T.9, [Pulido, 2014]T.7,[Grande-Olalla, 2013]T.10, [Pulido, 2014]T.5, [Sánchez-Ortiz, 2015c]T.1. ABSTRACT This document addresses methodologies for computation of the collision risk of a satellite. Two different approaches need to be considered for collision risk minimisation. On an operational basis, it is needed to perform a sieve of possible objects approaching the satellite, among all objects sharing the space with an operational satellite. As the orbits of both, satellite and the eventual collider, are not perfectly known but only estimated, the miss-encounter geometry and the actual risk of collision shall be evaluated. In the basis of the encounter geometry or the risk, an eventual manoeuvre may be required to avoid the conjunction. Those manoeuvres will be associated to a reduction in the fuel for the mission orbit maintenance, and thus, may reduce the satellite operational lifetime. Thus, avoidance manoeuvre fuel budget shall be estimated, at mission design phase, for a better estimation of mission lifetime, especially for those satellites orbiting in very populated orbital regimes. These two aspects, mission design and operational collision risk aspects, are summarised in Figure 3, and covered along this thesis. Bottom part of the figure identifies the aspects to be consider for the mission design phase (statistical characterisation of the space object population data and theory computing the mean number of events and risk reduction capability) which will define the most appropriate collision avoidance approach at mission operational phase. This part is covered in this work by starting from the theory described in [Sánchez-Ortiz, 2006]T.14 and implemented by this author in ARES tool [Sánchez-Ortiz, 2004b]T.15 provided by ESA for evaluation of collision avoidance approaches. This methodology has been now extended to account for the particular features of the available data sets in operational environment (section 4.3.3). Additionally, the formulation has been extended to allow evaluating risk computation approached when orbital uncertainty is not available (like the TLE case) and when only catastrophic collisions are subject to study (section 4.3.2.3). These improvements to the theory have been included in the new version of ESA ARES tool [Domínguez-González and Sánchez-Ortiz, 2012b]T.12 and available through [SDUP,2014]R.60. At the operation phase, the real catalogue data will be processed on a routine basis, with adequate collision risk computation algorithms to propose conjunction avoidance manoeuvre optimised for every event. The optimisation of manoeuvres in an operational basis is not approached along this document. Currently, American Two Line Element (TLE) catalogue is the only public source of data providing orbits of objects in space to identify eventual conjunction events. Additionally, Conjunction Summary Message (CSM) is provided by Joint Space Operation Center (JSpOC) when the American system identifies a possible collision among satellites and debris. Depending on the data used for collision avoidance evaluation, the conjunction avoidance approach may be different. The main features of currently available data need to be analysed (in regards to accuracy) in order to perform estimation of eventual encounters to be found along the mission lifetime. In the case of TLE, as these data is not provided with accuracy information, operational collision avoidance may be also based on statistical accuracy information as the one used in the mission design approach. This is not the case for CSM data, which includes the state vector and orbital accuracy of the two involved objects. This aspect has been analysed in detail and is depicted in the document, evaluating in statistical way the characteristics of both data sets in regards to the main aspects related to collision avoidance. Once the analysis of data set was completed, investigations on the impact of those features in the most convenient avoidance approaches have been addressed (section 5.1). This analysis is published in a peer-reviewed journal [Sánchez-Ortiz, 2015b]T.3. The analysis provides recommendations for different mission types (satellite size and orbital regime) in regards to the most appropriate collision avoidance approach for relevant risk reduction. The risk reduction capability is very much dependent on the accuracy of the catalogue utilized to identify eventual collisions. Approaches based on CSM data are recommended against the TLE based approach. Some approaches based on the maximum risk associated to envisaged encounters are demonstrated to report a very large number of events, which makes the approach not suitable for operational activities. Accepted Collision Probability Levels are recommended for the definition of the avoidance strategies for different mission types. For example for the case of a LEO satellite in the Sun-synchronous regime, the typically used ACPL value of 10-4 is not a suitable value for collision avoidance schemes based on TLE data. In this case the risk reduction capacity is almost null (due to the large uncertainties associated to TLE data sets, even for short time-to-event values). For significant reduction of risk when using TLE data, ACPL on the order of 10-6 (or lower) seems to be required, producing about 10 warnings per year and mission (if one-day ahead events are considered) or 100 warnings per year (for three-days ahead estimations). Thus, the main conclusion from these results is the lack of feasibility of TLE for a proper collision avoidance approach. On the contrary, for CSM data, and due to the better accuracy of the orbital information when compared with TLE, ACPL on the order of 10-4 allows to significantly reduce the risk. This is true for events estimated up to 3 days ahead. Even 5 days ahead events can be considered, but ACPL values down to 10-5 should be considered in such case. Even larger prediction times can be considered (7 days) for risk reduction about 90%, at the cost of larger number of warnings up to 5 events per year, when 5 days prediction allows to keep the manoeuvre rate in 2 manoeuvres per year. Dynamics of the GEO orbits is different to that in LEO, impacting on a lower increase of orbits uncertainty along time. On the contrary, uncertainties at short prediction times at this orbital regime are larger than those at LEO due to the differences in observation capabilities. Additionally, it has to be accounted that short prediction times feasible at LEO may not be appropriate for a GEO mission due to the orbital period being much larger at this regime. In the case of TLE data sets, significant reduction of risk is only achieved for small ACPL values, producing about a warning event per year if warnings are raised one day in advance to the event (too short for any reaction to be considered). Suitable ACPL values would lay in between 5•10-8 and 10-7, well below the normal values used in current operations for most of the GEO missions (TLE-based strategies for collision avoidance at this regime are not recommended). On the contrary, CSM data allows a good reduction of risk with ACPL in between 10-5 and 10-4 for short and medium prediction times. 10-5 is recommended for prediction times of five or seven days. The number of events raised for a suitable warning time of seven days would be about one in a 10-year mission. It must be noted, that these results are associated to a 2 m radius spacecraft, impact of the satellite size are also analysed within the thesis. In the future, other Space Situational Awareness Systems (SSA, ESA program) may provide additional catalogues of objects in space with the aim of reducing the risk. It is needed to investigate which are the required performances of those catalogues for allowing such risk reduction. The main performance aspects are coverage (objects included in the catalogue, mainly limited by a minimum object size derived from sensor performances) and the accuracy of the orbital data to accurately evaluate the conjunctions (derived from sensor performance in regards to object observation frequency and accuracy). The results of these investigations (section 5.2) are published in a peer-reviewed journal [Sánchez-Ortiz, 2015a]T.2. This aspect was not initially foreseen as objective of the thesis, but it shows how the theory described in the thesis, initially defined for mission design in regards to avoidance manoeuvre fuel allocation (upper part of figure 1), is extended and serves for additional purposes as dimensioning a Space Surveillance and Tracking (SST) system (bottom part of figure below). The main difference between the two approaches is the consideration of the catalogue features as part of the theory which are not modified (for the satellite mission design case) instead of being an input for the analysis (in the case of the SST design). In regards to the outputs, all the features computed by the statistical conjunction analysis are of importance for mission design (with the objective of proper global avoidance strategy definition and fuel allocation), whereas for the case of SST design, the most relevant aspects are the manoeuvre and false alarm rates (defining a reliable system) and the Risk Reduction capability (driving the effectiveness of the system). In regards to the methodology for computing the risk, the SST system shall be driven by the capacity of providing the means to avoid catastrophic conjunction events (avoiding the dramatic increase of the population), whereas the satellite mission design should consider all type of encounters, as the operator is interested on avoiding both lethal and catastrophic collisions. From the analysis of the SST features (object coverage and orbital uncertainty) for a reliable system, it is concluded that those two characteristics are to be imposed differently for the different orbital regimes, as the population level is different depending on the orbit type. Coverage values range from 5 cm for very populated LEO regime up to 100 cm in the case of GEO region. The difference on this requirement derives mainly from the relative velocity of the encounters at those regimes. Regarding the orbital knowledge of the catalogues, very accurate information is required for objects in the LEO region in order to limit the number of false alarms, whereas intermediate orbital accuracy can be considered for higher orbital regimes. In regards to the operational collision avoidance approaches, several collision risk algorithms are used for evaluation of collision risk of two pair of objects. Figure 2 provides a summary of the different collision risk algorithm cases and indicates how they are covered along this document. The typical case with high relative velocity is well covered in literature for the case of spherical objects (case A), with a large number of available algorithms, that are not analysed in detailed in this work. Only a sample case is provided in section 4.2. If complex geometries are considered (Case B), a more realistic risk evaluation can be computed. New approach for the evaluation of risk in the case of complex geometries is presented in this thesis (section 4.4.2), and it has been presented in several international conferences. The developed algorithm allows evaluating the risk for complex objects formed by a set of boxes. A dedicated Monte Carlo method has also been described (section 4.1.2.3) and implemented to allow the evaluation of the actual collisions among a large number of simulation shots. This Monte Carlo runs are considered the truth for comparison of the algorithm results (section 4.4.4). For spacecrafts that cannot be considered as spheres, the consideration of the real geometry of the objects may allow to discard events which are not real conjunctions, or estimate with larger reliability the risk associated to the event. This is of particular importance for the case of large spacecrafts as the uncertainty in positions of actual catalogues does not reach small values to make a difference for the case of objects below meter size. As the tracking systems improve and the orbits of catalogued objects are known more precisely, the importance of considering actual shapes of the objects will become more relevant. The particular case of a very large system (as a tethered satellite) is analysed in section 5.4. Additionally, if the two colliding objects have low relative velocity (and simple geometries, case C in figure above), the most common collision risk algorithms fail and adequate theories need to be applied. In this document, a low relative velocity algorithm presented in the literature [Patera, 2001]R.26 is described and evaluated (section 4.5). Evaluation through comparison with Monte Carlo approach is provided in section 4.5.2. The main conclusion of this analysis is the suitability of this algorithm for the most common encounter characteristics, and thus it is selected as adequate for collision risk estimation. Its performances are evaluated in order to characterise when it can be safely used for a large variety of encounter characteristics. In particular, it is found that the need of using dedicated algorithms depend on both the size of collision volume in the B-plane and the miss-distance uncertainty. For large uncertainties, the need of such algorithms is more relevant since for small uncertainties the encounter duration where the covariance ellipsoids intersect is smaller. Additionally, its application for the case of complex satellite geometries is assessed (case D in figure above) by integrating the developed algorithm in this thesis with Patera’s formulation for low relative velocity encounters. The results of this analysis show that the algorithm can be easily extended for collision risk estimation process suitable for complex geometry objects (section 4.5.3). The two algorithms, together with the Monte Carlo method, have been implemented in the operational tool CORAM for ESA which is used for the evaluation of collision risk of ESA operated missions, [Sánchez-Ortiz, 2013a]T.11. This fact shows the interest and relevance of the developed algorithms for improvement of satellite operations. The algorithms have been presented in several international conferences, [Sánchez-Ortiz, 2013b]T.9, [Pulido, 2014]T.7,[Grande-Olalla, 2013]T.10, [Pulido, 2014]T.5, [Sánchez-Ortiz, 2015c]T.1.

Fast algorithms for automatic mapping with space-limited covariance functions

In this paper we discuss a fast Bayesian extension to kriging algorithms which has been used successfully for fast, automatic mapping in emergency conditions in the Spatial Interpolation Comparison 2004 (SIC2004) exercise. The application of kriging to automatic mapping raises several issues such as robustness, scalability, speed and parameter estimation. Various ad-hoc solutions have been proposed and used extensively but they lack a sound theoretical basis. In this paper we show how observations can be projected onto a representative subset of the data, without losing significant information. This allows the complexity of the algorithm to grow as O(n m 2), where n is the total number of observations and m is the size of the subset of the observations retained for prediction. The main contribution of this paper is to further extend this projective method through the application of space-limited covariance functions, which can be used as an alternative to the commonly used covariance models. In many real world applications the correlation between observations essentially vanishes beyond a certain separation distance. Thus it makes sense to use a covariance model that encompasses this belief since this leads to sparse covariance matrices for which optimised sparse matrix techniques can be used. In the presence of extreme values we show that space-limited covariance functions offer an additional benefit, they maintain the smoothness locally but at the same time lead to a more robust, and compact, global model. We show the performance of this technique coupled with the sparse extension to the kriging algorithm on synthetic data and outline a number of computational benefits such an approach brings. To test the relevance to automatic mapping we apply the method to the data used in a recent comparison of interpolation techniques (SIC2004) to map the levels of background ambient gamma radiation. © Springer-Verlag 2007.

Performance of the Bayesian online algorithm for the perceptron

In this letter, we derive continuum equations for the generalization error of the Bayesian online algorithm (BOnA) for the one-layer perceptron with a spherical covariance matrix using the Rosenblatt potential and show, by numerical calculations, that the asymptotic performance of the algorithm is the same as the one for the optimal algorithm found by means of variational methods with the added advantage that the BOnA does not use any inaccessible information during learning. © 2007 IEEE.

Nested Sampling and its Applications in Stable Compressive Covariance Estimation and Phase Retrieval with Near-Minimal Measurements

Compressed covariance sensing using quadratic samplers is gaining increasing interest in recent literature. Covariance matrix often plays the role of a sufficient statistic in many signal and information processing tasks. However, owing to the large dimension of the data, it may become necessary to obtain a compressed sketch of the high dimensional covariance matrix to reduce the associated storage and communication costs. Nested sampling has been proposed in the past as an efficient sub-Nyquist sampling strategy that enables perfect reconstruction of the autocorrelation sequence of Wide-Sense Stationary (WSS) signals, as though it was sampled at the Nyquist rate. The key idea behind nested sampling is to exploit properties of the difference set that naturally arises in quadratic measurement model associated with covariance compression. In this thesis, we will focus on developing novel versions of nested sampling for low rank Toeplitz covariance estimation, and phase retrieval, where the latter problem finds many applications in high resolution optical imaging, X-ray crystallography and molecular imaging. The problem of low rank compressive Toeplitz covariance estimation is first shown to be fundamentally related to that of line spectrum recovery. In absence if noise, this connection can be exploited to develop a particular kind of sampler called the Generalized Nested Sampler (GNS), that can achieve optimal compression rates. In presence of bounded noise, we develop a regularization-free algorithm that provably leads to stable recovery of the high dimensional Toeplitz matrix from its order-wise minimal sketch acquired using a GNS. Contrary to existing TV-norm and nuclear norm based reconstruction algorithms, our technique does not use any tuning parameters, which can be of great practical value. The idea of nested sampling idea also finds a surprising use in the problem of phase retrieval, which has been of great interest in recent times for its convex formulation via PhaseLift, By using another modified version of nested sampling, namely the Partial Nested Fourier Sampler (PNFS), we show that with probability one, it is possible to achieve a certain conjectured lower bound on the necessary measurement size. Moreover, for sparse data, an l1 minimization based algorithm is proposed that can lead to stable phase retrieval using order-wise minimal number of measurements.