En aquesta tesi s'estudia el problema d'estimació cega de segon ordre en comunicacions digitals. En aquest camp, els símbols transmesos esdevenen paràmetres no desitjats (nuisance parameters) d'estadística no gaussiana que degraden les prestacions de l'estimador. En aquest context, l'estimador de màxima versemblança (ML) és normalment desconegut excepte si la relació senyal-soroll (SNR) és prou baixa. En aquest cas particular, l'estimador ML és una funció quadràtica del vector de dades rebudes o, equivalentment, una transformació lineal de la matriu de covariància mostral. Aquesta característica es compartida per altres estimadors importants basats en el principi de màxima versemblança com ara l'estimador ML gaussià (GML) i l'estimador ML condicional (CML). Així mateix, l'estimador MUSIC, i altres mètodes de subespai relacionats amb ell, es basen en la diagonalització de la matriu de covariància mostral. En aquest marc, l'aportació principal d'aquesta tesi és la deducció i avaluació de l'estimador òptim de segon ordre per qualsevol SNR i qualsevol distribució dels nuisance parameters.El disseny d'estimadors quadràtics en llaç obert i llaç tancat s'ha plantejat de forma unificada. Pel que fa als estimadors en llaç obert, s'han derivat els estimadors de mínim error quadràtic mig i mínima variància considerant que els paràmetres d'interès són variables aleatòries amb una distribució estadística coneguda a priori però, altrament, arbitrària. A partir d'aquest plantejament Bayesià, els estimadors en llaç tancat es poden obtenir suposant que la distribució a priori dels paràmetres és altament informativa. En aquest model de petit error, el millor estimador quadràtic no esbiaixat, anomenat BQUE, s'ha formulat sense convenir cap estadística particular pels nuisance parameters. Afegit a això, l'anàlisi de l'estimador BQUE ha permès calcular quina és la fita inferior que no pot millorar cap estimador cec que utilitzi la matriu de covariància mostral. Probablement, el resultat principal de la tesi és la demostració de què els estimadors quadràtics són capaços d'utilitzar la informació estadística de quart ordre dels nuisance parameters. Més en concret, s'ha demostrat que tota la informació no gaussiana de les dades que els mètodes de segon ordre són capaços d'aprofitar apareix reflectida en els cumulants de quart ordre dels nuisance parameters. De fet, aquesta informació de quart ordre esdevé rellevant si el mòdul dels nuisance parameters és constant i la SNR és moderada o alta. En aquestes condicions, es demostra que la suposició gaussiana dels nuisance parameters dóna lloc a estimadors quadràtics no eficients. Un altre resultat original que es presenta en aquesta memòria és la deducció del filtre de Kalman estès de segon ordre, anomenat QEKF. L'estudi del QEKF assenyala que els algoritmes de seguiment (trackers) de segon ordre poden millorar simultàniament les seves prestacions d'adquisició i seguiment si la informació estadística de quart ordre dels nuisance parameters es té en compte. Una vegada més, aquesta millora és significativa si els nuisance parameters tenen mòdul constant i la SNR és prou alta. Finalment, la teoria dels estimadors quadràtics plantejada s'ha aplicat en alguns problemes d'estimació clàssics en l'àmbit de les comunicacions digitals com ara la sincronització digital no assistida per dades, el problema de l'estimació del temps d'arribada en entorns amb propagació multicamí, la identificació cega de la resposta impulsional del canal i, per últim, l'estimació de l'angle d'arribada en sistemes de comunicacions mòbils amb múltiples antenes. Per cadascuna d'aquestes aplicacions, s'ha realitzat un anàlisi intensiu, tant numèric com asimptòtic, de les prestacions que es poden aconseguir amb mètodes d'estimació de segon ordre. / This thesis deals with the problem of blind second-order estimation in digital communications. In this field, the transmitted symbols appear as non-Gaussian nuisance parameters degrading the estimator performance. In this context, the Maximum Likelihood (ML) estimator is generally unknown unless the signal-to-noise (SNR) is very low. In this particular case, if the SNR is asymptotically low, the ML solution is quadratic in the received data or, equivalently, linear in the sample covariance matrix. This significant feature is shared by other important ML-based estimators such as, for example, the Gaussian and Conditional ML estimators. Likewise, MUSIC and other related subspace methods are based on the eigendecomposition of the sample covariance matrix. From this background, the main contribution of this thesis is the deduction and evaluation of the optimal second-order parameter estimator for any SNR and any distribution of the nuisance parameters.A unified framework is provided for the design of open- and closed-loop second-order estimators. In the first case, the minimum mean square error and minimum variance second-order estimators are deduced considering that the wanted parameters are random variables of known but arbitrary prior distribution. From this Bayesian approach, closed-loop estimators are derived by imposing an asymptotically informative prior. In this small-error scenario, the best quadratic unbiased estimator (BQUE) is obtained without adopting any assumption about the statistics of the nuisance parameters. In addition, the BQUE analysis yields the lower bound on the performance of any blind estimator based on the sample covariance matrix.Probably, the main result in this thesis is the proof that quadratic estimators are able to exploit the fourth-order statistical information about the nuisance parameters. Specifically, the nuisance parameters fourth-order cumulants are shown to provide all the non-Gaussian information that is utilizable for second-order estimation. This fourth-order information becomes relevant in case of constant modulus nuisance parameters and medium-to-high SNRs. In this situation, the Gaussian assumption is proved to yield inefficient second-order estimates.Another original result in this thesis is the deduction of the quadratic extended Kalman filter (QEKF). The QEKF study concludes that second-order trackers can improve simultaneously the acquisition and steady-state performance if the fourth-order statistical information about the nuisance parameters is taken into account. Once again, this improvement is significant in case of constant modulus nuisance parameters and medium-to-high SNRs.Finally, the proposed second-order estimation theory is applied to some classical estimation problems in the field of digital communications such as non-data-aided digital synchronization, the related problem of time-of-arrival estimation in multipath channels, blind channel impulse response identification, and direction-of-arrival estimation in mobile multi-antenna communication systems. In these applications, an intensive asymptotic and numerical analysis is carried out in order to evaluate the ultimate limits of second-order estimation.

Paired comparison models in loglinear form are generalised linear models and can be fitted using the IWLS algorithm. Unfortunately, the design matrices can become very large and thus a method is needed to reduce computational load (relating to both space and time). This paper discusses an algorithm for fitting loglinear paired comparison models in the presence of many nuisance parameters which is based on partition rules for symmetric matrices and takes advantage of the special structure of the design matrix in Poisson loglinear models. The algorithm is implemented as an R function. Some simple examples illustrate its use for fitting both paired comparison models and (multinomial) logit models. (author's abstract) / Series: Research Report Series / Department of Statistics and Mathematics

Le but de ce travail a été de proposer des méthodes d'inférence pour décrire l'histoire naturelle de la phase pré-diagnostic de la démence. Durant celle-ci, qui dure une quinzaine d'années, les trajectoires de déclin cognitif sont non linéaires et hétérogènes entre les sujets. Pour ces raisons, nous avons choisi un modèle à changement de pente aléatoire pour les décrire. Une première partie de ce travail a consisté à proposer une procédure de test pour l'existence d'un changement de pente aléatoire. En effet, dans certaines sous-populations, le déclin cognitif semble lisse et la question de l'existence même d'un changement de pente se pose. Cette question présente un défi méthodologique en raison de la non-identifiabilité de certains paramètres sous l'hypothèse nulle rendant les tests standards inutiles. Nous avons proposé un supremum score test pour répondre à cette question. Une seconde partie du travail concernait l'ordre temporel du temps de changement entre plusieurs marqueurs. La démence est une maladie multidimensionnelle et plusieurs dimensions de la cognition sont affectées. Des schémas hypothétiques existent pour décrire l'histoire naturelle de la démence mais n'ont pas été éprouvés sur données réelles. Comparer le temps de changement de différents marqueurs mesurant différentes fonctions cognitives permet d'éclairer ces hypothèses. Dans cet esprit, nous proposons un modèle bivarié à changement de pente aléatoire permettant de comparer les temps de changement de deux marqueurs, potentiellement non gaussiens. Les méthodes proposées ont été évaluées sur simulations et appliquées sur des données issues de deux cohortes françaises. Enfin, nous discutons les limites de ces deux modèles qui se concentrent sur une accélération tardive du déclin cognitif précédant le diagnostic de démence et nous proposons un modèle alternatif qui estime plutôt une date de décrochage entre cas et non-cas. / The aim of this work was to propose inferential methods to describe natural history of the pre-diagnosis phase of dementia. During this phase, which can last around fifteen years, the cognitive decline trajectories are nonlinear and heterogeneous between subjects. Because heterogeneity and nonlinearity, we chose a random changepoint mixed model to describe these trajectories. A first part of this work was to propose a testing procedure to assess the existence of a random changepoint. Indeed, in some subpopulations, the cognitive decline seems smooth and the question of the existence of a changepoint itself araises. This question is methodologically challenging because of identifiability issues on some parameters under the null hypothesis that makes standard tests useless. We proposed a supremum score test to answer this question. A second part of this work was the comparison of the temporal order of different markers changepoint. Dementia is a multidimensional disease where different dimensions of the cognition are affected. Hypothetic cascade models exist for describing this natural history but have not been evaluated on real data. Comparing change over time of different markers measuring different cognitive functions gives precious insight on this hypothesis. In this spirit, we propose a bivariate random changepoint model allowing proper comparison of the time of change of two cognitive markers, potentially non Gaussian. The proposed methodologies were evaluated on simulation studies and applied on real data from two French cohorts. Finally, we discussed the limitations of the two models we used that focused on the late acceleration of the cognitive decline before dementia diagnosis and we proposed an alternative model that estimates the time of differentiation between cases and non-cases.

This paper deals with the computation of OC- and ASN-function of sequential probability ratio tests in the multi-parameter case. In generalization of the method of conjugated parameter pairs Wald-like approximations are presented for the OC- and ASN-function. These characteristics can be used describing robustness properties of a sequential test in case of nuisance parameters. As examples tests are considered for the mean and the variance of a normal distribution.

Made available in DSpace on 2016-06-02T20:06:04Z (GMT). No. of bitstreams: 1 2751.pdf: 886213 bytes, checksum: 2f07f7329a7f25f1759ddb5d7a6edd66 (MD5) Previous issue date: 2010-01-15 / Financiadora de Estudos e Projetos / In this study, we used the capture-recapture procedure to estimate the size of a closed population. We analysed three di_erent statistics models. For each one of these models we determined - through several methods of eliminating nuisance parameters - the likelihood function and the pro_le, conditional, uniform integrated, Je_reys integrated and generalized integrated likelihood functions of the population size, except for the last model where we determined a function that is analogous to the conditional likelihood function, called integrated restricted likelihood function. In each instance we determined the respectives maximum likelihood estimates, the empirical con_dence intervals and the empirical mean squared errors of the estimates for the population size and we studied, using simulated data, the performances of the models. / Nesta dissertação utilizamos o processo de captura-recaptura para estimar o tamanho de uma população fechada. Analisamos três modelos estatísticos diferentes e, para cada um deles, através de diversas metodologias de eliminação de parâmetros perturbadores, determinamos as funções de verossimilhança e de verossimilhança perfilada, condicional, integrada uniforme, integrada de Jeffreys e integrada generalizada do tamanho populacional, com exceção do último modelo onde determinamos uma função análoga à função de verossimilhança condicional, denominada função de verossimilhança restrita integrada. Em cada capítulo determinamos as respectivas estimativas de máxima verossimilhança e construímos intervalos de confiança empíricos para o tamanho populacional, bem como determinamos os erros quadráticos médios empíricos das estimativas e estudamos, através de dados simulados, as performances dos modelos.

This paper deals with the computation of OC- and ASN-function of sequential probability ratio tests in the multi-parameter case. In generalization of the method of conjugated parameter pairs Wald-like approximations are presented for the OC- and ASN-function. These characteristics can be used describing robustness properties of a sequential test in case of nuisance parameters. As examples tests are considered for the mean and the variance of a normal distribution.

Made available in DSpace on 2016-06-02T20:04:51Z (GMT). No. of bitstreams: 1 4032.pdf: 1016886 bytes, checksum: 6e1eb83f197a88332f8951b054c1f01a (MD5) Previous issue date: 2011-11-18 / Financiadora de Estudos e Projetos / The capture-recapture process, largely used in the estimation of the number of elements of animal population, is also applied to other branches of knowledge like Epidemiology, Linguistics, Software reliability, Ecology, among others. One of the _rst applications of this method was done by Laplace in 1783, with aim at estimate the number of inhabitants of France. Later, Carl G. J. Petersen in 1889 and Lincoln in 1930 applied the same estimator in the context of animal populations. This estimator has being known in literature as _Lincoln-Petersen_ estimator. In the mid-twentieth century several researchers dedicated themselves to the formulation of statistical models appropriated for the estimation of population size, which caused a substantial increase in the amount of theoretical and applied works on the subject. The capture-recapture models are constructed under certain assumptions relating to the population, the sampling procedure and the experimental conditions. The main assumption that distinguishes models concerns the change in the number of individuals in the population during the period of the experiment. Models that allow for births, deaths or migration are called open population models, while models that does not allow for these events to occur are called closed population models. In this work, the goal is to characterize likelihood functions obtained by applying methods of elimination of nuissance parameters in the case of closed population models. Based on these likelihood functions, we discuss methods for point and interval estimation of the population size. The estimation methods are illustrated on a real data-set and their frequentist properties are analised via Monte Carlo simulation. / O processo de captura-recaptura, amplamente utilizado na estimação do número de elementos de uma população de animais, é também aplicado a outras áreas do conhecimento como Epidemiologia, Linguística, Con_abilidade de Software, Ecologia, entre outras. Uma das primeiras aplicações deste método foi feita por Laplace em 1783, com o objetivo de estimar o número de habitantes da França. Posteriormente, Carl G. J. Petersen em 1889 e Lincoln em 1930 utilizaram o mesmo estimador no contexto de popula ções de animais. Este estimador _cou conhecido na literatura como o estimador de _Lincoln-Petersen_. Em meados do século XX muitos pesquisadores se dedicaram à formula ção de modelos estatísticos adequados à estimação do tamanho populacional, o que causou um aumento substancial da quantidade de trabalhos teóricos e aplicados sobre o tema. Os modelos de captura-recaptura são construídos sob certas hipóteses relativas à população, ao processo de amostragem e às condições experimentais. A principal hipótese que diferencia os modelos diz respeito à mudança do número de indivíduos da popula- ção durante o período do experimento. Os modelos que permitem que haja nascimentos, mortes ou migração são chamados de modelos para população aberta, enquanto que os modelos em que tais eventos não são permitidos são chamados de modelos para popula- ção fechada. Neste trabalho, o objetivo é caracterizar o comportamento de funções de verossimilhança obtidas por meio da utilização de métodos de eliminação de parâmetros perturbadores, no caso de modelos para população fechada. Baseado nestas funções de verossimilhança, discutimos métodos de estimação pontual e intervalar para o tamanho populacional. Os métodos de estimação são ilustrados através de um conjunto de dados reais e suas propriedades frequentistas são analisadas via simulação de Monte Carlo.

Humanitarian clearance of minefields is most often carried out by hand, conventionally using a a metal detector and a probe. Detection is a very slow process, as every piece of detected metal must treated as if it were a landmine and carefully probed and excavated, while many of them are not. The process can be safely sped up by use of Ground-Penetrating Radar (GPR) to image the subsurface, to verify metal detection results and safely ignore any objects which could not possibly be a landmine. In this thesis, we explore the possibility of using Full Wave Inversion (FWI) to improve GPR imaging for landmine detection. Posing the imaging task as FWI means solving the large-scale, non-linear and ill-posed optimisation problem of determining the physical parameters of the subsurface (such as electrical permittivity) which would best reproduce the data. This thesis begins by giving an overview of all the mathematical and implementational aspects of FWI, so as to provide an informative text for both mathematicians (perhaps already familiar with other inverse problems) wanting to contribute to the mine detection problem, as well as a wider engineering audience (perhaps already working on GPR or mine detection) interested in the mathematical study of inverse problems and FWI.We present the first numerical 3D FWI results for GPR, and consider only surface measurements from small-scale arrays as these are suitable for our application. The FWI problem requires an accurate forward model to simulate GPR data, for which we use a hybrid finite-element boundary-integral solver utilising first order curl-conforming N\'d\'{e}lec (edge) elements. We present a novel `line search' type algorithm which prioritises inversion of some target parameters in a region of interest (ROI), with the update outside of the area defined implicitly as a function of the target parameters. This is particularly applicable to the mine detection problem, in which we wish to know more about some detected metallic objects, but are not interested in the surrounding medium. We may need to resolve the surrounding area though, in order to account for the target being obscured and multiple scattering in a highly cluttered subsurface. We focus particularly on spatial sensitivity of the inverse problem, using both a singular value decomposition to analyse the Jacobian matrix, as well as an asymptotic expansion involving polarization tensors describing the perturbation of electric field due to small objects. The latter allows us to extend the current theory of sensitivity in for acoustic FWI, based on the Born approximation, to better understand how polarization plays a role in the 3D electromagnetic inverse problem. Based on this asymptotic approximation, we derive a novel approximation to the diagonals of the Hessian matrix which can be used to pre-condition the GPR FWI problem.