Higher Order Statistics (HOS) and Second Order Statistics (SOS) approaches have certain advantages and disadvantages in signal processing applications. HOS approach provides more statistical information for non-Gaussian signals. On the other hand, SOS approach is more robust to the estimation errors than the HOS approach, especially when the number of observations is small. In this thesis, HOS and SOS approaches are jointly used in order to take advantage of both methods. In this respect, the joint use of HOS and SOS approaches are introduced for online calibration of sensor arrays with arbitrary geometries. Three different problems in online array calibration are considered and new algorithms for each of these problems are proposed. In the first problem, the positions of the randomly deployed sensors are completely unknown except the two reference sensors and HOS and SOS approaches are used iteratively for the joint Direction of Arrival (DOA) and sensor position estimation. Iterative HOS-SOS algorithm (IHOSS) solves the ambiguity problem in sensor position estimation by observing the source signals at least in two different frequencies and hence it is applicable for wideband signals. The conditions on these frequencies are presented. IHOSS is the first algorithm in the literature which finds the DOA and sensor position estimations in case of randomly deployed sensors with unknown coordinates. In the second problem, narrowband signals are considered and it is assumed that the nominal sensor positions are known. Modified IHOSS (MIHOSS) algorithm uses the nominal sensor positions to solve the ambiguity problem in sensor position estimation. This algorithm can handle both small and large errors in sensor positions. The upper bound of perturbations for unambiguous sensor position estimation is presented. In the last problem, an online array calibration method is proposed for sensor arrays where the sensors have unknown gain/phase mismatches and mutual coupling coefficients. In this case, sensor positions are assumed to be known. The mutual coupling matrix is unstructured. The two reference sensors are assumed to be perfectly calibrated. IHOSS algorithm is adapted for online calibration and parameter estimation, and hence CIHOSS algorithm is obtained. While CIHOSS originates from IHOSS, it is fundamentally different in many aspects. CIHOSS uses multiple virtual ESPRIT structures and employs an alignment technique to order the elements of rows of the actual array steering matrix. In this thesis, a new cumulant matrix estimation technique is proposed for the HOS approach by converting the multi-source problem into a single source one. The proposed algorithms perform well even in the case of correlated source signals due to the effectiveness of the proposed cumulant matrix estimate. The iterative procedure in all the proposed algorithms is guaranteed to converge. Closed form expressions are derived for the deterministic Cram&acute / er-Rao bound (CRB) for DOA and unknown calibration parameters for non-circular complex Gaussian noise with unknown covariance matrix. Simulation results show that the performances of the proposed methods approach to the CRB for both DOA and unknown calibration parameter estimations for high SNR.

Progressive Type-II censoring was introduced by Cohen (1963) and has since been the topic of much research. The question stands whether it is sensible to use this sampling plan by design, instead of regular Type-II right censoring. We introduce an asymptotic progressive censoring model, and find optimal censoring schemes for location-scale families. Our optimality criterion is the determinant of the 2x2 covariance matrix of the asymptotic best linear unbiased estimators. We present an explicit expression for this criterion, and conditions for its boundedness. By means of numerical optimization, we determine optimal censoring schemes for the extreme value, the Weibull and the normal distributions. In many situations, it is shown that these progressive schemes significantly improve upon regular Type-II right censoring.

Simultaneous operation of different wireless applications in the same geographical region and the same frequency band gives rise to undesired interference issues. Since licensed (primary) applications have been granted priority access to the frequency spectrum, unlicensed (secondary) services should avoid imposing interference on the primary system. In other words, secondary system’s activity in the same bands should be in a controlled fashion so that the primary system maintains its quality of service (QoS) requirements. In this thesis, we consider collocated point-to-multipoint primary and secondary networks that have simultaneous access to the same frequency band. Particularly, we examine three different levels at which the two networks may coexist: pure interference, asymmetric co-existence, and symmetric co-existence levels. At the pure interference level, both networks operate simultaneously regardless of their interference to each other. At the other two levels, at least one of the networks attempts to mitigate its interference to the other network by deactivating some of its users. Specifically, at the asymmetric co-existence level, the secondary network selectively deactivates its users based on knowledge of the interference and channel gains, whereas at the symmetric level, the primary network also schedules its users in the same way. Our aim is to derive optimal sum-rates (i.e., throughputs) of both networks at each co-existence level as the number of users grows asymptotically and evaluate how the sum-rates scale with the network size. In order to find the asymptotic throughput results, we derive two propositions; one on the asymptotic behaviour of the largest order statistic and one on the asymptotic behaviour of the sum of lower order statistics. As a baseline comparison, we calculate primary and secondary sum-rates for the time division (TD) channel sharing. Then, we compare the asymptotic secondary sum-rate in TD to that under simultaneous channel sharing, while ensuring the primary network maintains the same sum-rate in both cases. Our results indicate that simultaneous channel sharing at both asymmetric and symmetric co-existence levels can outperform TD. Furthermore, this enhancement is achievable when user scheduling in uplink mode is based only on the interference gains to the opposite network and not on a network’s own channel gains. In other words, the optimal secondary sum-rate is achievable by applying a scheduling strategy, referred to as the least interference strategy, for which only the knowledge of interference gains is required and can be performed in a distributed way.

Obstructive sleep apnoea hypopnoea syndrome (OSAHS) is a highly prevalent disease in which upper airways are collapsed during sleep, leading to serious consequences. The reference standard of clinical diagnosis, called Polysomnography (PSG), requires a full-night hospital stay connected to over 15 measuring channels requiring physical contact with sensors. The vast quantity of physiological data acquired during the PSG has to be manually scored by a qualified technologist to assess the presence or absence of the decease. The PSG is inconvenient, time consuming, expensive and unsuited for community screening. The limited PSG facilities around the world have resulted in long waiting lists and a large fraction of patients remain undiagnosed at present. There has been a flurry of recent activities in developing a portable technology to resolve this need. All the devices have at least one sensor that requires physical contact with the subject. Unattended systems have not led to sufficiently high sensitivity/specificity levels to be used in a routine home monitoring or a community screening exercise. OSAHS is a sleep respiratory disorder principally caused by functional deficiencies occurring in the upper airways during sleep. These conditions and the reduced muscle tone during sleep, cause the muscles in the upper airways to collapse partially or completely thus resulting in episodes of hypopnoea and apnoea respectively. During the process leading to collapse of upper airways, upper airways act as an acoustic filter frequently producing snoring sounds. The process of snore sound production leads us to hypothesise that snore sounds should contain information on changes occurring in the upper airways during the OSAHS. Snoring almost always accompanies the OSAHS and is universally recognised as its earliest symptom. At present, however, the quantitative analysis of snore sounds is not a practice in clinical OSAHS detection. The vast potential of snoring in the diagnosis/screening of the OSAHS remains unused. Snoring-based technology opens up opportunities for building community-screening devices that do not depend on contact instrumentation. In this thesis, we present our work towards developing a snore–based non-contact instrumentation for the diagnosis/screening of the OSAHS. The primary task in the analysis of Snore Related Sounds (SRS) would be to segment the SRS data as accurately as possible into three main classes, snoring (voiced non-silence), breathing (unvoiced non-silence) and silence. A new algorithm was developed, based on pattern recognition for the SRS segmentation. Four features derived from the SRS were considered to classify samples of the SRS into three classes. We also investigated the performance of the algorithm with three commonly-used noise reduction (NR) techniques in speech processing, Amplitude Spectral Subtraction (ASS), Power Spectral Subtraction (PSS) and Short Time Spectral Amplitude (STSA) Estimation. It was found that the noise reduction, together with a proper choice of features, could improve the classification accuracy to 96.78%. A novel model for the SRS was proposed for the response of a mixed-phase system (total airways response, TAR) to a source excitation at the input. The TAR/source model is similar to the vocal tract/source model in speech synthesis and is capable of capturing the acoustical changes brought about by the collapsing upper airways in the OSAHS. An algorithm was developed, based on the higher-order-spectra (HOS) to jointly estimate the source and the TAR, preserving the true phase characteristics of the latter. Working on a clinical database of signals, we show that the TAR is indeed a mixed phased signal and second-order statistics cannot fully characterise it. Nocturnal speech sounds can corrupt snore recordings and pose a challenge to the snore-based OSAHS diagnosis. The TAR could be shown to detect speech segments embedded in snores and derive features to diagnose the OSAHS. Finally presented is a novel technique for diagnosing the OSAHS, based solely on multi-parametric snore sound analysis. The method comprises a logistic regression model fed with a range of snore parameters derived from its features — the pitch and Total Airways Response (TAR) estimated using a Higher Order Statistics (HOS) based algorithm. The model was developed and its performance validated on a clinical database consisting of overnight snoring sounds simultaneously recorded during a hospital PSG using a high fidelity sound recording setup. The K-fold cross validation technique was used for validating the model. The validation process achieved an 89.3% sensitivity with 92.3% specificity (the area under the Receiver Operating Characteristic (ROC) curve was 0.96) in classifying the data sets into the two groups, the OSAHS (AHI >10) and the non-OSAHS. These results are superior to the existing results and unequivocally illustrate the feasibility of developing a snore-based non-contact OSAHS screening device.

Obstructive sleep apnoea hypopnoea syndrome (OSAHS) is a highly prevalent disease in which upper airways are collapsed during sleep, leading to serious consequences. The reference standard of clinical diagnosis, called Polysomnography (PSG), requires a full-night hospital stay connected to over 15 measuring channels requiring physical contact with sensors. The vast quantity of physiological data acquired during the PSG has to be manually scored by a qualified technologist to assess the presence or absence of the decease. The PSG is inconvenient, time consuming, expensive and unsuited for community screening. The limited PSG facilities around the world have resulted in long waiting lists and a large fraction of patients remain undiagnosed at present. There has been a flurry of recent activities in developing a portable technology to resolve this need. All the devices have at least one sensor that requires physical contact with the subject. Unattended systems have not led to sufficiently high sensitivity/specificity levels to be used in a routine home monitoring or a community screening exercise. OSAHS is a sleep respiratory disorder principally caused by functional deficiencies occurring in the upper airways during sleep. These conditions and the reduced muscle tone during sleep, cause the muscles in the upper airways to collapse partially or completely thus resulting in episodes of hypopnoea and apnoea respectively. During the process leading to collapse of upper airways, upper airways act as an acoustic filter frequently producing snoring sounds. The process of snore sound production leads us to hypothesise that snore sounds should contain information on changes occurring in the upper airways during the OSAHS. Snoring almost always accompanies the OSAHS and is universally recognised as its earliest symptom. At present, however, the quantitative analysis of snore sounds is not a practice in clinical OSAHS detection. The vast potential of snoring in the diagnosis/screening of the OSAHS remains unused. Snoring-based technology opens up opportunities for building community-screening devices that do not depend on contact instrumentation. In this thesis, we present our work towards developing a snore–based non-contact instrumentation for the diagnosis/screening of the OSAHS. The primary task in the analysis of Snore Related Sounds (SRS) would be to segment the SRS data as accurately as possible into three main classes, snoring (voiced non-silence), breathing (unvoiced non-silence) and silence. A new algorithm was developed, based on pattern recognition for the SRS segmentation. Four features derived from the SRS were considered to classify samples of the SRS into three classes. We also investigated the performance of the algorithm with three commonly-used noise reduction (NR) techniques in speech processing, Amplitude Spectral Subtraction (ASS), Power Spectral Subtraction (PSS) and Short Time Spectral Amplitude (STSA) Estimation. It was found that the noise reduction, together with a proper choice of features, could improve the classification accuracy to 96.78%. A novel model for the SRS was proposed for the response of a mixed-phase system (total airways response, TAR) to a source excitation at the input. The TAR/source model is similar to the vocal tract/source model in speech synthesis and is capable of capturing the acoustical changes brought about by the collapsing upper airways in the OSAHS. An algorithm was developed, based on the higher-order-spectra (HOS) to jointly estimate the source and the TAR, preserving the true phase characteristics of the latter. Working on a clinical database of signals, we show that the TAR is indeed a mixed phased signal and second-order statistics cannot fully characterise it. Nocturnal speech sounds can corrupt snore recordings and pose a challenge to the snore-based OSAHS diagnosis. The TAR could be shown to detect speech segments embedded in snores and derive features to diagnose the OSAHS. Finally presented is a novel technique for diagnosing the OSAHS, based solely on multi-parametric snore sound analysis. The method comprises a logistic regression model fed with a range of snore parameters derived from its features — the pitch and Total Airways Response (TAR) estimated using a Higher Order Statistics (HOS) based algorithm. The model was developed and its performance validated on a clinical database consisting of overnight snoring sounds simultaneously recorded during a hospital PSG using a high fidelity sound recording setup. The K-fold cross validation technique was used for validating the model. The validation process achieved an 89.3% sensitivity with 92.3% specificity (the area under the Receiver Operating Characteristic (ROC) curve was 0.96) in classifying the data sets into the two groups, the OSAHS (AHI >10) and the non-OSAHS. These results are superior to the existing results and unequivocally illustrate the feasibility of developing a snore-based non-contact OSAHS screening device.

abstract: Many products undergo several stages of testing ranging from tests on individual components to end-item tests. Additionally, these products may be further "tested" via customer or field use. The later failure of a delivered product may in some cases be due to circumstances that have no correlation with the product's inherent quality. However, at times, there may be cues in the upstream test data that, if detected, could serve to predict the likelihood of downstream failure or performance degradation induced by product use or environmental stresses. This study explores the use of downstream factory test data or product field reliability data to infer data mining or pattern recognition criteria onto manufacturing process or upstream test data by means of support vector machines (SVM) in order to provide reliability prediction models. In concert with a risk/benefit analysis, these models can be utilized to drive improvement of the product or, at least, via screening to improve the reliability of the product delivered to the customer. Such models can be used to aid in reliability risk assessment based on detectable correlations between the product test performance and the sources of supply, test stands, or other factors related to product manufacture. As an enhancement to the usefulness of the SVM or hyperplane classifier within this context, L-moments and the Western Electric Company (WECO) Rules are used to augment or replace the native process or test data used as inputs to the classifier. As part of this research, a generalizable binary classification methodology was developed that can be used to design and implement predictors of end-item field failure or downstream product performance based on upstream test data that may be composed of single-parameter, time-series, or multivariate real-valued data. Additionally, the methodology provides input parameter weighting factors that have proved useful in failure analysis and root cause investigations as indicators of which of several upstream product parameters have the greater influence on the downstream failure outcomes. / Dissertation/Thesis / Ph.D. Electrical Engineering 2011

Orientadores: João Marcos Travassos Romano, Francisco Rodrigo Porto Cavalcanti / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-04T00:19:46Z (GMT). No. of bitstreams: 1 Cavalcante_CharlesCasimiro_D.pdf: 8652621 bytes, checksum: bf432c4988b60a8e2465828f4f748b47 (MD5) Previous issue date: 2004 / Resumo: Esta tese é dedicada ao estudo de tecnicas de separação cega de fontes aplicadas ao contexto de processamento multiusuario em comunicações digitais. Utilizando estrategias de estimação da função de densidade de probabilidade (fdp), são propostos dois metodos de processamento multiusuario que permitem recuperar os sinais transmitidos pela medida de similaridade de Kullback-Leibler entre a fdp dos sinais a saida do dispositivo de separação e um modelo parametrico que contem as caracteristicas dos sinais transmitidos. Alem desta medida de similaridade, são empregados diferentes metodos que garantem a descorrelação entre as estimativas das fontes de tal forma que os sinais recuperados sejam provenientes de diferentes fontes. E ainda realizada a analise de convergencia dos metodos e suas equivalencias com tecnicas classicas resultando em algumas importantes relações entre criterios cegos e supervisionados, tais como o criterio proposto e o criterio de maxima a posteriori. Estes novos metodos aliam a capacidade de recuperação da informação uma baixa complexidade computacional. A proposição de metodos baseados na estimativa da fdp permitiu a realização de um estudo sobre o impacto das estatisticas de ordem superior em algoritmos adaptativos para separação cega de fontes. A utilização da expansão da fdp em series ortonormais permite avaliar atraves dos cumulantes a dinamica de um processo de separação de fontes. Para tratar com problemas de comunicação digital e proposta uma nova serie ortonormal, desenvolvida em torno de uma função de densidade de probabilidade dada por um somatorio de gaussianas. Esta serie e utilizada para evidenciar as diferenças em relação ao desempenho em tempo real ao se reter mais estatisticas de ordem superior. Simulações computacionais são realizadas para evidenciar o desempenho das propostas frente a tecnicas conhecidas da literatura em varias situações de necessidade de alguma estrategia de recuperação de sinais / Abstract: This thesis is devoted to study blind source separation techniques applied to multiuser processing in digital communications. Using probability density function (pdf) estimation strategies, two multiuser processing methods are proposed. They aim for recovering transmitted signal by using the Kullback-Leibler similarity measure between the signals pdf and a parametric model that contains the signals characteristics. Besides the similarity measure, different methods are employed to guarantee the decorrelation of the sources estimates, providing that the recovered signals origin from different sources. The convergence analysis of the methods as well as their equivalences with classical techniques are presented, resulting on important relationships between blind and supervised criteria such as the proposal and the maximum a posteriori one. Those new methods have a good trade-off between the recovering ability and computational complexity. The proposal os pdf estimation-based methods had allowed the investigation on the impact of higher order statistics on adaptive algorithms for blind source separation. Using pdf orthonormal series expansion we are able to evaluate through cumulants the dynamics of a source separation process. To be able to deal with digital communication signals, a new orthonormal series expansion is proposed. Such expansion is developed in terms of a Gaussian mixture pdf. This new expansion is used to evaluate the differences in real time processing when we retain more higher order statistics. Computational simulations are carried out to stress the performance of the proposals, faced to well known techniques reported in the literature, under the situations where a recovering signal strategy is required. / Doutorado / Telecomunicações e Telemática / Doutor em Engenharia Elétrica

Les principales contributions de cette thèse concernent l'identification à temps continu des systèmes par modèles non entiers dans un contexte à erreurs en les variables. Deux classes de méthodes sont développées : la première classe est fondée sur les statistiques d'ordre trois et la deuxième est fondée sur les statistiques d'ordre quatre. Dans chaque classe, deux cas différents sont distingués : le premier cas suppose que tous les ordres de dérivation non entiers sont connus a priori et seuls les coefficients de l'équation différentielle non entière sont estimés en utilisant les estimateurs fondés sur les statistiques d'ordre supérieur. Le deuxième cas suppose que les ordres de dérivation sont commensurables à un ordre nu estimé au même titre que les coefficients de l'équation différentielle non entière par des techniques d'optimisation non linéaire combinées aux estimateurs fondés sur les cumulants d'ordre trois et quatre. Des exemples de simulation numérique illustrent les développements théoriques. Des applications pratiques sur la modélisation du phénomène de diffusion de chaleur dans un barreau d'Aluminium et sur la modélisation d'un système électronique ont montré la pertinence des méthodes développées. / This thesis deals with continuous-time system identification by fractional models in the EIV context. Two classes of methods are developed : the first class is based on third-order statistics and the second one is based on fourth-order statistics. Firstly, all differentiation orders are known a priori and only the coefficients of the differential equation are estimated using the developed algorithms based on higher-order statistics. Then, they are extended to estimate both the fractional differential equation coefficients and the commensurate order. Simulation examples display the theoretical developments on system identification in the EIV context. A practical application for modeling heat transfer phenomena in an aluminium rod and for modeling an electronic real system have shown the efficiency of the developed methods.

Every year in the UK, tens of thousands of hospital patients suffer adverse events, such as un-planned transfers to Intensive Therapy Units or unexpected cardiac arrests. Studies have shown that in a large majority of cases, significant physiological abnormalities can be observed within the 24-hour period preceding such events. Such warning signs may go unnoticed, if they occur between observations by the nursing staff, or are simply not identified as such. Timely detection of these warning signs and appropriate escalation schemes have been shown to improve both patient outcomes and the use of hospital resources, most notably by reducing patients’ length of stay. Automated real-time early-warning systems appear to be cost-efficient answers to the need for continuous vital-sign monitoring. Traditionally, a limitation of such systems has been their sensitivity to noisy and artefactual measurements, resulting in false-alert rates that made them unusable in practice, or earned them the mistrust of clinical staff. Tarassenko et al. (2005) and Hann (2008) proposed a novelty detection approach to the problem of continuous vital-sign monitoring, which, in a clinical trial, was shown to yield clinically acceptable false alert rates. In this approach, an observation is compared to a data fusion model, and its “normality” assessed by comparing a chosen statistic to a pre-set threshold. The method, while informed by large amounts of training data, has a number of heuristic aspects. This thesis proposes a principled approach to multivariate novelty detection in stochastic time- series, where novelty scores have a probabilistic interpretation, and are explicitly linked to the starting assumptions made. Our approach stems from the observation that novelty detection using complex multivariate, multimodal generative models is generally an ill-defined problem when attempted in the data space. In situations where “novel” is equivalent to “improbable with respect to a probability distribution ”, formulating the problem in a univariate probability space allows us to use classical results of univariate statistical theory. Specifically, we propose a multivariate extension to extreme value theory and, more generally, order statistics, suitable for performing novelty detection in time-series generated from a multivariate, possibly multimodal model. All the methods introduced in this thesis are applied to a vital-sign monitoring problem and compared to the existing method of choice. We show that it is possible to outperform the existing method while retaining a probabilistic interpretation. In addition to their application to novelty detection for vital-sign monitoring, contributions in this thesis to existing extreme value theory and order statistics are also valid in the broader context of data-modelling, and may be useful for analysing data from other complex systems.

In this thesis we investigate how to improve the performance of MU-MIMO wireless system in terms of achieving Shannon capacity limit and efficient use of precious resource of radio spectrum in wireless communication. First a new suboptimal volume-based scheduling algorithm is presented, which can be applied in MU-MIMO downlink system to transmit signals concurrently to multiple users under the assumption of perfect channel information at transmitter and receiver. The volume-based scheduling algorithm utilises Block Diagonalisation precoding and Householder reduction procedure of QR factorisation. In comparison with capacity-based suboptimal scheduling algorithm, the volume-based algorithm has much reduced computational complexity with only a fraction of sum-rate capacity penalty from the upper bound of system capacity limit. In comparison with semi-orthogonal user selection suboptimal scheduling algorithm, the volume-based scheduling algorithm can be implemented with less computational complexity. Furthermore, the sum-rate capacity achieved via volume-based scheduling algorithm is higher than that achieved by SUS scheduling algorithm in the MIMO case. Then, a two-step scheduling algorithm is proposed, which can be used in the MU-MIMO system and under the assumption that channel state information is known to the receiver, but it is not known to the transmitter and the system under the feedback resource constraint. Assume that low bits codebook and high bits codebook are stored at the transmitter and receiver. The users are selected by using the low bits codebook; subsequently the BD precoding vectors for selected users are designed by employing high bits codebook. The first step of the algorithm can alleviate the load on feedback uplink channel in the MU-MIMO wireless system while the second step can aid precoding design to improve system sum-rate capacity. Next, a MU-MIMO cognitive radio (CR) wireless system has been studied. In such system, a primary wireless network and secondary wireless network coexist and the transmitters and receivers are equipped with multiple antennas. Spectrum sensing methods by which a portion of spectrum can be utilised by a secondary user when the spectrum is detected not in use by a primary user were investigated. A Free Probability Theory (FPT) spectrum sensing method that is a blind spectrum sensing method is proposed. By utilizing the asymptotic behaviour of random matrix based on FPT, the covariance matrix of transmitted signals can be estimated through a large number of observations of the received signals. The method performs better than traditional energy spectrum sensing method. We also consider cooperative spectrum sensing by using the FPT method in MU-MIMO CR system. Cooperative spectrum sensing can improve the performance of signal detection. Furthermore, with the selective cooperative spectrum sensing approach, high probability of detection can be achieved when the system is under false alarm constraint. Finally, spectrum sensing method based on the bispectrum of high-order statistics (HOS) and receive diversity in SIMO CR system is proposed. Multiple antennas on the receiver can improve received SNR value and therefore enhance spectrum sensing performance in terms of increase of system-level probability of detection. Discussions on cooperative spectrum sensing by using the spectrum sensing method based on HOS and receive diversity are presented.