Theses supervised by Prof. Dr. Orhan Arıkan
23 theses · İhsan Doğramacı Bilkent University
İyonküre değişkenlenlerin krigleme, eş-krigleme ve uzay-zaman krigleme teknikleri kullanılarak aradeğerlemesi
Long distance communication and navigation systems operating in the HF band use interacting signals as they travel through the ionosphere. It is important to accurately model ionospheric behavior to increase the performance of these systems. Delays occurring in the signals depend on the refrectivity which is a function of frequency of the signals, and the electron density on the signal path at the time of propagation. Depending on the change in the solar activities, the electron distribution in the ionosphere changes spatially and temporally. The change in ionosphere can be tracked by various parameters and the space-time distribution of these parameters. Total Electron Content (TEC), the total number of electrons in a cylinder with one meter square cross-sectional area over a ray path is used as an important descriptor for the ionosphere. It is possible to generate TEC maps with high spatial resolution using the information obtained by processing the GPS satellite signals by constantly operating reference stations (CORS) GPS receivers. In particular, there are two other parameters that are used in HF communication and direction finding applications: foF2, which is the highest plasma frequency of foF2 layer, and hmF2, which is the height of maximum ionization. Sensitive foF2 and hmF2 measurements can be made by ionosonde systems. However, these systems are highly sparser than TEC measurements. For this reason, the resolution of the foF2 and hmF2 maps is less than the TEC maps. In this study, we propose a space-time mapping technique based on Co-Kriging which is used in conjunction with TEC data, that is correlated to these parameters, to increase the resolutions of foF2 and hmF2 maps. The performance of the proposed technique is compared with the alternatives and the increase in performance achieved is described statistically.
Büyük ölçekli doğrusal en küçük kareler problemleri için hızlı ve gürbüz çözüm yöntemleri
Momentum Iterative Hessian Sketch (M-IHS) techniques, a group of solvers for large scale linear Least Squares (LS) problems, are proposed and analyzed in detail. Proposed M-IHS techniques are obtained by incorporating the Heavy Ball Acceleration into the Iterative Hessian Sketch algorithm and they provide significant improvements over the randomized preconditioning techniques. By using approximate solvers along with the iterations, the proposed techniques are capable of avoiding all matrix decompositions and inversions, which is one of the main advantages over the alternative solvers such as the Blendenpik and the LSRN. Similar to the Chebyshev Semi-iterations, the M-IHS variants do not use any inner products and eliminate the corresponding synchronization steps in hierarchical or distributed memory systems, yet the M-IHS converges faster than the Chebyshev Semi-iteration based solvers. Lower bounds on the required sketch size for various randomized distributions are established through the error analyses of the M-IHS variants. Unlike the previously proposed approaches to produce a solution approximation, the proposed M-IHS techniques can use sketch sizes that are proportional to the statistical dimension which is always smaller than the rank of the coefficient matrix. Additionally, hybrid schemes are introduced to estimate the unknown l2-norm regularization parameter along with the iterations of the M-IHS techniques. Unlike conventional hybrid methods, the proposed Hybrid M-IHS techniques estimate the regularization parameter from the lower dimensional sub-problems that are constructed by random projections rather than the deterministic projections onto the Krylov Subspaces. Since the lower dimensional sub-problems that arise during the iterations of the Hybrid M-IHS variants are close approximations to the Newton sub-systems and the accuracy of their solutions increase exponentially, the parameters estimated from them rapidly converge to a proper regularization parameter for the full problem. In various numerical experiments conducted at several noise levels, the Hybrid M-IHS variants consistently estimated better regularization parameters and constructed solutions with less errors than the direct methods in far fewer iterations than the conventional hybrid methods. In large scale applications where the coefficient matrix is distributed over a memory array, the proposed Hybrid M-IHS variants provide improved efficiency by minimizing the number of distributed matrix-vector multiplications with the coefficient matrix.
Elektronik taarruz sistemlerinde derin öğrenme: Otomatik darbe tespiti ve istemli darbe ici kipleme sınıflandırma
Detection and classification of radar systems based on modulation analysis on pulses they transmit is an important application in electronic warfare systems. Many of the present works focus on classifying modulations assuming signal detection is done beforehand without providing any detection method. In this work, we propose two novel deep-learning based techniques for automatic pulse detection and intra-pulse modulation recognition of radar signals. As the first nechnique, an LSTM based multi-task learning model is proposed for end-to-end pulse detection and modulation classification. As the second technique, reassigned spectrogram of measured radar signal and detected outliers of its instantaneous phases filtered by a special function are used for training multiple convolutional neural networks. Automatically extracted features from the networks are fused to distinguish frequency and phase modulated signals. Another major issue on this area is the training and evaluation of supervised neural network based models. To overcome this issue we have developed an Intentional Modulation on Pulse (IMOP) measurement simulator which can generate over 15 main phase and frequency modulations with realistic pulses and noises. Simulation results show that the proposed FFCNN and MODNET techniques outperform the current state-of-the-art alternatives and is easily scalable among broad range of modulation types.
Uzaktan algılama uygulamaları için yenilikçi teknikler
Remote sensing is the acquisition of information about an object or phenomenon without a direct contact. In this thesis, novel signal processing techniques are proposed for two different remote sensing applications. First, for the reconstruction of electron density profile in the ionosphere using the ionosonde measurements, a new technique, ISED, is proposed. By using a hidden Markov model, ISED technique first identifies actual ionosonde echoes reflected from the ionosphere. Then, vertical electron density profile is estimated as the solution to a model based convex optimization problem. Conducted experiments on real ionosonde data show that ISED overperforms the state of the art ionogram inversion tools. Second part, for Specific Emitter Identification (SEI) of radar transmitters, a new SEI technique is proposed. SEI is an important Electronic Warfare (EW) activity that aims to identify unique transmitters, even the emitters of same kind, by using the subtle differences on the transmitted signals. Proposed SEI consist of two main stages. In the first stage, received radar pulses are time aligned and coherently integrated to reveal subtle differences which could be used to distinguish different radar transmitters. In the second step, Variational Mode Decomposition (VMD) is used to decompose both the envelope and the instantaneous frequency of the received radar signal into a set of modes. Then, these mod signals are characterized by using a group of features for identification. Highly successful identification performance with the proposed method on real radar datasets is demonstrated.
Çeşitli sensör ağlarında makine öğrenimi ile anomali tespiti
Earthquake precursor detection is one of the oldest research areas that has the potential of saving human lives. Recent studies have enlightened the fact that strong seismic activities and earthquakes affect the electron distribution of the ionosphere. These effects are clearly observable on the ionospheric Total Electron Content (TEC) that shall be measured by using the satellite position data of the Global Navigation Satellite System (GNSS). In this dissertation, several earthquake precursor detection techniques are proposed and their precursor detection performances are investigated on TEC data obtained from different sensor networks. First, a model based earthquake precursor detection technique is proposed to detect precursors of the earthquakes with magnitudes greater than 5 in the vicinity of Turkey. Precursor detection and TEC reliability signals are generated by using ionospheric TEC variations. These signals are thresholded to obtain earthquake precursor decisions. Earthquake precursor detections are made by using Particle Swarm Optimization (PSO) technique on these precursor decisions. Performance evaluations show that the proposed technique is able to detect $14$ out of $23$ earthquake precursors of magnitude larger than $5$ in Richter scale while generating $8$ false precursor decisions. Second, a machine learning based earthquake precursor detection technique, EQ-PD is proposed to detect precursors of the earthquakes with magnitudes greater than 4 in the vicinity of Italy. Spatial and spatio-temporal anomaly detection thresholds are obtained by using the statistics of TEC variation during seismically active times and applied on TEC variation based anomaly detection signal to form precursor decisions. Resulting spatial and spatio-temporal anomaly decisions are fed to a Support Vector Machine (SVM) classifier to generate earthquake precursor detections. When the precursor detection performance of the EQ-PD is investigated, it is observed that the technique is able to detect 22 out of 24 earthquake precursors while generating 13 false precursor decisions during 147 days of no-seismic activity. Last, a deep learning based earthquake precursor detection technique, DL-PD is proposed to detect precursors of the earthquakes with magnitudes greater than 5.4 in the vicinity Anatolia region. The DL-PD technique utilizes a deep neural network with spatio-temporal Global Ionospheric Map (GIM)-TEC data estimation capabilities. GIM-TEC anomaly score is obtained by comparing GIM-TEC estimates with GIM-TEC recordings. Earthquake precursor detections are generated by thresholding the GIM-TEC anomaly scores. Precursor detection performance evaluations show that DL-PD shall detect 5 out of 7 earthquake precursors while generating 1 false precursor decision during 416 days of no-seismic activity.
Yapay zeka etkinlenmiş sensör ağlarında 6G nesnelerin interneti senaryoları için inovasyon bazlı iletim
Goal-oriented signal processing and communications are assured to play a key role in developing the next generation of sensor devices and networks, e.g., 6G IoT networks. A critical task of semantic signal processing is the detection of innovation and transmission scheduling based on the innovation in AI-enabled sensor networks and IoT for 6G. This thesis proposes efficient and optimal sampling and transmission strategies for goal-oriented sensor networks for various data models, investigates their performances both analytically and numerically; and introduces the use of dimensionality reduction algorithms in semantic signal processing and highlights its effectiveness in real life case studies. That is, the proposed methods are explained rigorously and demonstrated through simulations and case studies based on real-world computer vision examples with recorded video signals. Numerical results indicate that the next generation sensor devices and networks can benefit significantly from the proposed methods in terms of energy efficiency and semantic innovation detection performances.
Anlamsal iletişimde güvenilirliğiarttırma: Anlamsal-grafik modellemesine stokastik yaklaşım
Semantic communication is expected to play a critical role in reducing traffic load in future intelligent large-scale sensor networks. With advances in Machine Learning (ML) and Deep Learning (DL) techniques, design of semantically-aware systems has become feasible in recent years. This thesis focuses on improving the reliability of the semantic information represented in a graph-based language that was previously developed. Inaccuracies in the representation of the semantic information can arise due to multiple factors, such as algorithmic shortcomings or sensory errors, deteriorating the performance of the semantic extractor. This thesis aims to model the temporal evolution of semantic information, represented using the graph language, to enhance its reliability. Each unique graph configuration is treated as a distinct state, leading to a Hidden Semi-Markov Model (HSMM) defined over the state space of the graph configurations. The HSMM formulation enables the integration of prior knowledge on the semantic signal into the graph sequences, enhancing the accuracy in identifying semantic innovations. Within the HSMM framework, algorithms designed for graph smoothing, semantic information fusion, and model learning are introduced. The efficacy of these algorithms in improving the reliability of the extracted semantic-graphs is demonstrated through simulations and video streams generated in the CARLA simulation environment.
Artificial intelligence-based hybrid anomaly detection and clinical decision support techniques for automated detection of cardiovascular diseases and COVİD-19
Coronary artery diseases are the leading cause of death worldwide, and early diagnosis is crucial for timely treatment. To address this, we present a novel automated artificial intelligence-based hybrid anomaly detection technique composed of various signal processing, feature extraction, supervised, and unsupervised machine learning methods. By jointly and simultaneously analyzing 12-lead electrocardiogram (ECG) and cardiac sympathetic nerve activity (CSNA) data, the automated artificial intelligence-based hybrid anomaly detection technique performs fast, early, and accurate diagnosis of coronary artery diseases. To develop and evaluate the proposed automated artificial intelligence-based hybrid anomaly detection technique, we utilized the fully labeled STAFF III and PTBD databases, which contain 12-lead wideband raw recordings non-invasively acquired from 260 subjects. Using the wideband raw recordings in these databases, we developed a signal processing technique that simultaneously detects the 12-lead ECG and CSNA signals of all subjects. Subsequently, using the pre-processed 12-lead ECG and CSNA signals, we developed a time-domain feature extraction technique that extracts the statistical CSNA and ECG features critical for the reliable diagnosis of coronary artery diseases. Using the extracted discriminative features, we developed a supervised classification technique based on artificial neural networks that simultaneously detects anomalies in the 12-lead ECG and CSNA data. Furthermore, we developed an unsupervised clustering technique based on the Gaussian mixture model and Neyman-Pearson criterion that performs robust detection of the outliers corresponding to coronary artery diseases. By using the automated artificial intelligence-based hybrid anomaly detection technique, we have demonstrated a significant association between the increase in the amplitude of CSNA signal and anomalies in ECG signal during coronary artery diseases. The automated artificial intelligence-based hybrid anomaly detection technique performed highly reliable detection of coronary artery diseases with a sensitivity of 98.48%, specificity of 97.73%, accuracy of 98.11%, positive predictive value (PPV) of 97.74%, negative predictive value (NPV) of 98.47%, and F1-score of 98.11%. Hence, the artificial intelligence-based hybrid anomaly detection technique has superior performance compared to the gold standard diagnostic test ECG in diagnosing coronary artery diseases. Additionally, it outperformed other techniques developed in this study that separately utilize either only CSNA data or only ECG data. Therefore, it significantly increases the detection performance of coronary artery diseases by taking advantage of the diversity in different data types and leveraging their strengths. Furthermore, its performance is comparatively better than that of most previously proposed machine and deep learning methods that exclusively used ECG data to diagnose or classify coronary artery diseases. It also has a very short implementation time, which is highly desirable for real-time detection of coronary artery diseases in clinical practice. The proposed automated artificial intelligence-based hybrid anomaly detection technique may serve as an efficient decision-support system to increase physicians' success in achieving fast, early, and accurate diagnosis of coronary artery diseases. It may be highly beneficial and valuable, particularly for asymptomatic coronary artery disease patients, for whom the diagnostic information provided by ECG alone is not sufficient to reliably diagnose the disease. Hence, it may significantly improve patient outcomes, enable timely treatments, and reduce the mortality associated with cardiovascular diseases. Secondly, we propose a new automated artificial intelligence-based hybrid clinical decision support technique that jointly analyzes reverse transcriptase-polymerase chain reaction (RT-PCR) curves, thorax computed tomography images, and laboratory data to perform fast and accurate diagnosis of Coronavirus disease 2019 (COVID-19). For this purpose, we retrospectively created the fully labeled Ankara University Faculty of Medicine COVID-19 (AUFM-CoV) database, which contains a wide variety of medical data, including RT-PCR curves, thorax computed tomography images, and laboratory data. The AUFM-CoV is the most comprehensive database that includes thorax computed tomography images of COVID-19 pneumonia (CVP), other viral and bacterial pneumonias (VBP), and parenchymal lung diseases (PLD), all of which present significant challenges for differential diagnosis. We developed a new automated artificial intelligence-based hybrid clinical decision support technique, which is an ensemble learning technique consisting of two preprocessing methods, long short-term memory network-based deep learning method, convolutional neural network-based deep learning method, and artificial neural network-based machine learning method. By jointly analyzing RT-PCR curves, thorax computed tomography images, and laboratory data, the proposed automated artificial intelligence-based hybrid clinical decision support technique benefits from the diversity in different data types that are critical for the reliable detection of COVID-19 and leverages their strengths. The multi-class classification performance results of the proposed convolutional neural network-based deep learning method on the AUFM-CoV database showed that it achieved highly reliable detection of COVID-19 with a sensitivity of 91.9%, specificity of 92.5%, precision of 80.4%, and F1-score of 86%. Therefore, it outperformed thorax computed tomography in terms of the specificity of COVID-19 diagnosis. Moreover, the convolutional neural network-based deep learning method has been shown to very successfully distinguish COVID-19 pneumonia (CVP) from other viral and bacterial pneumonias (VBP) and parenchymal lung diseases (PLD), which exhibit very similar radiological findings. Therefore, it has great potential to be successfully used in the differential diagnosis of pulmonary diseases containing ground-glass opacities. The binary classification performance results of the proposed convolutional neural network-based deep learning method showed that it achieved a sensitivity of 91.5%, specificity of 94.8%, precision of 85.6%, and F1-score of 88.4% in diagnosing COVID-19. Hence, it has comparable sensitivity to thorax computed tomography in diagnosing COVID-19. Additionally, the binary classification performance results of the proposed long short-term memory network-based deep learning method on the AUFM-CoV database showed that it performed highly reliable detection of COVID-19 with a sensitivity of 96.6%, specificity of 99.2%, precision of 98.1%, and F1-score of 97.3%. Thus, it outperformed the gold standard RT-PCR test in terms of the sensitivity of COVID-19 diagnosis. Furthermore, the multi-class classification performance results of the proposed automated artificial intelligence-based hybrid clinical decision support technique on the AUFM-CoV database showed that it diagnosed COVID-19 with a sensitivity of 66.3%, specificity of 94.9%, precision of 80%, and F1-score of 73%. Hence, it has been shown to very successfully perform the differential diagnosis of COVID-19 pneumonia (CVP) and other pneumonias. The binary classification performance results of the automated artificial intelligence-based hybrid clinical decision support technique revealed that it diagnosed COVID-19 with a sensitivity of 90%, specificity of 92.8%, precision of 91.8%, and F1-score of 90.9%. Therefore, it exhibits superior sensitivity and specificity compared to laboratory data in COVID-19 diagnosis. The performance results of the proposed automated artificial intelligence-based hybrid clinical decision support technique on the AUFM-CoV database demonstrate its ability to provide highly reliable diagnosis of COVID-19 by jointly analyzing RT-PCR data, thorax computed tomography images, and laboratory data. Consequently, it may significantly increase the success of physicians in diagnosing COVID-19, assist them in rapidly isolating and treating COVID-19 patients, and reduce their workload in daily clinical practice.
Bilinen sinyal işleme problemlerine eniyileme çerçevesinde yeni çözümler
Novel approaches for three classical signal processing problems in optimization framework are proposed to provide further flexibility and performance improvement. In the first part, a new technique, which uses Hermite-Gaussian (HG) functions, is developed for analysis of signals, whose components have non-overlapping compact time-frequency supports. Once the support of each signal component is properly transformed, HG functions provide optimal representations. Conducted experiments show that proposed method provides reliable identification and extraction of signal components even under severe noise cases. In the second part, three different approaches are proposed for designing a set of orthogonal pulse shapes for ultra-wideband communication systems with wideband antennas. Each pulse shape is modelled as a linear combination of time shifted and scaled HG functions. By solving the constructed optimization problems, high energy pulse shapes, which maintain orthogonality at the receiver with desired timefrequency characteristics are obtained. Moreover, by showing that, derivatives of HG functions can be represented as a linear combination of HGs, a simple optimal correlating receiver structure is proposed. In the third part, two different methods for phase-only control of array antennas based on semidefinite modelling are proposed. First, antenna pattern design problem is formulated as a non-convex quadratically constraint quadratic problem (QCQP). Then, by relaxing the QCQP formulation, a convex semidefinite problem (SDP) is obtained. For moderate size arrays, a novel iterative rank refinement algorithm is proposed to achieve a rank-1 solution for the obtained SDP, which is the solution to the original QCQP formulation. For large arrays an alternating direction method of multipliers (ADMM) based solution is developed. Conducted experiments show that both methods provide effective phase settings, which generate beam patterns under highly flexible constraints.
Gürbüz sıkıştırılmış algılama teknikleri
Compressive Sensing theory details how a sparsely represented signal in a known basis can be reconstructed from an underdetermined linear measurements. However, in reality there is a mismatch between the assumed and the actual dictionary due to factors such as discretization of the parameter space defining basis components, sampling jitter in A/D conversion, and model errors. Due to this mismatch, a signal may not be sparse in the assumed basis, which causes significant performance degradation in sparse reconstruction algorithms. To eliminate the mismatch problem, this thesis presents two novel robust algorithm and an adaptive discretization framework that can obtain successful sparse representations. In the proposed techniques, the selected dictionary atoms are perturbed towards directions to decrease the orthogonal residual norm. The first algorithm named as Parameter Perturbed Orthogonal Matching Pursuit (PPOMP) targets the off-grid problem and the parameters of the selected dictionary atoms are perturbed. The second algorithm named as Perturbed Orthogonal Matching Pursuit (POMP) targets the unstructured basis mismatch problem and performs controlled rotation based perturbation of selected dictionary atoms. Based on detailed mathematical analysis, conditions for successful reconstruction are derived. Simulations show that robust results with much smaller reconstruction errors in the case of both parametric and unstructured basis mismatch problem can be obtained as compared to standard sparse reconstruction techniques. Different from the proposed perturbation approaches, the proposed adaptive framework discretizes the continuous parameter space depending on the estimated sparsity level. Once a provisional solution is obtained with a sparse solver, the framework recursively splits the problem into sparser sub-problems so that each sub-problem is exposed to less severe off-grid problem. In the presented recursive framework, any sparse reconstruction technique can be used. As illustrated over commonly used applications, the error in the estimated parameters of sparse signal components almost achieve the Cramer-Rao lower bound in the proposed framework.
EKG kayıtlarının öznitelikleri kullanılarak akut koroner sendromların otomatik olarak erken teşhisi
In patients with acute coronary syndrome, temporary chest pains together with changes in the ST/T sections of ECG occur shortly before the start of myocardial infarction. In order to diagnose acute coronary syndromes early, we propose a new technique which detects changes in the ST/T sections of ECG. For this purpose, by using real ECG recordings, we identify ECG features that are critical in the detection of acute coronary syndromes. By using support vector machines (SVM) operating with linear and radial basis function (RBF) kernels, we obtain classifiers that use 2 or 3 most discriminating features of the ST/T sections. To improve performance, classification results on multiple segments are fused. The obtained results over a considerable number of patients indicate that the proposed classification technique provides highly reliable detection of acute coronary syndromes. To develop a detection technique that can be used in the absence of unhealthy ECGs, we also investigate the detection of acute coronary syndromes based on ECG recordings of a patient obtained during healthy stage only. For this purpose, a Gaussian mixture model is used to represent the joint pdf of the selected features. Then, Neyman-Pearson type of approach is developed to provide detection of outliers that would correspond to acute coronary syndromes
Kerte arıtımı ile dışbükey eniyileme tabanlı FIR süzgeç tasarımı
Finite impulse response filters have been one of the primary topics of digital signal processing since their inception. Consequently, diverse class of design techniques including Chebyshev approximation, Fast Fourier Transform, and optimization based methods, had been proposed in the literature. With developments in computational tools, new design technique tools and formulations on filters including interior-point solvers and semidefinite programming (SDP), emerged. Since FIR filter design problem can be modelled as a quadratically constrained quadratic program, filter design problem can be solved via interior-point based convex optimization methods such as semidefinite programming. Unfortunately, SDP formulation of problem is nonconvex due to positive lower limit constraint in the passband. To overcome that problem, nonconvex problem can be cast into a convex SDP using semidefinite relaxation, which can be solved in polynomial time. Since relaxed formulation does not guarantee rank-1 solution matrix, recently proposed directed iterative rank refinement (DIRR) algorithm is used to impose a convex rank-1 constraint. Due to utilization of semidefinite relaxation and DIRR, addition of various constraints, such as phase and group delay masks, in convex manner is made possible. For feasibility type optimization formulations of filter design problem, a convergence rate improved version of DIRR is developed. Proposed techniques are applied on filter design problems with different set of constraints including phase and group delay constraints. Explicit simulations demostrate that the proposed technique is capable of solving nonlinear phase, phase constrained, and group delay constrained filter design problems.
IRI-Plas modeli ve YKS ölçümleri kullanarak iyonkürede 3 boyutlu elektron yoğunluğu kestirimi
Three dimensional imaging of the electron density distribution in the ionosphere is a crucial task for investigating the ionospheric effects. Dual-frequency Global Positioning System (GPS) satellite signals can be used to estimate the Slant Total Electron Content (STEC) along the propagation path between a GPS satellite and ground based receiver station. However, the estimated GPS-STEC are very sparse and highly non-uniformly distributed for obtaining reliable 3D electron density distributions derived from the measurements alone. Standard tomographic reconstruction techniques are not accurate or reliable enough to represent the full complexity of variable ionosphere. On the other hand, model based electron density distributions are produced according to the general trends of the ionosphere, and these distributions do not agree with measurements, especially for geomagnetically active hours. In this thesis, a novel regional 3D electron density distribution reconstruction technique, namely IONOLAB-CIT, is proposed to assimilate GPS-STEC into physical ionospheric models. The IONOLAB-CIT is based on an iterative optimization framework that tracks the deviations from the ionospheric model in terms of F2 layer critical frequency and maximum ionization height resulting from the comparison of International Reference Ionosphere extended to Plasmasphere (IRI-Plas) model generated STEC and GPS-STEC. The IONOLAB-CIT is applied successfully for the reconstruction of electron density distributions over Turkey, during calm and disturbed hours of ionosphere using Turkish National Permanent GPS Network (TNPGN-Active). Reconstructions are also validated by predicting the STEC measurements that are left out in the reconstruction phase. The IONOLAB-CIT is compared with the real ionosonde measurements over Greece, and it is shown that the IONOLAB-CIT results are in good compliance with the ionosonde measurements. The results of the IONOLAB-CIT technique are also tracked and smoothed in time by using Kalman filtering methods for increasing the robustness of the results.
Algılayıcı dizilimi kullanarak varış açısı sezimi ve takibi için sıkıştırma matrisi dizaynı
Direction of Arrival (DoA) estimation is extensively studied in the array signal processing with many applications areas including radar, sonar, medical diagnosis and radio astronomy. Since, in sparse target environments, Compressive Sensing (CS) provides comparable performance with the classical DoA estimation techniques by using fewer number of sensor outputs, there are a multitude of proposed techniques in the literature that focus on surveillance (detection) and tracking (estimation) of DoA in CS framework. Many of such works elaborate on recovery of compressed signal and employ random measurement matrices, such as Bernoulli or Gaussian matrices. Although random matrices satisfy Restricted Isometry Property (RIP) for reconstruction, the measurement matrices can be designed to provide improved performance in search sectors that they are designed for. In this thesis, a novel technique to design compressive measurement matrices is proposed in order to achieve enhanced DoA surveillance and tracking performance using sensor arrays. Measurement matrices are designed in order to minimize the Cramer-Rao Lower Bound (CRLB), which provides a lower bound for DoA estimation error. It is analytically shown that the proposed design technique attains the CRLB under mild conditions. Built upon the characteristics of proposed measurement design approach, a sequential surveillance technique using interference cancellation is introduced. A novel partitioning technique, which provides a greedy type solution to a minmax optimization problem, is also developed to ensure robust surveillance performance. In addition, an adaptive target tracking algorithm, which adaptively updates measurement matrices based on the available information of targets, is proposed. Via a comprehensive set of simulations, it is demonstrated that the proposed measurement design technique facilities significantly enhanced surveillance and tracking performance over the widely used random matrices in the compressive sensing literature.
Gürbüz minimum-maksimum kestiricinin kalman filtresine uyarlanması
Kalman filtering is one of the most essential tools in estimating an unknown state of a dynamic system from measured data, where the measurements and the previous states have a known relation with the present state. It has generally two steps, prediction and update. This filtering method yields the minimum mean-square error when the noise in the system is Gaussian and the best linear estimate when the noise is arbitrary. But, Kalman filtering performance degrades significantly with the model uncertainty in the state dynamics or observations. In this thesis, we consider the problem of estimating an unknown vector x in a state-space model that may be subject to uncertainties. We assume that the model uncertainty has a known bound and we seek a robust linear estimator for x that minimizes the worst case mean-square error across all possible values of x and all possible values of the model matrix. Robust minimax estimation technique is derived and analyzed in this thesis, then applied to the state-space model and simulation results with different noise perturbation models are presented. Also, a radar tracking application assuming a linear state dynamics is also investigated.Modifications to the James-Stein estimator are made according to the scheme we develop in this thesis, so that some of its limitations are dealt with. In our scheme, James-Stein estimation can be applied even if the observation equation is perturbed and the number of observations are less than the number of states, still yielding robust estimations.
Sürü zekası kullanılarak yeni dizilim sinyal işleme tekniklerinin geliştirilmesi
In this thesis, novel array signal processing techniques are proposed for identificationof multipath communication channels based on cross ambiguity function(CAF) calculation, swarm intelligence and compressed sensing (CS) theory. Firsttechnique detects the presence of multipath components by integrating CAFs ofeach antenna output in the array and iteratively estimates direction-of-arrivals(DOAs), time delays and Doppler shifts of a known waveform. Second techniquecalled particle swarm optimization-cross ambiguity function (PSO-CAF) makesuse of the CAF calculation to transform the received antenna array outputs todelay-Doppler domain for efficient exploitation of the delay-Doppler diversity ofthe multipath components. Clusters of multipath components are identified byusing a simple amplitude thresholding in the delay-Doppler domain. PSO isused to estimate parameters of the multipath components in each cluster. Thirdproposed technique combines CS theory, swarm intelligence and CAF computation.Performance of standard CS formulations based on discretization of the multipathchannel parameter space degrade significantly when the actual channelparameters deviate from the assumed discrete set of values. To alleviate this?off-grid?problem, a novel technique by making use of the PSO, that can also beused in applications other than the multipath channel identification is proposed.Performances of the proposed techniques are verified both on sythetic and realdata.
Sentetik arazi üzerinde darbe-doppler radarı kullanarak yapılan arazi profili kestirimi
The systems used for terrain profile estimation arise when the safety flight issues in civil flight transport and in military applications become important. These systems are developed for the purpose of terrain avoidance and safe flight. In this thesis, we study two techniques in estimating the terrain profile of the synthetically generated terrain which is achieved by means of signal processing. The estimation performance of the techniques is observed according to the results of flight simulations realized on the simulation environment. In the simulations, an aircraft with a pulse-Doppler radar scans a synthetic terrain according to the scanning patterns to generate the received signals. The techniques that we propose, are applied to the output of the pulse-Doppler process. The first technique is based on the usage of the first and the middle reflection range points in the clutter received signal. An adaptive thresholding method is developed for robust detection of these points. Accurate detection of these range points is crucial in the estimation performance of the first approach. The other technique uses the relation between the elevation angle theta and the clutter received signal amplitude ratio of the two receiver antennas R1 and R2 in finding the theta angles of the reflections in corresponding range values. In this approach, accurate estimation of the angle of arrival is important on the performance of estimation. Especially for far ranges, the errors in the estimation become more sensitive to the errors in the elevation angle theta. Finally, over a set of synthetically generated terrain profiles, the error performance of these two techniques are investigated and compared.
Belirsiz denklem sistemleri
In this thesis, new theoretical and practical results on linear equations with various types of uncertainties and their applications are presented. In the first part, the case in which there are more equations than unknowns (overdetermined case) is considered. A novel approach is proposed to provide robust and accurate estimates of the solution of the linear equations when both the measurement vector and the coefficient matrix are subject to uncertainty. A new analytic formulation is developed in terms of the gradient flow to analyze and provide estimates to the solution. The presented analysis enables us to study and compare existing methods in literature. We derive theoretical bounds for the performance of our estimator and show that if the signal-to-noise ratio is low than a treshold, a significant improvement is made compared to the conventional estimator. Numerical results in applications such as blind identification, multiple frequency estimation and deconvolution show that the proposed technique outperforms alternative methods in mean-squared error for a significant range of signal-to-noise ratio values. The second type of uncertainty analyzed in the overdetermined case is where uncertainty is sparse in some basis. We show that this type of uncertainty on the coefficient matrix can be recovered exactly for a large class of structures, if we have sufficiently many equations. We propose and solve an optimization criterion and its convex relaxation to recover the uncertainty and the solution to the linear system. We derive sufficiency conditions for exact and stable recovery. Then we demonstrate with numerical examples that the proposed method is able to recover unknowns exactly with high probability. The performance of the proposed technique is compared in estimation and tracking of sparse multipath wireless channels. The second part of the thesis deals with the case where there are more unknowns than equations (underdetermined case). We extend the theory of polarization of Arikan for random variables with continuous distributions. We show that the Hadamard Transform and the Discrete Fourier Transform, polarizes the information content of independent identically distributed copies of {\it compressible} random variables, where compressibility is measured by Shannon's differential entropy. Using these results we show that, the solution of the linear system can be recovered even if there are more unknowns than equations if the number of equations is sufficient to capture the entropy of the uncertainty. This approach is applied to sampling compressible signals below the Nyquist rate and coined "Polar Sampling". This result generalizes and unifies the sparse recovery theory of Compressed Sensing by extending it to general low entropy signals with an information theoretical analysis. We demonstrate the effectiveness of Polar Sampling approach on a numerical sub-Nyquist sampling example.
Bir boyutlu ve iki boyutlu sinyallerin polinom uyumu ve toplam değişime dayalı gürültü bastırma teknikleri
New techniques are developed for signal denoising and texture recovery. Geometricaltheory of total variation (TV) is explored, and an algorithm that usesquadratic programming is introduced for total variation reduction. To minimizethe staircase effect associated with commonly used total variation basedtechniques, robust algorithms are proposed for accurate localization of transitionboundaries. For this boundary detection problem, three techniques are proposed.In the first method, the 1?D total variation is applied in first derivative domain.This technique is based on the fact that total variation forms piecewise constantparts and the constant parts in the derivative domain corresponds to lines intime domain. The boundaries of these constant parts are used as the transitionboundaries for the line fitting. In the second technique proposed for boundarydetection, a wavelet based technique is proposed. Since the mother wavelet canbe used to detect local abrupt changes, the Haar wavelet function is used for thepurpose of boundary detection. Convolution of a signal or its derivative familywith this Haar mother wavelet gives responses at the edge locations, attaining local maxima. A basic local maximization technique is used to find the boundarylocations. The last technique proposed for boundary detection is the wellknown Particle Swarm Optimization (PSO). The locations of the boundaries arerandomly perturbed yielding an error for each set of boundaries. Pursuing thepersonal and global best positions, the boundary locations converge to a set ofboundaries. In all of the techniques, polynomial fitting is applied to the part ofthe signal between the edges.A more complicated scenario for 1?D signal denoising is texture recovery. Inthe technique proposed in this thesis, the periodicity of the texture is exploited.Periodic and non-periodic parts are distinguished by examining total variationof the autocorrelation of the signal. In the periodic parts, the period size wasfound by PSO evolution. All the periods were averaged to remove the noise, andthe final signal was synthesized.For the purpose of image denoising, optimum one dimensional total variationminimization is carried to two dimensions by Radon transform and slicingmethod. In the proposed techniques, the stopping criterion for the procedures ischosen as the error norm. The processes are stopped when the residual norm iscomparable to noise standard deviation. 1?D and 2?D noise statistics estimationmethods based on Maximum Likelihood Estimation (MLE) are presented.The proposed denoising techniques are compared with principal curve projectiontechnique, total variation by Rudin et al, total variation by Willsky et al, andcurvelets. The simulations show that our techniques outperform these widelyused techniques in the literature.
Gürbüz kısıtlı gauss karışım modellerinin enbüyük olabilirlik kestirimi
Density estimation using Gaussian mixture models presents a fundamental trade o between the exibility of the model and its sensitivity to the unwanted/unmodeled data points in the data set. The expectation maximization (EM) algorithm used to estimate the parameters of Gaussian mixture models is prone to local optima due to nonconvexity of the problem and the improper selection of parameterization. We propose a novel modeling framework, three dierent parameterizations and novel algorithms for the constrained Gaussian mixture density estimation problem based on the expectation maximization algorithm, convex duality theory and the stochastic search algorithms. We propose a new modeling framework called Constrained Gaussian Mixture Models (CGMM) that incorporates prior information into the density estimation problem in the form of convex constraints on the model parameters. In this context, we consider two dierent parameterizations where the rst set of parameters are referred to as the information parameters and the second set of parameters are referred to as the source parameters. To estimate the parameters, we use the EM algorithm where we solve two optimization problems alternatingly in the E-step and the M-step. We show that the M-step corresponds to a convex optimization problem in theinformation parameters. We form a dual problem for the M-step and show that the dual problem corresponds to a convex optimization problem in the source parameters. We apply the CGMM framework to two dierent problems: Robust density estimation and compound object detection problems. In the robust density estimation problem, we incorporate the inlier/outlier information available for small number of data points as convex constraints on the parameters usingthe information parameters. In the compound object detection problem, we incorporate the relative size, spectral distribution structure and relative location relations of primitive objects as convex constraints on the parameters using the source parameters. Even with the propoper selection of the parameterization,density estimation problem for Gaussian mixture models is not jointly convex in both the E-step variables and the M-step variables. We propose a third parameterization based on eigenvalue decomposition of covariance matrices which is suitable for stochastic search algorithms in general and particle swarm optimization (PSO) algorithm in particular. We develop a new algorithm where global search skills of the PSO algorithm is incorporated into the EM algorithm to doglobal parameter estimation. In addition to the mathematical derivations, experimental results on synthetic and real-life data sets verifying the performance ofthe proposed algorithms are provided.Key Words: Gaussian mixture models, expectation maximization, convex optimization, duality, particle swarm optimization.
SAR görüntüleme problemleri için yeni metodlar
Synthetic Aperture Radar (SAR) provides high resolution images of terrain reflectivity. SAR systems are indispensable in many remote sensing applications. High resolution imaging of terrain requires precise position information of the radar platform on its flight path. In target detection and identification applications, imaging of sparse reflectivity scenes is a requirement. In this thesis, novel SAR image reconstruction techniques for sparse target scenes are developed. These techniques differ from earlier approaches in their ability of simultaneous image reconstruction and motion compensation. It is shown that if the residual phase error after INS/GPS corrected platform motion is captured in the signal model, then the optimal autofocused image formation can be formulated as a sparse reconstruction problem. In the first proposed technique, Non-Linear Conjugate Gradient Descent algorithm is used to obtain the optimum reconstruction. To increase robustness in the reconstruction, Total Variation penalty is introduced into the cost function of the optimization. To reduce the rate of A/D conversion and memory requirements, a specific under sampling pattern is introduced. In the second proposed technique, Expectation Maximization Based Matching Pursuit (EMMP) algorithm is utilized to obtain the optimum sparse SAR reconstruction. EMMP algorithm is greedy and computationally less complex resulting in fast SAR image reconstructions. Based on a variety of metrics, performances of the proposed techniques are compared. It is observed that the EMMP algorithm has an additional advantage of reconstructing off-grid targets by perturbing on-grid basis vectors on a finer grid. Keywords: Synthetic Aperture Radar, Phase Error Correction, Compressed Sensing, Total Variation, Expectation Maximization Based Matching Pursuit.
İyonküre elektron içeriği kullanılarak deprem öncül tespit başarımı incelenmesi
Ionosphere region of Earth's upper atmosphere ranging from 90 km to 1000 km altitude, has a significant effect on military and civilian communications, satellite communications and positioning systems. Solar, geomagnetic, gravitational and seismic activities cause variations in the electron distribution of the atmosphere. The number of electrons within a vertical column of 1 m2 cross section, which is called as Total Electron Content (TEC), is a measurable feature of the ionosphere that provides valuable information about the ionosphere. TEC can be measured fast and accurately by using the phase difference between transmitted satellite positioning signals such as in the Global Positioning System (GPS). To investigate the reliability of earthquake prediction based on detection of local ionospheric anomalies, TEC measurements obtained from a network of GPS receivers over a period of 2 years in 2010 and 2011 are used to generate detection signals. For a day of interest, after selecting a receiver station surrounding GPS stations that are located within 150 km of the chosen station used to estimate TEC measurements at the chosen station. In one of the proposed techniques, detection of ionospheric anomalies is based on distance between measured TEC and its estimate. Detection threshold is obtained based on statistical variation of this distance for the days with insignificant seismic activities. Also, another detection technique based on temporal variation of TEC measurements is proposed. Both individual and fused detection performances of these techniques are investigated for a given level of false alarms. It is observed that the fused detection has superior performance and able to detect 15 out of 23 earthquakes of magnitude larger than 5 in Richter scale while generating 8 false alarms.
Anlamsal ve hedefe yönelik sinyal işleme: Anlamsal çıkarma
Advances in machine learning technology have enabled real-time extraction of semantic information in signals, which has the potential to revolutionize signal processing techniques and drastically improve their performance for next-generation applications. A graph-based semantic language and a goal-oriented semantic signal processing framework are adopted for structured and universal representation and efficient processing of semantic information. In the adopted framework, preprocessing of input signals is followed by a semantic extractor which identifies components from a set of application-specific predefined classes where the states, actions, and relations among the identified components are described by another application-specific predefined set called predicates. For additional information, the resulting semantic graph is also embedded with a hierarchical set of attributes. In this thesis, we focus on the crucial semantic extractor block, and to illustrate the proposed framework's applicability, we present a real-time computer vision application on video-stream data where we adopt a tracking by detection paradigm for the identification of semantic components. Next, we show that with the adopted semantic representation and goal-filtering, the semantic signal processing framework can achieve an extremely high reduction in data rates compared to traditional approaches. Finally, we demonstrate a way to identify points of significant innovation over extended periods of time by tracking the evolution of multi-level attributes and discussing future research directions.