Ph.D. Theses
Permanent URI for this collection
Browse
Recent Submissions
Item Novel time-series based DDOS attack detection schemes for traditional networks and software defined networks(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Fouladi, Ramin Fadaei.; Anarım, Emin.Distributed Denial of Service (DDoS) attacks are always one of the most signifi cant threats for computer networks since they affect the user satisfaction by degrading the availability of on-line services. Although some countermeasures such as Intrusion Detection Systems (IDSs) provide effective mechanisms to discriminate various types of DDoS attacks, they become impotent of detection when bogus packets similar to normal ones are dispatched by the attacker. One promising approach for the DDoS detection in traditional networks is to use the time-series representation of the network traffic while analyzing the incoming packets. Particularly, discriminating features are extracted from the representation of the traffic flow in order to be used with several data analytic techniques such as statistical measures or machine learning algorithms. In this thesis, we first improve the previous works in the literature for the traditional networks by introducing three methods using frequency domain analysis and statistical measures. Later, we extend our findings for SDNs and we propose three different DDoS detection and countermeasure schemes for SDN by employing: (i) Auto-Regressive Integrated Moving Average and a dynamic thresholding method, (ii) Discrete Wavelet Transform and Auto-Encoder Networks, and (iii) Continuous Wavelet Transform and Convolu tional Neural Network. Experimental results show that proposed schemes have high detection and low false alarm rates. Finally, we compare proposed schemes in terms of their attack detection performance and computational complexity cost analysis.Item Multi-modal tensor representations of brain networks(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Durusoy, Göktekin.; Acar, Burak.Considering the economic, social and psychological burdens of Alzheimer’s disease (AD), the most common form of dementia, it is essential to gain insight into the process and underlying mechanisms of the disease. Using structural and functional brain connectomes obtained by in-vivo MRI techniques as biomarkers is a promising approach. In this thesis, the B-Tensor structure that allows the representation of brain connectomes defined in structurally and functionally with a uni-modal and multi-modal fashion is presented. With the projection of structural connectomes onto known func tional networks, patients with AD and healthy control group are distinguished in a 7-dimensional space with a separation performance of over 90%. In addition, with the uni-modal and multi-modal tensor factorization methods, 47 patients with different levels of AD, are diagnosed with an accuracy of 77% - 100% in a 5-dimensional space. The results show that the multi-modal factorization technique performs better than the uni-modal one by successfully fusing the structural and functional networks which offer complementary information. While the neurological evaluations of the obtained sub-networks are highly consistent with previous literature, new findings regarding the progression of the disease are also recommended.Item DDoS attack detection using signal processing and statistical approaches(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Erhan, Derya.; Anarım, Emin.DDoS attacks cause a variety of changes in the properties of the attributes in the network traffic. Modeling these changes using signal processing and statistical ap proaches provides detection of these attacks. This thesis focuses on detecting DDoS attacks using time series analysis, sparse signal representation methods, and statistical modeling. We also investigate the effect of DDoS attacks on traffic features in a sta tistical manner. In addition, we propose two simple but effective network-based DDoS attack detection methods based on the statistical signal processing approach, using the advantage of statistical changes in traffic features. We propose a novel DDoS detection framework using the Matching Pursuit algo rithm to detect resource depletion type DDoS attacks. We use multiple characteristics of network traffic simultaneously to detect low-density DDoS attacks efficiently. The proposed method uses the dictionary produced from the parameters of the network traf fic using the K-SVD algorithm. Dictionary generation using network traffic provides legitimate and attack traffic models and adds adaptability to the proposed method to network traffic. We also implement DDoS detection approaches that use Matching Pursuit and Wavelet techniques and compare them using two different data sets. Addi tionally, we offer a hybrid DDoS detection framework that combines these approaches with a decision-making mechanism using an artificial neural network. We evaluate the proposed methods with two different data sets. In the hybrid intrusion detection sys tem with more than one attack, the detection performances of other approaches have decreased. In contrast, the proposed method achieves true-positive rates higher than 99% with a false positive rate lower than 0.7%.Item Multilayer perceptron neural network in analog VLSI-A system level study(Thesis (Ph.D.)- Bogazici University. Institute for Graduate Studies in Science and Engineering, 1999., 1999.) Öğrenci, Arif Selçuk.; Dündar, Günhan,; Balkır, Sina.Analog neural networks exhibit a potential for proper/suitable hardware implementation of artificial neural networks. Their advantages such as small size, low power and high speed, however, are seriously questioned/confronted by the difficulty in the training of analog neural network circuitry. Especially, hardware training by software, i.e., training of the circuitry by software based on models, so as to avoid on-chip and chip in-the-loop training methods, is threatened by circuit nonidealities and variations at outputs of identical blocks. The performance of the analog neural network is severely degraded in the presence of those unwanted effects caused mainly by statistical variations in the production process. We propose a new paradigm for the backpropagation algorithm in hardware training of multilayer perceptron type analog neural networks. The variations at outputs of analog neural network circuitry are modeled based on the transistor level mismatches occurring between identically designed transistors. Those variations can be used as additive noise during the training, and it has been shown that this will increase fault tolerance of the trained neural network drastically. The method has been compared to the method of injecting random noise, and our method outperforms the latter where injecting random noise is seen to be inadequate for establishing a satisfactory level of fault tolerance in the presence of mismatch based variations. The concept of mismatch based variations has been verified by measurements on our test chip.Item Multi-dimensional yield-aware optimization of the analog and heterogeneous circuits(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Pak, Murat.; Dündar, Günhan,Even though the design of digital circuits is well supported by several CAD tools, this is not the case for analog circuits and MEMS where the design is typically hand-crafted by expert designers. Several tools have been implemented for trade-off exploration in analog circuits; however, yield-optimization is a hot and vital topic considering that process variations have deteriorated along with the feature sizes scaling down; hence, physical variations originating from manufacturing process have a huge impact on yield. Therefore, efficient yield-aware optimization methodologies for analog ICs are needed. MEMS design, on the other hand, requires a lot of expert knowledge, which implies long design times and increased cost due to this physical heterogeneity. The approach followed by the industry, based on composing separately designed sensors and read-out circuitry, has several issues such as inappropriate partitioning of system specifications or potential violation of system level constraints during the coupling process of the these devices. Hence, design methodologies which can obtain globally optimal MEMS by performing co-optimization of the sensor and the circuit are needed. This study is mainly focused on developing and implementing novel and generic design methodologies for multi-objective yield-aware optimization of analog circuits and MEMS. A novel yield optimization technique has been proposed and compared with the existing approaches and has provided very promising results for yield-aware Pareto Front generation. Besides the work conducted for yield-aware optimization, co-optimization of MEMS and analog circuits has been performed for the first time by jointly optimizing a mechanical accelerometer sensor and an electronic read-out circuitry. The implemented yield-aware optimization techniques have been integrated into the co-optimization loop to enable yield-aware multi-objective optimization of MEMS.Item Design, fabrication and characterization of micromachined THz absorbers(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Mamaghani, Ayda.; Yalçınkaya, Arda Deniz.; Torun, Hamdi.The purpose of this research is to develop a new class of THz absorbers for imaging and sensing applications that can provide improved functionality and perfor mance. These absorbers consist of periodic arrays of layered cells made of a back-plane metal layer, opaque to transmitted radiation, a dielectric substrate as a spacer, and a metallic patterned layer as resonating part. Once electromagnetic wave impinges, the structure reflects the incident wave, except in a specific frequency band determined by the absorber’s physical properties. In the presence of a blocking layer which eliminates transmitted power, absorption in that specific frequency band occurs. The effective per mittivity and permeability of absorber unit cell reach the same value and be negative simultaneously to have perfect absorption. Negative values of and µ do not happen in nature. At this point, ”metamaterials” are introduced. Metamaterials (MMs) are artificially engineered structures used in THz absorbers. MMs’ concept is derived from replacing the natural materials with engineered materials where their sizes are much smaller than the given wavelength. Intriguing properties of MM are achieved from the degree of the skillfulness of its structure (geometry, shape, orientation, and size), while its chemical construction plays an insignificant role. Currently, THz absorbers are in use as medical and security applications. Many types of MM based absorbers have been introduced based on their shapes, sizes, and configurations. This research has focused on absorbers with a low dependency of absorber’s performance on the exciting wave’s incidence and polarization angle. Besides, we demand these structures to operate in a wide range of frequency spectra designed to, with small dimensions of the patterned array compared to the wavelength; in other words, we concentrate on the design of wideband THz absorbers. In Chapter 1, we start with a brief introduction to THz radiation and its benefits and applications. It also comprises an overview of the history and theory of MMs. Var ious designs demonstrating the performance and operating frequency bands of MMs and their applications have been studied. A comprehensive study of wave propagation in the right and left-handed media is presented in Chapter 2. Basic models of proposed MM absorbers are introduced in Chapter 3. The principle of resonance is based on a lumped LC circuit and broadband operation provided in this chapter. Afterward, the design, Fabrication, and characterization of a MEMS-based THz metamaterial based absorber are described in Chapter 4. A new 3D design inspired by honeycomb structures as broadband and incident wave independent absorber has been proposed. Chapter 5 includes the design, simulation, fabrication process, and measurement re sults of this 3-dimensional MM based broadband absorber. The proposed absorber exhibits excellent absorption characteristics. The unique feature of the proposed ap proach is having very low sensitivity with respect to the incidence angle. The ability to maintain their absorption properties in THz frequencies, unlike their small feature sizes compared with traditional approaches, made these porous structures attractive for sensing applications. Throughout this dissertation, both physical and numerical interpretations of the spectral behavior of new designs are provided. Physical insight into the operating mechanism is the key to further improvements and creates more complex surfaces with desired frequency responses.Item Estimating biological changes in human by different markers from RF exposure using modeling and measurements(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Uluaydın, Niyazi Korkut.; Çiçekoğlu, Oğuzhan.; Şeker, Selim.This thesis dissertation reviews the works of academia on natural occurring electrical, magnetic, and electromagnetic (EM) phenomena and their subtle relations with living organisms. It also investigates the artificial electromagnetic waves from different generations of mobile technology radio frequency (RF) exposure with their possible paths of impact. It proposes new scenarios of mobile technology usage through finite element method (FEM) simulations with new radio possibilities of mobile terminals and base stations in view of new antenna technologies. The model theoretical limitations are cross-checked by Mie Theory. The FEM simulations are first time oriented according to organ specific effect investigations in IEEE SAM phantom model. The simulations are also improved in terms of organ details such as eye-based ones with detailed initial conditions. This dissertation also first time projects the known EM effects on salivary glands by biological markers over hypothalamus-pituitary-adrenal (HPA) axis and pineal glands, which are similar in biological and electrical properties and cannot be studied without invasive methods. This is a solid non-thermal effect proposition on the homeostasis of human beings. This dissertation proposes a detailed measurement of the multi-operator multi technology urban base stations with spectrum and antenna aware methodology. Combining this with the statistics of mobile radio network maintenance activities with their periodicity and duration, which have never been reported before in academia, it provides a totally different perspective for occupational RF exposure and increased risks. Result of both probable non-thermal and thermal risks are clear findings new to the academia.Item Machine heath monitoring for cyber-physical systems(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Aydemir, Gürkan.; Acar, Burak.Estimating the failure time of the machinery that are used in the production is crucial to achieve an e cient maintenance in Industry 4:0 era. Remaining useful life (RUL) is the term that refers to the length of time in terms of the raw time intervals or usage that a machine will continue to operate before it requires a repair or replacement. Machine learning (ML), especially deep learning, provides industry practitioners with e cient tools for estimating the RUL. However, ML is far from being fully utilized, since domain knowledge is generally ignored in current studies. This thesis focuses on three main domain speci c problems in machine condition monitoring to improve the performance of ML based RUL estimation. First, RUL is ill-de ned during the healthy operation period of the machinery, hence enforcing ML with respect to a ctitious true RUL during these periods adversely a ects the overall RUL estimation accuracy. In this thesis, a system level anomaly detection triggered RUL estimation method is proposed to detect degradation onset point in sensor data to prevent ML models to estimate RUL in this period, and hence to increase the accuracy. Secondly, the operating conditions of the machines a ect their degradation pattern and related sensor measurements. Thus, the accuracy of ML based RUL estimation models decreases when the machinery operate in varying conditions. A siamese neural network based operating conditioninvariant feature extraction method is introduced to alleviate this problem. These two approaches are veri ed using a benchmark turbofan engine degradation data. Lastly, most of the ML models su er from lack of data in RUL estimation. If the data are high dimensional such as image, pro le, etc., the problem becomes more challenging. Two deep learning architectures are proposed to resolve curse of dimensionality in case of degradation data scarcity. E ciency of the proposed models is demonstrated with an infrared image data.Item Split ring resonators in microwave regime for sensing of reagents in aqueous solutions(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Çamlı, Berk.; Yalçınkaya, Arda Deniz.In this thesis, design, fabrication, and characterization of different microwave band split ring resonator-based sensors used for sensing of glucose in aqueous solu tions are presented. Following an introduction to metamaterials and split ring res onators, a literature survey of their applications in sensing is given. After this, an analytical model that describes the operation of a single loop SRR is demonstrated along with additional theoretical considerations. Exploratory work in which simulated and measured electromagnetic interaction of different split ring resonator geometries with dielectric loads is reported. Change of resonance characteristics of the split ring resonator is shown to be correlated with different dielectric loads, such as aqueous glucose solutions of different concentrations. After the exploratory work, a biosensor application incorporating a glucose specific element glucose oxidase immobilized in a PEDOT:PSS matrix is discussed. The biosensor application had a sensitivity of 0.107 MHz/mg mL−1 to glucose. Its specificity was demonstrated by comparison of its re sponse to other reagents, such as sucrose, fructose, and NaCl. Precision, linearity, and repeatability improvements to this design was done by adoption of conceptual develop ments by switching to a differential measurement scheme, use of loop antennas instead of monopole antennas, and incorporation of microfluidic elements to the system. An improved iteration of the original sensor system was presented at the end, fabricated from materials highly compatible with relevant biosensor applications.Item Improving the error floor performance of LDPC codes(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Sarıduman, Abdullah.; Pusane, Ali Emre.; Taşkın, Zeki Caner.Low-density parity-check codes (LDPC) have become one of the most popular error-correcting code families in recent years due to their high error correction performance and easy implementation. They became one of the biggest candidates to become a standard in the next-generation wireless communication systems. In particular, quasi-cyclic (QC)-LDPC codes have been chosen as the standard codes for 5G mobile broadband. However, at high signal-to-noise ratio values, the error-correcting performance of LDPC codes decreases due to harmful structures called trapping sets. Designing an LDPC code with few or no harmful structures is one of the popular topics in recent years. In this thesis, rstly, an adaptive linear decoder is designed that can decode LDPC codes with high error-correcting performance. Then, simulated annealing algorithms are proposed to reduce the number of cycles and trapping sets. Finally, new QC-LDPC codes have been designed with the proposed algorithm. In terms of trapping sets distribution, the best short-length QC-LDPC code constructions in the literature have been achieved.Item A high voltage triboelectric energy harvesting system utilizing parallel-SSHI rectifier and DC-DC converters for SUB-5 hz motions(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Kara, İsmail.; Mutlu, Şenol.; Doğan, Hakan.In this thesis, the first integrated circuit (IC) implementation of parallel synchronized switching harvesting on inductor (parallel-SSHI) is presented for triboelectric energy harvester targeting 1 Hz to 5 Hz mechanical motions. It is accompanied by on-chip buck and switched-capacitor DC-DC converters, all capable of handling 70 V levels. Unlike piezoelectric harvesters, triboelectric nanogenerators (TENGs) can produce very high open-circuit voltages; thus, the proposed system utilizes this property within the technology limits to maximize the extracted power. An in-house manufactured TENG using steel and polytetra uoroethylene (PTFE) is modeled for sub-5 Hz motions. The energy is extracted and stored in an external capacitance until its voltage reaches 70 V, which is achieved in three press-and-release mechanical cycles. 70-to-2 V down conversion is carried on by a 70-to-10 V buck converter followed by a 10-to-2 V switched-capacitor DC-DC converter. A chip is manufactured in TSMC 0.18 m HV BCD process with an active area of 6.25 mm2. End-to-end peak e ciency is measured as 32.71% for 1 Hz motion with a 722 W total power delivery to the load for 4 ms.Item Practical channel coding methods for channels with input-dependent noise(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Ülkar, Mehmet Görkem.; Pusane, Ali Emre.With the emergence of new applications and needs, communications theory started to be applied to outside of the traditional radio frequency (RF) bands or even in completely di erent channels. Molecular communication and visible light communication (VLC) are examples of such emerging use cases that communication takes place in a speci c channel. Not surprisingly, those channels bring their own challenges and di erences compared to the conventional wireless channels. One common point among many of those new channels is that the noise depends on the input signal. This situation is contrary to the prevalent assumption of existence of white noise in the design of wireless communication blocks. Since white noise assumption is not valid, applying directly conventional methods to the new channels yields unsatisfactory performances. In this thesis work, our aim was to develop practical channel coding methods for the channels with input dependent noise. For molecular communication via di usion (MCvD), we propose 2 novel decoding methods coupled with constant low weight codes. Iterative sorting decoder is a decision-feedback heuristic method that iteratively calculates the intersymbol interference (ISI) from a better estimation at each step. The second proposed method is the super trellis decoder, which is a maximum a posteriori sequence estimator. Iterative sorting and super trellis decoders bring substantially better performances than the existing methods. For VLC, a deep learning based VLCnet is proposed. VLCnet has a novel activation unit, FRAU, to achieve icker reduction and dimming, which are two main illumination needs. By allowing joint optimization of both the encoder and decoder, VLCnet performs superior compared to the other proposed techniques in the literature.Item Development of a high level synthesis tool specialized on fir-based multirate systems(Thesis (Ph.D.)- Bogazici University. Institute for Graduate Studies in Science and Engineering, 1999., 1999.) Yurdakul, Arda.; Dündar, Günhan,; Tansal, Sabih.Digital Signal Processing (DSP) is the most studied area in design automation, because it is one of the most well-established branches of electrical engineering for several years. In the last few years, it is stimulated by the progression of multirate techniques. The key property of multirate algorithms is their computational efficiency. In this thesis, a silicon compiler is developed to reduce design time for the hardware realization of FIR-based multirate DSP algorithms. This is a brand new study, because there does not exist a silicon compiler of this type according to our knowledge. Although multirate algorithms contain decimators and interpolators changing the effective sample rate, the design of synchronous systems using a single-clock signal is possible by this newly developed tool. The designer can achieve this by folding nodes of similar type into a single node. Additionally, the FIR filters followed by a decimator or following an interpolator can be entered as a single node while defining a system at the input of the tool. Also multiplications with the tap coefficients in FIR-based nodes in a fold are handled at the same time to exploit common terms so as to realize those multiplications without multipliers. As a result, the tool produces very efficient layouts in terms of area, power and clock signals. It can also determine the quantization levels of tap coefficients in FIR-based nodes and fractional parts of data bus if the system output error is specified. It also handles module selection under given power, area and delay constraints and scheduling like other well-known silicon compilers. The compiler is programmed to process bit-parallel-digit-serial architectures.Item Stable - matching based resource allocation methods in wireless communication systems(Thesis (Ph.D.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Sabuncu Sandal, Yağmur.; Pusane, Ali Emre.; Karabulut Kurt, GüneşAs the number of smart devices increases day by day, the need for new resource allocation techniques also increases. 5G technologies such as Long-Term Evolution Advanced (LTE-A), carrier aggregated heterogeneous networks (HetNets), or Internet of things (IoT) networks with mobile edge computing (MEC) features require high data rates and ultra low latency with the restricted resources. Resource allocation and secure communication are two of the most important challenges in wireless communication. Graph-based algorithms are proposed in order to achieve resource allocation with a low computational complexity. The primary objective of this thesis is to address the resource allocation challenges, e.g., fairness and stability, by using stable matching (SM)-based approaches. SM algorithm requires channel state information (CSI) before starting allocation. However, CSI transmission may cause an overload of the up-link channel. The problem is extensively elaborated in many aspects for different wireless communication systems such as HetNets and IoT networks. The overload on the uplink channel, through CSI transmission, is decreased significantly by the proposed partial feedback matching (PFM) algorithm. Finally, IoT networks are very fragile against attackers in physical layer as the number of connected smart devices increases. A three state SM-based attacker identification and punishment policy, is proposed in order to increase the robustness of the network. Both analytical and simulation results are presented to demonstrate that the proposed SM-based approaches have better performance than the algorithms that exist in the literature.Item Design of high efficiency switching converters for mobile applications(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Ozanoğlu, Kemal.; Dündar, Günhan,High power efficiency is a key design specification in mobile applications, mainly to increase battery life. To answer this need, switching converters are preferred in such applications to convert the battery voltage to voltage domains of various blocks. In this thesis, design improvements and novel solutions aiming to increase power efficiency and to enhance system performance for mobile platform switching converters are proposed for buck, boost, and buck-boost topologies. After a brief description of switching converter operation, a novel technique to improve power efficiency in buck converters is given, through optimized resistive and capacitive power losses of the output stage. Then, a charge recycling technique for single inductor dual output buck converters is described. Next, two improved control techniques for buck-boost converters based on hysteretic control and current mode control have been proposed. Finally, two novel techniques addressing the lock-out phenomenon occurring in boost and buck-boost converters are described. Simulation results show that the targeted performance improvements are achieved, thus demonstrating promising solutions for various future mobile platforms.Item Novel signal processing techniques for fiber optic distributed acoustic sensing(Thesis (Ph.D.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Ölçer, İbrahim.; Öncü, Ahmet.In this thesis, to reduce the effect of noise in phase-sensitive optical time domain based (φ-OTDR) distributed acoustic sensing (DAS) systems, two novel approaches are proposed and a real experimental φ-OTDR system is developed for validation. The first approach is the temporal adaptive processing of φ-OTDR signals which is based on maximizing the signal-to-noise ratio (SNR) at the output of an adaptive linear filter. When the vibration frequency of interest is known a priori, it is called the adaptive matched filter (AMF). The second approach is based on the largest eigenvalue computation of the optical covariance matrix which does not require any prior information about the vibration frequencies. Both methods utilize the correlation properties of the measured data. In the first method, the noise covariance matrix is estimated to compute an adaptive weight vector for optimum linear filtering. In the second method, the eigenvalues of the covariance matrix are computed and the maximum eigenvalue is used as the test statistic for detecting the vibrations along the fiber optic cable route. This so called maximum eigenvalue detection (MED) technique is assisted by the random matrix theory (RMT) to establish the binary detection threshold. First, the efficacy of the proposed methods was demonstrated with Monte Carlo simulations. In the second phase, a φ-OTDR system was developed in the laboratory to gather real data and to verify the AMF and MED techniques with indoor experiments. In the last phase, extensive field tests, with both buried fibers and fibers on fences, were carried out to validate the proposed techniques in real-world conditions. The results show that more than 20 dB of SNR values can be achieved without any reduction in the system bandwidth and using any optical amplifier stage in the hardware.Item Unsupervised learning of word alignments for statistical machine translation(Thesis (Ph.D.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Mermer, Coşkun.; Saraçlar, Murat.; Sarıkaya, Ruhi.Word alignment is a crucial first step in learning statistical translation models. In this dissertation, we propose a Bayesian approach to unsupervised learning of word alignments by introducing a sparse prior on the parameters of IBM word alignment models. In the original approach, word translation probabilities are estimated using the expectation-maximization (EM) algorithm. In the proposed approach, they are random variables with a prior and are integrated out during inference, where collapsed Gibbs sampling is used. The inferred word alignments are evaluated in a statistical ma chine translation (SMT) setting, experimenting with several language pairs and sizes of corpora and comparing against the EM and variational Bayes (VB) methods. We show that Bayesian inference outperforms both EM and VB in the majority of test cases, effectively addresses the high-fertility rare word problem in EM and unaligned rare word problem in VB, achieves higher agreement and vocabulary coverage rates than both, and leads to smaller phrase tables. We also propose a method for un supervised learning of the optimal segmentation for SMT. We augment the original Morfessor monolingual segmentation model with a word alignment model so that the new model optimizes the posterior probability of the parallel training corpus according to a generative segmentation-translation model. In order to speed up computation, we propose an incremental method for approximate translation likelihood calculation and a parallelizable search algorithm, which improves the performance of even the mono lingual segmentation. We use the proposed method to segment the Turkish side in a Turkish-to-English SMT system and find that the bilingual model results in more intuitive segmentations but does not yield a further significant increase in BLEU scores.Item Magnetically actuated MEMS scanners for laser scanning confocal microscopy(Thesis (Ph.D.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Oyman, Hilmi Artun.; Yalçınkaya, Arda Deniz.; Gökdel, Yiğit Dağhan.Magnetically actutated micro-scanners are fabricated via three-dimensional (3D) printingandlasermachiningtechnologiestouseinalaserscanningconfocalmicroscopy application. First, a 3D printed polymer based scanning mirror with magnetic actuation is designed to meet a confocal microscopy application providing 100 µm × 100 µm field of view and less than 3-µm lateral resolution. Stress distribution along the circular-profiledflexureiscomparedwitharectangularcounterpartinfinite-elementenvironment. The resonance frequencies of the device were analytically modeled. Finally, imagingexperimentswereconductedonaresolutiontarget,showcasingthedesiredscan area and resolution. In the second part of thesis, laser machining technology is used to produce stainless steel the micro-scanners. First device is developed for a 2D confocal imaging application. This device tested with the United States Air Force target accomplishing a 200 µm × 200 µm field of view and sub 10 µm resolution. In the following study, a micro-scanner with multi-gimbaled structure is designed and produced for 3D Lissajous confocal imaging application. This device can work in three different out-of-plane modes in order to control the focus of the confocal system. Final study contains a micro-scanner for 3D beam steering with better performance specifications, such as higher TOSA for less current consumption, increased resonance frequencies and smaller total size of the device as opposed to the previously designed micro-scanners. Also, fabricated micro-scanner is integrated in a confocal system and 2D image of a biological sample; convallaria rhizome, is obtained. Furthermore, a novel 3D confocal microscopy configuration is proposed and it’s validty is tested for 3D beam steering on a custom confocal setup.Item Molecular communication in diffusion based channel(Thesis (Ph.D.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2018., 2018.) Akdeniz, Bayram Cevdet.; Pusane, Ali Emre.This thesis deals with diffusion based molecular communication. Unlike conven tional communication systems, the information is encoded with molecules and these molecules are emitted by nanodevices in a diffusive environment. Since the molecules diffuse through the environment, their movement is governed by Brownian Motion, resulting in very slow and random movement that leads to excessive interference com pared to conventional communication channels. In this thesis, various modulation, equalization and coding schemes are proposed to combat with this interference issue. The main motivation for the determination of these schemes is proposing computa tionally simple and/or sparse communication methods suitable for nanodevices. All proposed methods diminish the interference at the received signal and improve the performance of the molecular communication channels compared to the other exist ing methods in the literature. In particular, pulse position modulation is adopted for molecular communication channel to diminish interference without any additional complexity and less channel information compared to other proposed modulations. A specific sparse channel code for molecular communication is also proposed by present ing its improved performance. For equalization, the received signal is equalized by analytically determining the optimum reception delay analytically that minimizes in terference and maximizes the signal power. With the same goal, optimum aperture region of the receiver is determined by deriving the joint angle-time distribution of the absorbed molecules at the receiver. Finally, network coding methods are proposed for two-way one-hop and multi-hop nanonetworks.Item Equivalence checking of designs modeled in simulink implemented in low level(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2018., 2018.) Sağlamdemir, Muharrem Orkun.; Dündar, Günhan,; Şen, Alper.We propose methodologies for checking the equivalence of analog and digital Simulink models and their low level implementations. For digital methodology, we develop a flow utilizing commonly used tools used in digital integrated circuit design. Digital model can be designed by built-in Simulink libraries or user-defined functions. These models are converted to register-transfer level (RTL) implementations by us age of either Matlab’s HDL Coder tool or Real Time Workshop followed by Catapult. Manual RTL implementation is compared to the converted RTL via Formality tool to decide equivalence. For the analog methodology, equivalence is decided by comparing predetermined performance parameters from simulations of model and the circuit level implementation. Simulations are done for different input parameters, which charac terize the design. By means of this, a space of input parameters is found, where the model and the circuit level implementation are equivalent. For the analog methodol ogy, we utilize Matlab’s optimizer firstly, then we develop an evolutionary computation approach, which is a modified version of SPEA2 algorithm. We utilize Quasi Monte Carlo method to generate the samples, which are the input parameter set. This makes it possible to reach the result with less samples. Finally, we integrate process variation analysis to the equivalence checking methodology. To the best of our knowledge, the methodologies we propose for both digital and analog models are the first for check ing equivalence for Simulink. Also the analog equivalence checking methodology with evolutionary computation can be applied to non-Simulink models. We validate digital methodology on designs distributed with Matlab package and Advanced Encryption Standard (AES). Proposed analog methodologies and process variation analysis are validated on inverter, operational amplifier and buck converter.