Abstract

In view of the stringent requirements of measurement of capacitances associated with capacitive sensors, there is considerable work going on in the area of VLSI design. Typically MEM based capacitive sensors used in gyroscopes produce changes of atto Farads in nominal capacitance values of the order of pico Farads. The requirements of the measuring equipment are very stringent like the noise conditions, the temperature stability and frequency response etc. There are no readily available interface circuits to measure these order of capacitances. In view of this an attempt has been made to analyze the various issues that are associated with measuring the small capacitances. In the first instance the capacitance variations are converted to voltage variations. Even though there are several approaches for carrying out this operation it is found that a switched capacitor approach with DC source voltage is a suitable technique . In the first stage, amplification of the signal is carried out by minimizing the noise, temperature variations and offset variations. Keeping these in view a switched capacitor fully differential folded cascode amplifier with an open loop gain of 120dB has been designed and simulated using HSPICE simulator. The clock frequency chosen for this operation is much higher than the rate at which the capacitance changes. The high linearity of 0.05% in the maximum output of 0.6V could be achieved only through extensive simulations by varying circuit components. Finally it has been possible to arrive at a circuit with correlated double sampling, common mode feedback and optimal values of feedback capacitors. This circuit is designed as a low noise circuit to give the required resolution such that it can measure a small voltage of 10μV corresponding to 10aF. Since the circuit itself contributes to flicker noise the effect of flicker noise is countered by employing the correlated double sampling and designing the amplifier for low noise. The output of the amplifier is given to a sigma delta modulator so that the noise can be tackled and a digital output of high resolution is obtained. The sigma delta modulator is chosen to be a fourth order cascaded modulator which is realized using two second order modulators operating at same clock frequency as the main switched capacitor amplifier. The integrated blocks of switched capacitor amplifier and sigma delta modulator has been found to give a linearity of 0.05% with a maximum output value of 0.6V and minimum output value of 20μV. This thesis mainly concentrated on these two blocks and also suggests that these two blocks can be integrated in to single chip. Further processing of output of sigma delta modulator involves a process called decimation which converts the sigma delta modulator output into accurate digital output.

Abstract

The growth of Internet, like no other communication medium, has given a "Globalized dimension" to the world. It has become the Universal source of information for millions of people, at home, at school, and at work. Due to numerous contributors across the world, the amount of information available on the web is huge and is distributed across various languages. Hence, there is a need to develop applications to manage such huge amount of varied information. Multilingual Document Clustering (MDC) is one such technique which is highly useful for organizing huge amount of data present in different languages. One of its main applications is Cross-lingual Information Retrieval (CLIR), where search engine takes query in one language and retrieves results in different languages. Instead of providing results as a single long list, it would be very convenient for users if the search engine retrieves them as a list of clusters, where each cluster contains multilingual documents that are similar. It encourages users to opt for cluster based browsing which is very convenient for processing the results. Existing approaches in Multilingual document clustering uses various tools and resources such as Manually Annotated (MA) bilingual dictionaries, lemmatizers, Named Entity Recognizers (NERs), etc., for processing the data. However, many languages across the world do not have the luxury of such tools and resources. Hence, it is very difficult to process and organize the data of languages which do not have sufficient resources. In this thesis, we focus on clustering multilingual documents of resourcepoor languages. We propose three main approaches which do not make use of any language dependent resources or tools for multilingual document clustering. In the rst approach, we perform clustering of multilingual documents using Wikipedia as external knowledge. We avail various features of Wikipedia such as cross-lingual links, categories, outlinks, Infobox information, etc., to enrich the document representation. It helps in comparing two multilingual documents eciently for forming good multilingual clusters. The use of external knowledge improves the clustering performance because it considers the semantic association between the words. In the second approach, we propose a method to identify the Named Entities (NEs) present in resource-poor languages such as Hindi and Marathi. We use the Named Entities present in English, which is a resource-rich language for this purpose. The identied Named Entities help in detecting the topic of a given document, which is highly useful for comparing two documents efficiently for improving the performance of multilingual document clustering. In our final approach, we propose a method which combines our first and second approaches. In this approach, Named Entities and Wikipedia features are used together in enriching the document representation for forming good multilingual clusters. In all our approaches we use Bisecting k-means algorithm for forming multilingual clusters. We do not make use of any non-English linguistic tools or resources such as WordNet, Part-Of-Speech tagger etc., which makes our proposed work easy to extend to other languages. Experiments are conducted on English, Hindi and Marathi news datasets provided by FIRE1 for their 2010 Ad-hoc Cross-Lingual document retrieval task on Indian languages. All our approaches are evaluated using F-score, Purity and Normalized Mutual Information (NMI) measures and the results obtained are encouraging.

Abstract

There are many ways in which science is learnt by students: classroom lectures, real physical activities or computer generated interactions. Computer based simulation tools offer students a means to explore, experience and express themselves through computer generated interactions. Simulation tools are used for enhancing the learning by offering a different way of presenting the information via visualization. In simulation tools, the students can pose hypotheses about a science concept, conduct as many experiments as they want, observe and record data and draw conclusions. This might not always be possible in a real laboratory because of many reasons. The physical phenomenon could itself be complicated to observe or the experiment setup could be time consuming. Laboratory experiments are conducted in a more step-by-step manner following processes of examination and observation, which are very important to build necessary scientific skills in a student. Hence with a simulation tool students can miss out critical examination and observation skill required in a real laboratory. Therefore many researchers have suggested that the simulation can be very useful if used for experiencing before classroom instruction, and applying learned contents after classroom instruction. The classroom instruction and guidance by teachers are very important for making simulation tool an effective learning tool. Development of simulation tools for science education has been explored widely in literature. The results from these efforts indicate that the comfort level of a user with simulations is influenced by their prior experiences with other software environments. Hence for a teacher to be comfortable with a simulation tool, it needs to be customized to their preferences. Development and modification of simulations according to the specific need of teachers, while desirable, is a difficult and time-consuming process as it may require an iterative tailoring of the user interface to meet multiple constraints like their teaching style and students’ learning style. The developer must learn a software package, design the visual composition (menu, tools) and temporal behavior (stored program or task) of user interface to meet requirements. Hence, an authoring tool which allows a user to create various simulations and is easy to use will be an effective educational resource for content developers. An authoring tool, refers to a system that allows a user to create content, by defining the relationship between different media (i.e., text, video, images, etc.) and sequencing them in an appropriate order. In this thesis, we present such an authoring tool, ViPLab, to understand different components required in an authoring tool. We present the design details of the physics engine and user interface. This tool has been evaluated by science teachers and a number of undergraduate engineering students. Based on the preliminary feedback from teachers and students a set of guidelines are identified for authoring tool development. We propose an authoring tool framework that adheres to these guidelines by decoupling the user interface from the physics engine. This enables the integration of user interface with different physics engines. We present a Graphical interface construction Kit (GicK) which helps in creating user interfaces and combine it with a physics engine for creating interactive simulations. A 3D-ViPLab, an authoring tool, is developed as an exemplar of the proposed framework. We end by showing a way to convert simulations developed using 3D-ViPLab into simulation games.

Abstract

The importance of transformers, with their role in transmission and distribution of electrical power and with the effect of their performance on the system, is an obvious axiom in the modern day’s power systems. In addition to their momentous share in the capital investment of a power system, transformer outages have a considerable economic impact on the operation of the power systems. In the course of continuous efforts to make the existing power network smarter and efficient, thermal modeling and monitoring of transformers has become important in the field of transformer engineering. With all the advances in the design techniques as well as material engineering, it is the transformer thermal limitations that decide the loading and designing of the transformer from the purview of user as well as manufacturer. With the research in power systems on the whole progressing towards development of a ‘smart grid’, which infers that each of the equipment should be ‘smart’, that includes that the monitoring of each individual equipment should be intelligent, accurate as well as fast and economical, the problem of thermal Modeling of transformers has been gaining momentum all the more. The maximum temperature in the transformer interior is a significant parameter governing a transformer’s performance and life expectancy. Though the temperature rise in the transformer interior by itself may not have immediate effects, it does trigger other undesirable consequences like excessive deterioration of insulation, which in the long run will reduce the life of the transformer, thus affecting the economics of the power system. Thus the possible maximum temperature rise in the transformer for certain kind of loading needs to be estimated so as to be able to decide on the operational conditions as well as estimate the remaining life of the transformer and plan accordingly. In the perspective of the user, temperatures in a transformer are important to determine the amount and duration of over load it can sustain, and to estimate the effects on the life of the transformer by operation at various temperatures. For a transformer design engineer, prediction of temperatures at various points becomes necessary to determine the amount of copper to place in the coils, leads and outlet bushings, type of cooling and ducts, position of ducts, insulation class, design and settings of control equipment. Apart from this, increased market competency demands for accurate determination of the thermal profile across the transformer, which might result in a more economical as well as efficient manufacturing. Existing thermal models calculate the winding hotspot temperature and top oil temperature using the lumped values of heat generation inside the transformer and the rate of heat transfer and retention in the surrounding media that finally result in the temperature rise. The heat generation is due to the energy losses in the transformer which are the iron losses in the core and ohmic losses in the coils. These temperatures served as an index for the interior temperature rise in the transformer. To calculate the hotspot temperatures, the existing models used the lumped values of losses and lumped values of heat transfer and retention in the different media that surrounded the heat generating elements and the loss distribution across the transformer geometry was not calculated and used in those models. However, advancements in computing capabilities and ever ongoing research enables better transformer interior temperature modeling, which may be a better indicator of transformer thermal status. In the current work, the use of finite element analysis technique was made to calculate the loss distribution across the transformer geometry, which is a different approach. With the calculation of loss distribution across the transformer geometry, the current work proposes a new approach for thermal model of transformer and discusses the development of this thermal model that aims at computing the interior temperatures at different as well as desired points across the transformer geometry. The proposed thermal model has been successfully implemented on four real transformer data to calculate the thermal profiles of transformers that show the real life use of proposed thermal model.

Abstract

The selection of a subset of ’interesting’ regions in a given image forms an important first step in many computer vision algorithms, for example for computing image correspondences , scene recognition, object detection or for learning object categories. The algorithms that are used to detect and represent these subset of regions are known as interest point operators. In order to develop an interest point operator, two key issues need to be addressed: how to select the subset of the image for subsequent analysis and how to represent the subset? In this thesis we examine both the issues of detection and representation. The first topic of this thesis deals with the problem of interest point/region detection. Several interest point/region detectors such as Harris corner detector, SIFT, Maximally Stable Extrema Regions have been developed, all of which by and large rely on some ad hoc filters to quantify ”interest”. This could be due to the fact that the challenging aspect of this problem is to quantify the ”interestingness” of a region in the given image. We address this topic by looking into the mechanisms adopted by Human Visual System (HVS) for identifying visual regions of interest. It is believed that HVS constantly relies on saccades to analyze the scene in front of it in a serial fashion, attending to the most relevant/interesting regions first. Saccades are common eye movements that are used to shift gaze, scan visual scenes or track moving targets. The thesis analyzes the ways in which existing interest point operators and systems developed for predicting saccades are related to each other. The latter rely more on feature contrast known to be influencing HVS in contrast to ad hoc approaches used in the former. This has prompted us to study the role of features like local symmetry, curvature and phase in predicting saliency. The role of each feature is quantified by adding them to an existing feature-based model for predicting saccades and analyzing how well the resultant prediction correlate with fixation maps. The results indicate that of the features studied, phase only reconstruction of an image correlates best with HVS fixations. Phase only reconstruction is one of several spectral domain methods reported recently for predicting interesting/salient regions. We establish a generalization of these spectral domain methods using a simple algorithm known as Frequency self filtering which has been used for automatic image inpainting. The second topic of the thesis deals with the problem of interest region representation. The representation of a region is supposed to capture the spatial arrangement of the pixels while preserving invariance to common distortions, in a compact format. The typical trade-off here is between the detail the representation can capture and the robustness to distortions. Popular representations like SIFT, vGiLOH, Shape Context etc, rely on histograms of edge-like features computed over predefined grids. The amount of spatial distribution information captured by these representations is very much dependent on the granularity of the grid over which histograms are computed. As an alternate way of encoding spatial distribution information we propose a hypothesis that spatial distribution information can be captured using texture information in the projection domain. In order to test this hypothesis a representation based on Radon transform and local binary patterns is developed. This is tested in three different scenarios, namely binary shape retrieval, gray scale image matching and abnormality classification in medical images. In all the three scenarios the proposed representation has shown good performance. The results indicate that spatial distribution information can be encoded into representation using the proposed hypothesis.

Abstract

With increase in technology scaling, semiconductor devices are becoming increasingly less predictable. The device failure rates increase significantly with decrease in device sizes. Memory unit usually has more hardware components than other circuits and therefore is the most unreliable part. The SRAM memory cell has highest density and is more vulnerable to defects compared to other logic circuits. As the technology scales the process variations also affect the functionality of the memory. There are computer applications where extreme reliability of computing system is essential. Even with advances in manufacturing technology to reduce the faults occurring, the errors in memory are still present. So, methods to increase reliability by providing some internal fault-tolerance are gaining importance in memories. This thesis focuses on developing a memory architecture which enables self test of the memories after manufacturing and avoids the faulty rows by incorporating additional test circuits. The effectiveness of the proposed circuits is checked on a word oriented two port memory. We assume that a single fault in any memory cell within a particular row makes the entire row faulty. During the isolation and replacement, the entire row is isolated and replaced with a healthy row. To carry out the replacement, some spare rows are provided in the memory array. A row can have a single or multiple faulty cells within it; it is replaced by a spare row. The fault model considered in this thesis is the stuck-at-fault i.e., stuck-at-0 and stuck-at-1 faults occurring in the SRAM cell. The stuck-at-faults present in the SRAM cell alone are considered in this thesis for testing, isolation and replacement process. The idea has been demonstrated through the simulation for a word oriented two port memory architecture and additional circuits were designed which incorporate all the above discussed facilities. In a word oriented memory the entire row is written or read at a time in a clock cycle. In the two port memories, the read and write ports are completely independent and hence the read and write operations can be performed simultaneously though simultaneous reading and writing of same row cannot be performed. The architecture proposed in this thesis consists of word oriented two port SRAM memory cells consisting of 4 words, and each word consists of 4bits. The memory further has additional circuits for fault detection, fault isolation and replacement of faulty rows. During testing it is assumed that a Built-in Self Test (BIST) controller is present which includes circuits to generate the signals necessary to start and end the test mode in the memory; a Test Pattern Generator (TPG) to generate the necessary test addresses and test input vectors necessary to carry out the testing during the test mode. The BIST controller signals the start of testing process at every power-on of the memory. First the testing is carried which provides the input vectors which sensitize the fault present in the memory and allows to identify them. A compare circuit is used to identify the faulty row(s) in the memory. The signal generated from the compare circuit is used by the isolation circuit to isolate the faulty row from the SRAM array. The same signal generated from the compare circuit is used by the replacement circuit to redirect the signals corresponding to the faulty row to the healthy row. Additional spare rows are provided in the memory architecture which are used during the replacement process. The self healing structure proposed in this thesis enables one to have reliable memory after the testing and replacement process. It helps in improving the memory yield which directly translates into chip yield and chip reliability. The circuits that are proposed and tested through simulation are designed with the same technology as the SRAM memory is designed. Hence it is possible to integrate them with the SRAM memory. The architecture is designed at circuit level and the simulations are carried using HSPICE. The circuit is designed at 65nm technology node and the 65nm Predictive Technology Model (PTM) is used as the model for simulation. Thus the thesis proposes a self healing architecture through BIST and additional circuits proposed for memories which may result in considerable improvement in yield of memories manufactured in the latest technology nodes.

Abstract

The importance of environment modelling as a prerequisite for the navigation of mobile robots can- not be understated. It is essential for a mobile robot to have the ability to autonomously build a model of the environment, as human intervention is not practical in all circumstances. The unavailability of environment modelling algorithms until recently, has hampered the feasibility of the envisaged vision of mobile robotics. A robot has to address both the mapping and the localization problems in order to au- tonomously solve the map learning problem. Due to the coupling between these two tasks, autonomous map learning has been a challenging problem. This thesis is primarily focused on the implementation of an environment modelling technique based on the Extended Kalman Filter(EKF). The mapping and localization problem, called Simultaneous Localization And Mapping(SLAM) is defined as the process of concurrently building a map of the environment while using this map to obtain improved estimates of the robot location. The SLAM algorithm begins with the mobile robot in an unknown location, with no a priori knowledge of the map. The robot then uses a sensor to model the visible environment and simul- taneously computes an estimate of its own position. Continuing in motion, the mobile robot eventually builds a model of the environment which is useful for navigation. In spite of the ample literature available on SLAM, there is hardly any literature on the development of a real SLAM system. A majority of the work in this thesis is attributed to filling the gap between theory and implementation by the identification and diagnosis of the issues faced during implementation. The work in this thesis can be divided into two sections: contributions in the implementation front and those in the theoretical front. The implementation part deals with the system specific customization of the algorithm, a novel calibration method for the sensor with respect to the robot, extraction and parametrization of features, provision for manual exploration during map building among other issues. This part of the thesis concludes with successful demonstrations of robust EKF-SLAM implementation on the Pioneer P3-Dx mobile robot using a SICK LMS200 range scanning sensor. The theoretical work in this thesis deals with better map building by novel parametrizations of features to keep error estimates low, representative features for better association and trajectory estimation from the online SLAM problem.

Abstract

The present work attempts to build an automatic translation system of nominal compound (NC) from English to Hindi. A noun compound is a sequence of nouns acting as a single noun, e.g., colon cancer, suppressor protein, colon cancer tumor suppressor protein. They comprise 3.9% and 2.6% of all tokens in the Reuters corpus and the British National Corpus (BNC), respectively. As of today, no good system exists for the translation of multi-word expressions from English to any Indian languages. We have evaluated two state-of-the-art systems, Moses and Google Translation system, to check the Noun Compound translation accuracy from English to Hindi. Google translation system results in an accuracy of 57% while Moses, a statistical machine translation system, returns an accuracy of 48% on a test data of 300 Noun Compounds. The above figures indicate that automatic NC translation from English to Hindi is an important subtask of machine translation system. We build a Noun Compound Translation system (NCT) which returns an accuracy of 64% on the same set of test data. This thesis examines two approaches for translation of Noun Compounds from English to Hindi. We have done a manual study on 50K parallel sentences from English to Hindi and have found out that Noun Compounds in English are translated into Noun Compound in Hindi in over 40% of the cases. In other cases they are translated into varied syntactic constructs. Among them the most frequent construction type is “Modifier + Post-Position + Head” which occurs in 35% of all the cases. Some examples are “cow milk” ! “gAya kA dUXa”, “wax work” ! “mOMa para ciwroM”. This observation motivates both the approaches for translation in the present thesis. The approaches are called in this work as: a) Translation of NC by paraphrasing on source side and mapping the paraphrase to target construct and b) Context based translation by searching and ranking translation candidate on target side. In the first approach English nominal compounds are automatically paraphrased and the paraphrases are translated into Hindi constructions. The paraphrasing is done with prepositions following [Lauer 1995] approach of paraphrasing of nominal compound. For example, “cow milk” is paraphrased as ‘milk from cow’, “blood sugar” is paraphrased as ‘sugar in blood’. Since English prepositions have one-to-one mapping to post-position in Hindi, English paraphrases are easily translated into Hindi using the mapping schema. Assuming that lexical substitution for component nouns of the compound is correct, this method examines how paraphrasing of English nominal compound acts as an aid for translation. In the second approach, we, at first, generate translation templates for the target language. These templates are all possible Hindi construction types that English nominal compounds can be translated into. Context based translation system take context into consideration while translating. We translate noun compound by taking the sentence in which the compound occurs as the context. For example, the expression “finance minister” is the nominal compound to be translated in the sentence “The finance minister declared the financial budget for this year”. Other content words in the sentence such as ‘declared’, ‘financial’, ‘budget’, ‘year’ form the context. We apply a Word-sense-disambiguation tool for selecting the correct sense of the component nouns of NC in the given context. We use a bilingual dictionary to get the Hindi translation of the component nouns in the sense selected by WSD tool. Thus context based lexical substitution is accomplished for the target language. The output of lexical substitution is placed in the translation templates and the resulted construction is searched on a Hindi indexed corpus of 28 million words. For ranking, a reference ranking based on the frequency of occurrence of the translate candidates in full in the TL corpora is taken as baseline. To improve on the baseline, a stronger ranking measure is borrowed from [Tanaka & Baldwin 2003b]. The context based translation system approach is adopted in the present work for building the noun compound translation system (NCT) which is integrated to Moses. The outputs of Moses and Moses integrated with NCT are compared. Evaluation of the system is carried out at two levels: by automatic evaluation metric BLEU and by manual evaluation technique. The issue of automatic evaluation is discussed in detail which motivates manual evaluation under the given circumstance.

Abstract

Byzantine Agreement (BA) and reliable Point-to-point communication are two of the most important and well-studied problems in distributed computing. Informally, the challenge of BA is to maintain a coherent view of the world among non-faulty players interacting over a network with few players (possibly with malicious intent) trying to disrupt the entire process. In a synchronous network over n nodes of which up to any t are corrupted by a Byzantine adversary, BA is possible only if all pair point-to-point reliable communication is possible [Dol82, DDWY93]. Specically, in the standard unauthenticated model, (2t + 1)- connectivity is necessary whereas in the authenticated setting (t+1)-connectivity is required. Thus, a folklore is that maintaining global consistency is at least as hard as the problem of all pair point-to-point communication. Equivalently, it is widely believed that protocols for BA over incomplete graphs exist only if it is possible to simulate an overlay-ed complete graph. Surprisingly, we show that the folklore is far from true | achieving global consistency can be strictly easier than all-pair point-to-point communication. In the authenticated model, it is assumed that the adversary can forge the signatures of only those nodes under its control. In contrast, the unauthenticated model assumes that the adversary can forge the signatures of all the nodes (that is, secure signatures are not used). We initiate a study on the entire gamut of BA's in between, viz., the adversary can forge the signatures of up to any k nodes apart from the up to t nodes that it can actively corrupt. We completely characterize the possibility of BA across the spectrum. Thus, our work attempts to unify the extant literature on agreement. It is, however, more than a mere attempt towards unication as it provides insights into the eld. Specically, apart from the extremes (of k = 0 and k = n

Abstract

Autonomous driving is a long-standing problem in robotics, interest in which has spiked in recent years due to the series of DARPA Grand Challenges. One of the basic components in the operation of such autonomous vehicles is the capability to autonomously navigate in terrains particularly in ex- ploratory applications, the robots must deal with terrains whose models at best are only partially-known. On-road navigation is a crucial component of autonomous navigation outdoors and presents interesting challenges, the prominent among those being the ability to discern the traversability of a terrain patch. Modified cars became the new platform in robotics for outdoor navigation which has the possibility to travel longer distances, carry more sensors and thus being more suitable for mapping larger areas. This offers the opportunity to address robotic tasks on a larger scale but on the other hand requires efficient solutions to common problems like mapping or localization. The objective of the work is to develop. simulate and test a terrain modelling system based on laser range data for slow moving vehicles. The terrain modelling system will enable the unmanned ground vehicle to identify the next best way-point to reach. More formally, given a laser range finder fitted on a vehicle moving largely on flat terrain and a pose measurement system that is accurate enough up to a bounded error, develop a terrain modelling system based on laser range data for the vehicle. Robot must possess that enable it to perceive the outside world. We have used three sensors namely GPS, IMU and LIDAR for estimating pose of the vehicle, its orientation and the range measurements respectively. All these devices are mounted on top of a vehicle to acquire data from the sensors. After processing and applying transformations over the data, we will get the map representation in the form of point cloud. Two algorithms for terrain classification for a single tilted laser are developed and extensively tested. One algorithm is plane extraction which will represent the local terrain of the vehicle in terms of planes and points to discern the traversable region. Efficient detection and plane parametrization from scan-line data obtained incrementally to discern the navigable and non-navigable regions confirms the efficacy of the proposed method. However, plane detection is less convincing when dealing with certain structures that arise in natural environments like poles etc. Another algorithm is Spatio-temporal classification which is not too constrained by the nature of the structure of the environment. The essential assumption is that corresponding laser readings in two successive scans are correlated and in general see neighbouring areas. An obstacle or non-navigable region is detected when the sequence indicates decreasing laser distances with corresponding increase in height.This method needs extra rules like searching neighbourhood, relative thresholding to classify obstacles that are not in the frontal viewing cone of the laser. The map obtained is too sparse as we have used a single LIDAR. So, the techniques we have used were not able to overcome few sensor noises. Also, AHRS was giving angle errors while the vehicle pitches and rolls over bumps. This work has lot of future scope by deploying state of the art sensors and multiple laser scanners.