Indian Institute of Science Bangalore
etd@IISc Electronic Theses and Dissertations at Indian Institute of ScienceNot a member yet
6204 research outputs found
Sort by
Hardware-based Device Identification for Systems with Commercially Off-the-shelf Components
No full textThe Identity of an electronic device is a fundamental property, that bootstraps several applications such as authentication and traceability. For the purpose of device identification, conventional methods generate a unique number using techniques such as the chip’s wafer-ID and the XY location, or through a random number generator. More recently, Physically unclonable functions (PUFs) are emerging as an alternative to the conventional methods. PUFs exploit the inherent variations in the device characteristics occurring due to tolerances in the manufacturing processes.
Our work has focussed on developing PUF-based identification methodologies for systems with Commercially-off-the-shelf (COTS) components. The inherent tolerances in the parameters available in these components have been exploited and mapped to perform their identification. The benefits of our mechanism are: (a) No custom PUF circuits have been used, (b) No requirement for manual hardware reconfiguration and power-cycling, (c) identification has
been performed in real-time using simple software Application Programming Interfaces (APIs).
In our first work, we have constructed an identifier that we call IoT-ID. This identifier is based on the variations in clock oscillators and Analog to digital converters (ADCs) that are commonly present in SoCs. We have demonstrated that IoT-ID is repeatable and unique. We have also shown the scalability of our approach through numerical simulations.
In our second work, we have developed an Acoustic PUF that combines the Uniqueness signature of a device with its Position signature. The Uniqueness signature has exploited the clock tolerances in the devices, making the signature unclonable. The Position signature has been derived using Acoustic fingerprinting, giving a sticky identity to the device. Our evaluation has demonstrated the Uniqueness and Repeatability and further shows the use of temperature coefficients for device identification.
In our third work, we have constructed a digital identifier by exploiting the inter-channel variance in errors for a multi-channel simultaneous sampling sigma-delta ADC. Such a device is common in power instrumentation such as Intelligent Electronic Devices (IEDs), and thus our methodology can be used to determine their identity. The suggested approach for identifier generation is resilient by construction, and is thus minimally impacted due to external factors such as voltage and temperature variations. We have also evaluated the randomness of the identifier to explore its suitability as a random key.
General purpose input outputs (GPIOs) are the most common interfaces present in almost all microcontrollers, including low-end systems. By performing identification based on GPIOs in our fourth work, we have demonstrated the generic nature of our approach and its adaptability to a wide variety of microcontrollers. Since a large number of GPIOs are used for the construction of GPIO PUF, in this work, we have evaluated the redundancy among different components and presented a step-by-step method to identify the significant contributors.
Different devices may map to the same identifier, causing a ‘collision.’ Our final work presents a framework that computes the collision probabilities based on inter-device and intradevice variations. In particular, we have derived the probability of none of the devices in collision and upper bound on the probability of there being L distinguishable devices. We have also computed the expectation of number of collision-free devices. The framework can be utilized to tune PUF attributes and compare various PUF implementations.
Our research has established the feasibility of PUF-based device identification for systems with COTS components, paving the way for its wider adoption in deployments
Characterization of Interconnections in Smart-X Applications
No full textSmart-X applications are realized by interconnecting several objects to achieve real-time capability by improving safety, reliability, and efficiency. These interconnections can be broadly classified as physical connections and logical network connections. A malfunction or damage in a physical connection may be catastrophic, leading to system downtime and sometimes fatal accidents. Thus, monitoring these connections is crucial to identify the onset of damage or a malfunction. Logical network connections are used in several applications to track and monitor a mesh-connected group of things called the Mesh of Things (MoT). In our research work, we focus on the characterization of physical and logical network connections. We apply these characterizations to: (a) reliably detect insulation damage in signaling and indoor power cables, (b) reliably detect and report a detangled MoT network, and (c) develop a suitable testbed to demonstrate measures and algorithms for anomaly reasoning in Smart-X connectivity. \\
In our first use case, we implement Power Line Communication (PLC) based measurement methods and propose algorithms to detect, classify and localize cable insulation faults. The solution considers factors such as the type and length of the cable, the width, and length of the fault, the structure of the network wiring, and source and load variations on the cable. In order to increase the accuracy of cable fault detection and localization, we use a conglomerate diagnostic solution. We perform extensive measurements using an in-situ non-invasive Software Defined Radio (SDR) based composite diagnostic kit. We develop a Bayesian inversion framework to estimate the Health Index (HI) of the cable. The HI assists in determining the cable state in real time and thus can suggest the need for preventive maintenance. Secondly, we develop an anomaly reasoning framework for the cable insulation fault by utilizing our experimental data collected from the testbed. Since source and load variations can contribute to an anomaly, we derive insights using measures such as SNR, S parameter, and Reflectogram. We identify the most probable root cause for an anomaly in a cable section. In particular, we are interested in estimating the belief for the severity of a cable insulation fault.\\
In our second use case, we are interested in the real-time detection of a detangled MoT. Our solution uses Bluetooth Low Energy (BLE) based mesh and PLC backbone network. Measures such as latency and Packet Delivery Ratio (PDR) are characterized for this heterogeneous network which is subjected to interference and impulse noise. Our use case is related to air cargo monitoring and tracking within an airport terminal. Finally, we analyze the performance of the integrated BLE-PLC network through numerical simulations. We use the generalized Noisy-OR model for mesh reliability and Dijkstra’s shortest path model for latency analysis. For the backbone network, we use a measurement-aided channel frequency response model for latency and reliability analysis. We validate the simulation results with our empirical setup
Influence of Soil's Electrical Parameters on Lightning Stroke-current Evolution and Fields in the Close Range
No full textThe lightning return stroke forms one of the severest natural sources of electromagnetic interference for ground-based and airborne systems. Many physical fields are involved in this complex physical phenomenon. Several pertinent aspects are somewhat unclear, and it is not practical to conduct the field measurements to resolve them. One such important aspect, which is of practical relevance, is the influence of soil's electrical properties on the stroke current evolution and the fields in the soil. It formed the genesis of the present work.
The collection of the required data from on-field measurements would be nearly impossible, and hence suitable theoretical approach was considered. For that, an appropriate model for the return stroke is necessary. Among different models for the lightning return stroke, only the 'Self-consistent return stroke' model is found to be suitable. This model employs a macroscopic electrical representation of the underlying physical phenomenon and accounts for the associated dynamic electric field to emulate the stroke current evolution. However, in the past works, only perfectly conducting earth was considered, and it relied on the time-domain thin-wire formulation to evaluate the associated dynamic electromagnetic fields.
On the other hand, a more realistic representation of the soil, with its frequency-dependent and non-linear parameters, is required for the present work. This necessitated a suitable adoption of the domain-based 'Finite difference time domain' (FDTD) method for field computation. It turned out that, in an FDTD framework, the modeling of the channel and its corona sheath, soil-ionization, and soil-dispersion is a challenging exercise.
For the simulation, a straight vertical channel of 5 km is considered. A complex-frequency-based PML (perfectly matched layer) is employed to truncate the problem domain. The high aspect ratio of the channel does not permit the application of standard FDTD update equations with a realistic spatial discretization. The conventional subcell approach, generally used to model thin-wire structure in an FDTD framework, was also not usable for two reasons. Firstly, the channel has a dynamic conductivity, and secondly, the presence of corona-sheath surrounding the channel produces a typical field profile in the region. The channel in the soil and the non-linear ionization around it also posed a similar problem. A ‘Modified subcell approach’ was developed to handle the lightning channel, which is one of the essential contributions of the present work. In the ‘Modified subcell approach’, the spatial field variation is computed at each time step, taking into account all the relevant field contributions in the respective region. The radial current produced by the charge deposited in the corona sheath is also be taken into account separately.
The frequency-dependent conductivity and permittivity of the soil require a convolution in time domain formulation. This would require a repeated calculation of the integral over each cell, a forbidden task. Based on one of the recent literature, a suitable simplification is adopted, thereby drastically minimizing the computational requirement. The soil ionization, a strongly field-dependent phenomenon, required a different set of developments. Each cell is divided into subgrids to account for the local field variation and the dynamic conductivity profile.
The developed FDTD formulation is deployed to investigate the role of soil's electrical properties on the stroke current evolution and the field in the soil using the self-consistent return stroke model. For the first time, it is shown that the soil's electrical conductivity has some noticeable influence on the stroke current magnitude (up to about 45 %), and the ionization phenomenon in soil tends to reduce this influence . It is shown that the current magnitude varies most for a low magnitude fast-rising current as the soil ionization is minimal for these cases. On the other hand, for high-level slow-rising currents, the ionization process significantly matures, and as a result, the dependence of current magnitude on soil resistivity is reduced substantially. It is noted that the effect of soil permittivity and the frequency-dependent soil parameters on the return-stroke current is minimal.
From the results of the detailed simulation, it is found that soil resistivity also affects the field in the soil significantly. The field in the air is increased with decreasing soil resistivity, and the increase is primarily due to the increase in channel current magnitude. For the field in the soil, in addition to modulating the field magnitude, soil resistivity also affects the temporal nature, with the field becoming peakier for lower resistivity. A comparison of the computed field demonstrates that the field is underestimated significantly by the prevalent quasi-static approach, and the difference increases with the radial distance from the channel. The frequency-dependency of the soil's conductivity, and permittivity to a lesser extent, significantly reduces the field in the soil. It is also seen that the current concentration near the surface due to skin-effect is altered at later periods by the field produced by the channel current. The presence of a second layer of lower resistivity at a shallow depth, on the other hand, effectively controls the current and field in the top layer. It is also shown that the field for a strike to a mountain can depend significantly on the mountain height.
In summary, significant contributions have been made in the present work towards the FDTD formulations for modeling lightning phenomena and assessing the soil’s electrical parameters on lightning stroke current evolution and the resulting field
Implicit Gradient Reconstruction for Unstructured Mesh Finite Volume Method
No full textIn the present work, an Implicit Gradient Reconstruction (IGR) method is proposed in the
context of Finite Volume Methodology (FVM). There are three computationally intensive steps
involved in a typical finite volume framework for a spatially second-order accurate upwind
scheme. These are solution reconstruction, solution limiting and flux finding. Solution re-
construction involves determination of gradients while solution limiting requires comparison
of double precision numbers. Further, computation of gradient and solution limiting is cell
based procedure while flux finding is edge/face based procedure. The proposed IGR procedure,
which is edge/face based procedure, integrates all these steps obviating explicit reconstruction
and limiting steps resulting in a considerable reduction in computational effort and associated
memory footprint.
In the modified CIR (MCIR) scheme of linear convection equation, a parameter φ is in-
troduced to control dissipation. The IGR procedure is derived from the MCIR scheme. The
relation between the φ parameter and solution reconstruction in a finite volume procedure is
systematically established. The methodology is extended to multidimensions, where the use of
φ implicitly represents a reconstruction step. Hence, this procedure is referred to as Implicit
Gradient Reconstruction. In addition, it is brought out that the use of φ also serves the purpose
of solution limiting. The spatial accuracy of this procedure is demonstrated by computing the
2-D circular convection problem.
The methodology, when extended to the Euler equations of Gas dynamics, results in the
reconstruction of the characteristic variables. Consequently, three steps in the computation of
explicit residual, namely, solution reconstruction, limiting and flux computation, are seamlessly
merged into a single step. Owing to its significantly smaller memory footprint, the procedure
is particularly relevant to large scale parallel computing. This procedure can be effortlessly
incorporated into any of the existing finite volume solvers where inviscid flux formulation is
based on characteristic decomposition. The capability of IGR procedure is established through
several test cases involving inviscid and viscous flow computations in one and two dimensions.
The results obtained from IGR procedure are compared with reconstruction based solvers and wind tunnel data wherever available. The IGR procedure produces results comparable to the
classical reconstruction based procedure.
The aforementioned IGR procedure is applicable to any flux formulation involving charac-
teristic decomposition. An Edge Based Reconstruction Limiting (EBRL) is proposed for flux
formulation not involving characteristic decomposition. This procedure is very simple and does
not require classical diamond path reconstruction. The results obtained from EBRL based Roe
and AUSM Plus flux formulation on transonic viscous flow over RAE 2822 airfoil are very good
and compare well with wind tunnel experiments
Polymerizable Porogen – Direct generation of internally functionalized porous polymers
No full textBicontinuous functional porous polymers are desirable in terms of pore accessibility, particularly for applications in templating, chromatographic separation, and catalysis. Bicontinuous morphologies can be obtained by block copolymer self-assembly only over a very narrow compositional window, which makes it synthetically challenging. On the other hand, polymerization induced phase separation (PIPS) methodology for generating bicontinuous porous structures, which was discovered several decades ago, is an easy strategy, but it does not permit precise control over the pore size, especially at smaller sizes. To achieve multiple objectives in a single step, Seo and Hillmyer ingeniously combined the concepts of PIPS and BCP self-assembly to generate crosslinked polymer matrices, wherein they introduced a covalent linkage between the pore-forming segment and the matrix, allowing microphase separation to occur at the nanoscale dimensions; thus generating nanoporous polymers. The bicontinuous pores were obtained because of kinetic trapping of the microphase-separated domains via in situ crosslinking.
The main objective of my thesis is to develop an alternate strategy using a polymerizable porogen, wherein a polymerizable Styryl unit is linked to the pore-forming PEG segment via a thermally labile linker, namely a urethane. Copolymerization of the polymerizable porogen (PolyPo) with a crosslinker, divinyl benzene (DVB), leads to the formation of a microphase separated crosslinked matrix; one of the defining ideas of the study is the exploitation of the thermal reversibility of the urethane linkage between the pore-forming segment and the matrix, which not only disconnects the porogen to generate the porous matrix but also leaves behind amine groups (upon reaction with water) that lines walls of the pore. Careful examination of the in situ microphase separation process during the copolymerization of the PolyPo, revealed that slowing down the polymerization by using controlled radical polymerization is essential to prevent premature crosslinking and allow effective microphase separation. The pore volume and surface area in these systems could be easily controlled by varying the ratio of porogen to the crosslinker, whereas the average pore size depended only on the length of the PEG porogenic segment. This concept was extended to the distyryl systems where both ends of the porogen PEG segment were linked to polymerizable groups; it was shown that the pore size was dependent only on the size of the porogenic segment and was unaffected by the presence of two polymerizable units. Further, it was shown that carrying out the copolymerization of the PolyPo in the presence of free PEG porogen could be an effective strategy for controlling the average pore dimension, despite broadening of the size distribution; likewise, copolymerization of two PolyPos carrying different PEG segments proved to be an alternate approach for fine-tuning the average pore size.
Finally, we showed that the pore-generating PEG segment could also be designed as a counter-ion to a suitable polymerizable unit; in this case PEG-trimethylammonium, 4-vinylbenzoate was designed as an ionic PolyPo. Here, a simple MeOH-HCl wash at 70°C was adequate for near-complete removal of the porogen PEG segment. The pore size, pore volume, and surface area of these counter ion-based systems varied much like their covalent counterparts. Most importantly, the porous crosslinked matrix that was generated using this approach carried carboxylic acid groups, unlike the earlier urethane-based strategy that left behind amine groups. In summary, we demonstrated novel single-step strategies for the preparation of crosslinked bicontinuous porous polymers carrying tailorable functional groups on the pore walls, with excellent control over the pore size and surface area.CSIR NE
Studies on Nanostructured Transition Metal Oxides and Related Composites for Supercapacitor Electrodes
No full textSupercapacitors have acquired considerable scientific and technological position in energy storage field owing to their compelling power capability, good energy density, excellent cycling stability and ideal safety. Supercapacitor is the burgeoning candidate to cope with the ever-growing need for green and renewable energy. High-performance supercapacitors are realized by nanostructured electrode designs, which provide ameliorated surface area for abundant electrode-electrolyte interaction, ease of electron transfer and movement, and short ion-diffusion pathways, that lead to increased performance. Transition metal oxide (TMO)-based electroactive materials are of significant interest owing to the remarkable combination of structural, mechanical, electrical, and electrochemical properties. Besides their high specific capacitance and energy density, the stable redox chemistry, highly reversible and fast charge-discharge processes, low cost, and environment-friendly processes make them the most promising materials for next-generation supercapacitors. In the light of the foregoing, in the thesis efforts are made to synthesize transition metal oxides (TMOs) and related composites with varying nanostructures and characterize them as supercapacitor electrodes. The thesis comprises six chapters. Chapter I is an introduction to pristine TMOs with different nanostructured dimensions namely, 3D, 2D, 1D, and 0D, and their composite structures as electrode materials for supercapacitors. Design of different pristine and composite nanostructures, synthesis strategies, comprehensive structure-dependent electrochemical properties, present challenges and future perspectives are reviewed.
Chapter II presents an unpretentious method for anchoring pseudocapacitive materials on multi-walled carbon nanotubes (CNTs) to create high-performance electrode materials for asymmetric supercapacitors (ASCs). Anchoring mechanism involves the direct decomposition of the metal-hexacyanoferrate complex on the CNT surface. The nanoparticles (NPs) are discretely attached to the CNT surface without forming a homogeneous layer, making practically the entire NP surface open for electrochemical reactions. As a result, when compared to a pure CNT electrode, the CNT-Mn3O4 nanocomposite cathode exhibits significantly increased capacitive performance, demonstrating the usefulness of the composite electrode design. As a paired anode, CNT-Fe3O4 nanocomposite was used. At 10 mV s-1 scan rate hybrid ASC achieves a gravimetric capacitance of 135.2 F g-1 in 1 mol/L aq. Na2SO4 electrolyte within 0-1.8V potential window and gives excellent cycling performance (100%) even after 15000 cycles.
Chapter III discusses charge-storage mechanism of free-standing MoS2/r-GO hybrid nanoflakes on molybdenum (Mo) foil in Na2SO4 solution is elucidated for realizing a high-performance asymmetric supercapacitor (ASC). Thiourea that acts primarily as sulfur source also helps intercalating ammonium ions, which along with r-GO facilitate in-situ exfoliation of MoS2, producing hierarchical MoS2 with expanded interlayer spacing. This interlayer expansion in MoS2 facilitates Na+-ions intercalation/de-intercalation, and ensures enhanced capacitance, rate capability and cycling stability of the capacitor. Besides exhibiting attractive energy-cum-power traits, the 2V MoS2/r-GO//Fe2O3/MnO2 ASC shows compelling cycling performance for over 20,000 cycles in an aqueous electrolyte.
Chapter IV presents the study on a one-step synthesis of carbon encapsulated Fe/Fe3C nanoparticles by pyrolysis of a single source precursor of Prussian Blue (Iron (III) ferrocyanide) and used as anode material in high-performance supercapacitors. The synthetic approach creates porous structures in the shape of a 3D doughnut, with many interconnected Fe/Fe3C nanoparticles completely enclosed within layers of graphitic carbon. During charge storage on Fe/Fe3C nanoparticles via surface or near-surface-based faradaic reactions, such a porous structure enables electrolytic ion diffusion, while the metallic iron helps in increasing the composite electronic conductivity. As a result, at lower scan rates, capacitive as well as diffusion-controlled mechanisms dictate charge storage in carbon-encapsulated Fe/Fe3C nanoparticles, while capacitive processes take over at higher scan rates. At 10 mV s-1 scan rate, nanocomposite material could deliver 223 F g-1 gravimetric capacitance and gives good cycling performance for over 20000 cycles with very little loss in capacity.
Chapter V briefly describes a one-step synthesis of sheet-like RuS2 nanostructures exhibiting traits of a potential cathode material for designing high-performance asymmetric supercapacitors (ASCs). The synthesis includes direct sulfurization of RuO2 in an inert atmosphere at high temperature that results in densely packed nanosheets of RuS2 with moderate surface area. Such a structure provides abundant sites for surface or near-surface based faradaic/non-faradaic reactions for energy storage while facilitating ion migration during charge/discharge processes. Furthered from these traits, RuS2 electrode exhibits substantially enhanced electrochemical performance as compared to the RuO2 electrode. Detailed analyses suggest that the charge storage in such RuS2 nanosheets is governed by capacitive as well as diffusion-controlled processes at lower scan rates but is dominated by capacitive processes at higher scan rates.
The thesis culminates with an application study on a 36 V substrate-integrated lead-carbon hybrid ultracapacitor with and without charge-balance circuit developed and performance tested in gel electrolyte presented in Chapter VI. A 36 V hybrid ultracapacitor is realized by connecting three 12 V hybrid ultracapacitors in series. The three 12 V hybrid ultracapacitors in the 36 V hybrid ultracapacitor are found to have uneven performance; to circumvent this problem, a voltage-management cell-balancing circuitry is employed for realizing synchronized performance from each of the 12 V hybrid ultracapacitor unit in the series arrangement.DS
Optimizing the Interval-centric Distributed Computing Model for Temporal Graph Algorithms
No full textGraphs with temporal characteristics are increasingly becoming prominent. Their vertices, edges and attributes are annotated with a lifespan, allowing one to add or remove vertices and edges. Such graphs can grow to millions of vertices, billions of edges, and have months or years of data. Time-dependent algorithms such as temporal reachability and shortest paths are designed over such materialised graphs. These algorithms find important use-cases in digital contact tracing, optimising transit routes, and analysing information diffusion over temporal graphs.
Interval-centric Computing Model (ICM) is a recent abstraction over temporal graphs, enabling intuitive development of temporal graph algorithms while ensuring efficient computation and communication. It uses a bulk-synchronous parallel model of execution with data-parallel computation on interval-vertices and message passing at superstep boundaries. To ease the design of temporal algorithms, ICM introduces a novel TimeWarp phase for temporally aligning messages and grouping them against vertex states. However, this warp operator is super-linear in time complexity with the number of messages received at a vertex. It also has additional overheads in the form of message replications. Further, in pipelining the computation and communication phases, ICM may create stale or redundant messages. This thesis primarily attempts to design techniques to mitigate these performance limitations of ICM, and also extends ICM toward incremental graph processing.
We propose three different techniques to accelerate the execution model of ICM: Local Warp Unrolling (LU), Deferred Message Scatter (DS) and Windowed ICM (WICM). LU unrolls the messages processed in the TimeWarp phase to reduce the time complexity of the warp operator. DS results in lazy scatter operations that reduce redundant calls to messaging. WICM partitions the temporal graph along the temporal dimension and processes the sub-interval graphs in parts, ensuring proper carryover of vertex states. While LU and DS apply locally to each vertex, WICM is applicable at the global interval graph level and can be coupled with the other two techniques.
While developing these techniques, we identify necessary constraints identifying the algorithms that can be modelled using the optimisations. Further, we also prove the equivalence of the new execution model to ICM's execution for a large class of temporal traversal algorithms. For WICM, not all temporal partitioning strategies give the same execution performance. Hence, we also develop heuristics that use statistics on the global graph topology with an analytical modelling of TimeWarp to determine the interval partitioning used with WICM.
We extensively evaluate these optimisations for six large temporal graphs with up to 133M vertices, 5.5B edges and 365 snapshots, and six graph algorithms on an 10-node commodity cluster. LU+DS reduce the runtime of ICM by an average of 56%; WICM reduces the runtime by 48% on average over native ICM, and combining these techniques offers an average reduction of 61%. We also conduct experiments to confirm the effectiveness of the heuristic partitioning technique. We also present preliminary results on extending the WICM model to operate over a graph that arrives incrementally, by batching the incoming updates and forming a window out of them to be executed using the WICM. This also has the benefit of reducing the memory footprint since the entire historic graph does not need to be retained in memory
Perceptual Quality Assessment of Lowlight Restored and Authentically Distorted Images
No full textThe capability of hand-held devices to acquire high-definition visual content has led to a tremendous increase in the number of images and videos captured daily. However, camera hardware and pipelines are not perfect and lead to multiple distortions in the captured content. This makes quality assessment (QA) imperative to advance the qualitative capability of different devices and the pipelines used. More particularly, the aim of perceptual quality assessment is to quantitatively analyze the perceptual quality of the captured content with respect to the distortions observed by the human visual system. This thesis focuses on two aspects of perceptual quality assessment. Firstly, we focus on the subjective and objective quality assessment of low-light restored images. Then we consider the problem of unsupervised quality assessment methods for authentically distorted images.
The quality assessment of restored low-light images is an important tool for benchmarking and improving low-light restoration (LLR) algorithms. While several LLR algorithms exist, the subjective perception of the restored images has been much less studied. Challenges in capturing aligned low-light and well-lit image pairs and collecting a large number of human opinion scores of quality for training warrant the design of unsupervised (or opinion unaware) no-reference (NR) QA methods. In this part, we study the subjective perception of low-light restored images and their unsupervised NR QA. Our contributions are two-fold. We first create a dataset of restored low-light images using various LLR methods, conduct a subjective QA study, and benchmark the performance of existing QA methods. The lack of good perceptual quality metrics designed explicitly for the low-light scenario is an important limitation in advancing the design of restoration methods. To tackle this, we present a self-supervised contrastive learning technique to extract distortion-aware features from the restored low-light images. We show that these features can be effectively used to build an opinion unaware image quality analyzer. Detailed experiments reveal that our unsupervised NR QA model achieves state-of-the-art performance among all such quality measures for low-light restored images.
The quality assessment of camera captured authentically distorted images is challenging due to the lack of a reference. While there is a plethora of supervised no reference image QA algorithms, there is a need to study unsupervised or opinion unaware algorithms based on their superior generalization performance. We explore self-supervised learning (SSL) for the feature design on authentically distorted images to predict quality without training on human labels. While SSL on synthetic distortions has recently shown promise, there is a need to enrich the feature learning on authentic distortions. We propose a novel two-stage learning approach on synthetic and authentically distorted images with different learning methodologies. We perform contrastive learning with positives and negatives that vary with quality on synthetic data to capture quality features. While learning on authentically distorted images, we only consider positives due to the difficulty in obtaining negatives that vary in quality alone. We employ the SimSiam framework to enrich features by fine-tuning on authentically distorted images. We show that the self-supervised features we learn can be used to make perceptually consistent image quality predictions on authentically distorted images without training on any human opinion scores. We achieve state-of-the-art performance on multiple authentically distorted datasets without training on them
2-Level Page Tables (2-LPT): A Building Block for Efficient Address Translation in Virtualized Environments
No full textEfficient address translation mechanisms are gaining more and more attention as the virtual
address range of the processors keeps expanding and the demand for machine virtualization
increases with cloud and data center-based services. Traditional radix-tree based address translations
can incur significant overheads in big data applications, particularly under virtualization,
due to multi-level tree walks and nested translation. The overheads stem primarily from the
unnecessary generality — ability to support several hundreds of thousands of virtual memory
regions in the virtual address space — supported by current processors.
We observe that in the common case, however, a process’s virtual address space contains
only a few contiguously allocated sections, which can be efficiently translated using a shallow
tree with two levels. We propose such a compact structure, called 2-Level Page Table(2-LPT),
which serves as a key building block for address translation in virtualized environment. A key
advantage of 2-LPT is that it maintains two levels of page tables irrespective of the size of the
virtual address space. Translating a virtual address using 2-LPT is fast. A walk on a 2-LPT
requires up to two memory accesses. In practice, however, the root level table is well cached in
the Page Walk Caches, thus, single memory access is sufficient. Under native execution, 2-LPT
reduces the page walk latency by up to 20.9% (9.38% on average) and improves performance by
up to 10.1% (1.66% on average) over the conventional four-level radix tree page tables, on a set
of memory-intensive applications.
2-LPT is more beneficial under virtualization. A naive extension of 2-LPT reduces the cost
of nested page walk from 24 to 8 memory accesses. To achieve further reduction, we propose
two optimizations: (i) Enhanced Partial Shadow Paging (ePSP) which employs a limited form
of shadow paging for the root-level of 2-LPT, and (ii) Host PTE Mirroring (HPM) which
allows accessing the host page table entry without performing host page table walk. These
optimizations reduce the number of memory accesses to just one, on average, while avoiding slow
VM exits. 2-LPT speeds up applications by 5.6%-50.9% (24.6%, on average) over the baseline
nested page walks. Compared to the best performing state-of-the-art proposal, 2-LPT achieves
an average reduction of 41.5% and 15.7% in page walk latency and execution cycle
Explorations in the Space of S-Matrices
No full textS-matrix is one of the fundamental observables of the quantum theory of relativistic particles. The quantum dynamics of relativistic particles can be abstractly understood in terms of S-matrix bypassing a Lagrangian formulation of quantum field theory. Equivalently, the space of possible S-matrices defines an abstract theory space. In this talk, I will discuss how to constrain the spectrum of physical theories in the theory space using the basic physical requirements of Poincare invariance, quantum unitarity, and causality.
The thesis discusses two distinct but related ways of such exploration. The first part of the thesis explores a novel mathematical way of cruising the space of S-matrices using the techniques from geometric function theory (GFT). The discussion will be centred on a crossing symmetric dispersive representation of scattering amplitudes due to Auberson and Khuri (1972), which enables us to carry out this exploration. In particular, the dispersion kernel turns out to be a univalent function in a suitable complex variable. Univalent functions are known to satisfy various bounding relations. The most famous of them is the de Branges’ theorem, previously known as the Bieberbach conjecture, which bounds the Taylor coefficient of univalent functions. Using this theorem, we put double-sided bounds on Wilson coefficients of EFT amplitudes. Using another theorem, the Koebe growth theorem, we were able to put double-sided bounds on the amplitude itself. We also explore the connection with another kind of function from GFT, the typically real function, which is also known to satisfy various bounding relations. Using these GFT techniques, we study elastic scattering amplitudes of identical massive scalar Bosons, EFT amplitudes of elastic 2-2 photon and graviton scattering and try to constrain the space of low energy effective field theories.
In the second part of the thesis we turn our attention to holographic S-matrices. The conjectural holography provides a way to construct flat space scattering amplitudes from the Mellin amplitudes of a conformal field theory (CFT) by taking a large radius limit of the dual space. Various analytic properties of flat space scattering amplitudes are encoded in corresponding properties of the CFT Mellin amplitude. Flat space scattering amplitudes are known to satisfy high energy bounds called the Froissart-Martin bound which follows from axiomatic analyticity and unitarity properties of the S-matrix. Froissart-Martin bound is one of the robust consistency tests for a flat space scattering amplitude. Therefore if a holographic construction of the S-matrix is to work, one should be able to obtain a systematic derivation of the Froissart-Martin bound starting with point Mellin amplitude for a holographic CFT. We provide such a derivation in the second part of the thesis. We find that our holographic derivation gives the exact Froissart-Martin bound in spacetime dimensions, while in greater spacetime dimensions, we get weaker bounds. We attempt to argue the possible reason for this behaviour