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
Posturing for Autonomous Digital Human Model
No full textDigital Human Models (DHMs) are representations of humans in virtual environments that have applications in diverse disciplines. In the context of engineering design, DHMs are used to assess if a proposed design satisfies ergonomics requirements. This is done by inserting a DHM into the CAD environment and making it interact with the virtual prototype of the system being designed. The main challenge in this process is the difficulty in controlling the DHM for simulating the interaction. The DHM systems that are currently in use have low autonomy, meaning controlling the DHM requires the designer to spend significant amounts of time and effort, as well as possess considerable expertise in the domain.
This thesis endeavours to enhance the autonomy of DHM systems, so that task simulations can be performed quickly and easily, even by users with less or no expertise. While a full-fledged autonomous DHM system would have numerous components, the current work focuses on the aspect of automating the process of generating postures. This thesis presents a computational framework that takes high-level commands as input and automatically generates physically valid postures and motions of human performing manual tasks. The first part of the thesis focuses on synthesising static postures and the later part builds on the developments of the first part for generating body-motions.
Firstly, an optimisation-based computational framework is developed to simulate functional reach postures that account for factors of stability and biomechanical effort. With this framework, simulation of reach postures and generation of reach-envelopes for extreme reach-tasks in various standing postures are demonstrated.
The capability to simulate functional reach is then extended to simulation of postures that take supports from the environment. Here, along with the computation of postures, the optimal location for support-contact and reaction forces at the support-contact are also computed through optimisation. In addition to that, a novel support-taking behaviour is introduced, which enables the DHM to automatically choose supporting surfaces that are suitable for the task to be performed.
Based on the facility to generate static postures, the problem of simulating posture-transition tasks is then addressed. Posture-transition deals with generation of body-motion connecting the given initial and final postures of the human, placed anywhere in the environment. Here, an approach for automatically generating and composing primitive motions that executes the posture-transition is developed. The posture-transition facility is demonstrated with the simulations of tasks such as walking, stair-climbing and vehicle ingress/egress.
After this, tasks involving manipulation of objects are characterised and methods for generating manipulation motions are developed. Finally, the methods to simulate manipulation and posture-transition tasks are combined to develop the capability to simulate long and complex operations in a fully autonomous mode. This unique feature is demonstrated with a simulation of a manual-assembly process.
The theories and methods developed in this thesis were implemented in the form of a software application called Maya-Manav. The details of the computer implementation of Maya Manav tool are presented at the end.
In summary, this thesis presents a generic and practical posture-generation framework that can be seen as a low-level infrastructure or a platform on which highly autonomous DHM simulation systems can be built
Investigation of twisted bilayer graphene using electrical and thermoelectric transport
No full textThe ability to tune the twist angle between di erent layers of two-dimensional materials has opened up a new
dimension to band engineering in van der Waals (vdW) heterostructures. By taking advantage of the formation
of a moir e superlattice arising from a small lattice mismatch or twist angle between two adjacent atomic layers,
one can create materials with tailored electronic, optical, mechanical, thermal and optoelectronic properties. In
particular, twisted bilayer graphene (tBLG) has attracted considerable scienti c interest owing to exceptional band
tunability. The coupling between the two graphene layers depends strongly on the twist angle, leading to angledependent
electronic and phononic hybridization. In addition, when the relative rotation is close to the magic angle
( m = 1:1 ), the low-energy electronic bands are nearly
at, leading to a multitude of interaction-driven phases. This
includes correlated insulators, superconductivity, magnetism, non-trivial band topology, nematicity and signatures
of non-Fermi liquid (NFL) excitations with linear-in-temperature resistivity that persists down to temperatures well
below the Bloch{Gr uneisen temperature. Although signi cant progress has been made in understanding the vast
phase diagram of tBLG, there is no consensus on the origin of the superconductivity, metallic states and the role
of electron-phonon coupling at very low temperatures ( 1 K). In this thesis, we study the in-plane and cross-plane
electrical and thermoelectric properties of tBLG with varying twist angles to understand the nature of metallic states
and the role of layer breathing phonon modes at low twist angles.
The cross-plane thermoelectric transport in large angle tBLG is driven by the scattering of electrons and inter-layer
breathing phonon modes. However, the relevance of layer hybridized phonons in thermoelectric transport remains
unclear when the electronic hybridization of the two layers becomes strong at low . In the rst part of the thesis, we
show out-of-plane thermoelectric measurements across the vdW gap in tBLG, which exhibits an interplay of twistdependent
interlayer electronic and phononic hybridization. We show that at large twist angles, the thermopower is
entirely driven by a novel phonon-drag e ect at the subnanometer scale. In contrast, the electronic component of the
thermopower is recovered only when the misorientation between the layers is reduced to < 6 . Our experiment shows
that cross-plane thermoelectricity at low angles is exceptionally sensitive to the nature of band dispersion.
Although the T-linear resistivity in tBLG at low temperatures has been attributed to the absence of a well-de ned
quasiparticle spectrum, experimentally, the manifestation of NFL e ects in transport properties of twisted bilayer
graphene remains ambiguous. In the next part of the thesis, we have performed simultaneous measurements of
electrical resistivity ( ) and thermoelectric power (S) in tBLG for several twist angles between 1:0 1:7 . We
observe an emergent violation of the semiclassical Mott relation (MR) in the form of excess S close to half- lling for
1:6 that vanishes for & 2 . In addition, for a device with 1:24 , excess S is observed at fractional band
lling. The combination of non-trivial electrical transport and violation of Mott relation provides strong evidence of
NFL physics intrinsic to tBLG.
Next, we study the electrical and thermoelectric transport in marginally tBLG ( 0:5 ), where the electronic band
structure at a low twist angle is expected to become qualitatively di erent as compared to magic-angle because of large
moir e unit cell and strain-accompanied lattice reconstruction. We observe a strong metallic behaviour accompanied
by a T-linear and an emergent violation of the semi-classical Mott relation in the vicinity of van Hove singularities
(vHSs). Our experiments show that the thermopower is exceptionally sensitive to the band dispersion in small-angle
tBLG even at high temperatures, and the low-T transport is governed by a network of topological channels formed
at domain boundaries between AB and BA regions.
Finally, we have demonstrated the working of a thermoelectric generator consisting of dual-junction tBLG. We
show that the thermopower in cross-plane tBLG can be enhanced by using a dual-junction device. Further, using
an external resistor enables us to measure the current-voltage characteristics of the device and estimate the power
generated in the system
Efficient Resource Allocation for Underlay Device-to-device Communication Networks with Limited Channel State Information
No full textDevice-to-device (D2D) communication is finding applications in future wireless networks such as vehicular networks and internet-of-things. It offloads traffic from the base station (BS), improves energy efficiency, and reduces latency by enabling direct communication between the users. In underlay D2D, the D2D users share subchannels with the cellular users (CUs). While this improves spatial reuse, it causes interference between the D2D users and CUs. Hence, interference-aware resource allocation is an important research problem for underlay D2D networks.
In this thesis, we consider a practically feasible partial CSI model in which the BS only knows the channel state information (CSI) of the CU-to-BS and D2D Receiver (DRx)-to-BS links. The D2D pair knows the CSI of the D2D transmitter (DTx)-to-DRx and CU-to-DRx links, and the statistics of inter-D2D and inter-cell interference powers. We propose a feedback model in which the DRx computes the signal-to-interference-plus-noise ratio (SINR) estimate and feeds a quantized version of it back to the BS. The SINR estimate is such that the corresponding rate has an outage probability within a pre-specified value.
We first consider a subchannel allocation problem in which at most K D2D pairs are allowed to share a subchannel and a minimum rate with a pre-specified probability of outage is guaranteed for the CUs. We propose a polynomial-time algorithm called cardinality-constrained subchannel assignment algorithm (CCSAA) based on a submodular maximization approach. We prove that it gives a D2D sum rate that is at least one-third of the optimal D2D sum rate. We also propose a lower-complexity locally greedy algorithm (LGA) that provides the same theoretical guarantee but is applicable when K is equal to the number of D2D pairs. We then propose a modification of LGA called cardinality-constrained LGA (CCLGA) that applies to all values of K. We propose a rate upgradation scheme employed at the D2D pair to improve the D2D rate after subchannel allocation by exploiting the asymmetry in the rate information at the BS and the D2D pairs.
Next, we consider a statistical CSI model in which the DRx computes and feeds back the SINR estimate by knowing only the statistics of the CSI of DTx-to-DRx, CU-to-DRx links, and the inter-cell and inter-D2D interference powers. We propose a relaxation-pruning algorithm (RPA) based on a linear program relaxation and rounding approach. It provides a D2D sum rate that is at least half of the optimal D2D sum rate. We present numerical results to investigate the interplay between the CSI model and resource allocation algorithm design by considering partial and statistical CSI models, and RPA and CCSAA. RPA outperforms CCSAA for the partial CSI model, while CCSAA outperforms RPA for the statistical CSI model even though it has a lower theoretical sum rate guarantee than RPA. We connect this to the different sensitivities of the algorithms to the variation of rates across subchannels for the considered CSI models. We find that the optimal value of K depends on the CSI model, algorithm, and feedback resolution. We also propose a statistical rate upgradation scheme in which the D2D pair exploits the broadcast subchannel allocation information to upgrade its rate.
In the last part of our work, we study the subchannel allocation problem with a disjunctivity constraint between the D2D pairs. It prevents two D2D pairs from sharing a subchannel if they cause significant interference to each other. This approach is naturally applicable for dense D2D networks, where D2D pairs are closely spaced and avoids the conservative rate estimates generated by our earlier approaches. We address the subchannel allocation problem for two cases. In the first case, a D2D pair is allowed to transmit on multiple subchannels. We propose a branch-and-bound algorithm to assign subchannels to the D2D pairs. In the second case, a D2D pair is allowed to transmit on only one subchannel. We propose a submodular maximization-based approach in which, for each subchannel, we apply the branch-and-bound algorithm to assign subchannels to the D2D pairs. This approach provides at least half of the optimal D2D sum rate. We look at the disjunctivity constraint and the SINR computation based on path-loss and fading-averaged interference power between the D2D pairs. Our results show that considering fading-averaged interference power that includes the path-loss and shadowing leads to improved system performance than considering only path-loss, which is often considered in the literature
Epithelial-to-mesenchymal transition and cellular cooperation in cancer progression: Novel roles for AMPK
No full textEpithelial-mesenchymal transition (EMT) is a developmental program hijacked by cancer cells to facilitate metastasis —a multi-step process involving the spread of cancer cells from the primary tumor site and culminating in the formation of secondary tumors at distant anatomical sites. Majority of cancer-associated deaths are due to metastasis; however, currently, there is a dearth of effective treatment once the tumors have metastasized. A better comprehension of the molecular players that govern the process of EMT will aid in the development of effective therapeutic strategies. AMP-activated protein kinase (AMPK) —an evolutionarily conserved energy-sensing kinase— is known to be activated by pathophysiological cues that induce EMT, such as hypoxia and TGFβ. Thus, the role of AMPK in the regulation of EMT was investigated. Activation of AMPK induced EMT in multiple solid tumor cell lines, as observed by enhanced expression of mesenchymal markers, decrease in epithelial markers, and an increase in migration and invasion. In contrast, inhibition or depletion of AMPK resulted in the reversal of EMT. Notably, AMPK activity was necessary for the induction of EMT by pathophysiological cues such as hypoxia and TGFβ treatment. Mechanistically, AMPK mediated EMT activation via multi-factorial regulation of the EMT-transcription factor Twist1. The present study identifies AMPK as a critical regulator of the EMT program, thus suggesting that strategies targeting AMPK might provide novel approaches to curb the spread of cancer.
Cancer is increasingly being viewed as an “ecosystem” that enables cellular interactions between cancer cells and neighboring cancer-associated cells. Furthermore, intra-tumoral heterogeneity arising from the presence of functionally distinct cancer cells is being recognized as a major player in cancer progression and therapy failure. Reversible EMT changes, serving as a source of intra-tumoral epithelial-mesenchymal (EM) heterogeneity, could generate heterogeneous cancer cells with functional differences in migration, invasion, stemness, immune evasion, and drug response. In this study, we investigated the effect of cellular interactions between epithelial and mesenchymal cancer cells within a tumor population in the outcome of chemotherapeutic drug treatment. Epithelial (E) and mesenchymal (M) subpopulations were segregated from the parental A549 cell line that exhibits inherent EM heterogeneity. Contrary to the prevalent notion in the field, mesenchymal (M) cells segregated from within the parental heterogeneous population displayed enhanced susceptibility to DNA-damaging chemotherapies, such as doxorubicin and mitoxantrone, compared to epithelial (E) cells. More importantly, E cell-derived exosomes transmitted chemoresistance to the sensitive M cells. Blockade of exosome production impairs the survival advantage conferred during the coculture of E and M cells, revealing ‘intercellular’ cooperation between epithelial and mesenchymal cancer cells in overcoming chemotherapeutic challenges. Mechanistically, exosome proteomics identified several proteins as potential candidates responsible for conferring chemoresistance. Furthermore, a novel role for AMPK in regulating exosome biogenesis was identified. AMPK inhibition reduced the number of exosomes released by the donor E cells and altered the exosomal protein cargo, suggesting that AMPK-targeted therapeutics might sabotage the cooperative survival advantage established by heterogeneous epithelial and mesenchymal cancer cells
Algorithms and Testbed for Synchronous Generator Parameter Estimation
No full textThe development of dynamic power system component models became increasingly important in the modern grids dominated by high penetration of renewables because of the increased dependency of planning and operational decisions on dynamic simulation studies. The parameters of synchronous machines and associated control models play significant role in the overall model of the grid, which need to be updated regularly by the utilities. So, the parameters of the power plants are calibrated/estimated either using off-line testing or online measurements from phasor measurement units (PMU) or digital fault recorders (DFR). Development of individual generator models is feasible only if the PMU/DFR data is available for each generator in a power plant. Otherwise, they can provide only aggregate model of a generating plant as PMU/DFRs are usually placed in substations. Digital protective relay (DPR) records are available for individual generators in any generating plant.
This thesis explores the possibilities of utilizing DPR records of individual generators for parameter estimation. About 36 relay records have been collected from a 247 MVA, 15.75 kV generator of a thermal plant in Karnataka. It is found that most of the records contain at the most 3 seconds data. The relay records should contain prefault data, during fault data and some post-fault data for accurate estimation. However, from the collected records only a small percentage of the records are found to be useful. Existing methods of parameter estimation using PMU/DFR data failed to work with the short duration records. There is no prior work reported in the literature which uses short relay records for parameter estimation of the synchronous generators. Constrained iterated unscented Kalman filter (CIUKF) and enhanced scattered search (eSS) algorithms are proposed for the parameter estimation using DPR records in this thesis. Parameters of the turbo alternator and its excitation system are estimated from the relay records collected using the proposed algorithms and the results are found be accurate.
For the holistic validation of the developed algorithms and faster adaptation by GENCOs, realistic testbeds are needed. A scaled-down generalized substation model for translational research in smart grids is developed, which can be configured to operate in 7 widely used substation bus bar schemes with prevalent current transformer (CT) configurations. All the potential transformers (PT) and CT measurements, circuit breaker (CB), isolator and earth switch status signals are made available to configure any protection strategy like bus-bar protection, unit protection schemes, etc. precisely the same way they get implemented in the field.
For studying the control interactions between renewable and conventional sources, frequency dependent (FD) transmission line models need to be physically realized. A new algorithm is proposed to fit a reduced-order R-L equivalent circuit to the frequency response of the modal impedances of a transmission lines. A close enough fitting is achieved with lesser number of passive elements using the proposed method compared to the widely used vector fitting algorithm. A scaled-down model of WECC 3-machine 9-bus system is developed with frequency dependent lines by selecting suitable tower and conductor configurations. Reduced order lumped parameter FD (LPFD) line models are derived for the 230 kV transmission lines in WECC system using the proposed fitting algorithm. A systematic procedure to scale down the 230 kV LPFD line models to 220 V laboratory model is presented. An experimental prototype of the scaled-down LPFD line is developed. Clarke and inverse Clarke transformations are implemented using specially designed 1-φ transformers. The inductances of the scaled-down model are realized using amorphous cores. Based on the prototype testing results, the six lines of WECC system are fabricated considering manufacturing tolerances.
Parameter estimation using practical DPR records, development of substation model including detailed station configurations and CT arrangements, and physical realization of a frequency dependent power transmission line model in the laboratory are first of its kind efforts in the literature to the best of our knowledge.MHRD, Govt. of India for the financial support through scholarship and DST, Govt. of India for supporting through the Fund for Improvement of Science and Technology (FIST) program and Robert Bosch Center for Cyber-Physical Systems (RBCCPS), IISc for the financial research gran
Atomization characteristics of alternative aviation biofuels, Jet A-1, and water from a hybrid airblast atomizer
No full textStudies on the atomization of sustainable aviation fuels (SAF) from aircraft engine atomizers are essential to replace the present fossil type jet fuel to counter the rise in aviation-caused CO2 emissions in the atmosphere. The characteristics of spray droplets resulting from the breakup of liquid film in atomizers are crucial for the description of primary atomization process and combustion dynamics in aircraft engines. The thesis investigates the atomization of camelina- and jatropha-derived drop-in aviation biofuels from a hybrid airblast atomizer (HAA) used in aircraft jet engines. The main focus of the study is on the evaluation of spray droplet characteristics in the near-region of liquid film breakup. The experiments are carried out in a spray test facility. The images of sprays at different flow conditions are captured using backlighted shadowgraphy technique. The measurements of spray droplet characteristics are obtained using phase Doppler interferometry (PDI) and Spraytec at different spatial locations of the spray below the atomizer exit.
In the first part of the study, extensive experiments of liquid atomization from the HAA using water are conducted. The size and velocity characteristics of droplets resulted from the liquid film breakup in the simplex swirl atomizer (central atomizer in the HAA), measured within millimetre distance from the actual location of the liquid film breakup, are analysed. The mean axial velocity of the spray droplets measured at the film breakup point is independent of droplet size, which is different from the correlation characteristics of the spray droplets observed in other regions of the spray. The linear film breakup theory overpredicts Sauter mean diameter (SMD) of the spray measured at the breakup point significantly, and an existing scaling law for the determination of volume median diameter of the spray captures the present experimental trend of the droplet size recorded at the breakup point. The droplet size distribution measured at the breakup point is well described by a Gamma distribution with index parameter n governing the corrugation features of ligaments formed in the film breakup.
Further, by using the self-similarity analysis of droplet size and velocity, the demarcation region between the near- and far-region of liquid film breakup in the spray is established. A systematic comparison of spray characteristics in the near- and far-region of the liquid film breakup is reported.
In the second part of the study, the atomization characteristics of water and drop-in aviation biofuel sprays from the HAA is carried out. The droplet characteristics in the near-region of liquid film breakup are obtained at a distance 19 mm from the atomizer exit using Spraytec. The present droplet size data compare well with the predictions obtained using previously reported empirical correlation with a modified proportionality constant. The spray characteristics of the aviation biofuel sprays from the HAA are almost same as that of the standard fuel (Jet A-1) spray, which confirms the drop-in behavior of the chosen alternative fuels. By using the present experimental data of HAA spray from six experimental fluids, an empirical correlation for the estimation of nondimensionalized SMD in terms of liquid and gas Weber numbers and Ohnesorge number is propose
Scanning Probe microscopy of van der Waals heterostructures and non-equilibrium magnetotransport in graphene
No full textGraphene is a two-dimensional semimetal that has linear dispersion in energy-momentum
space. When graphene is subjected to a perpendicular magnetic field, the dispersion is no
longer linear, resulting in discrete energy levels because of the formation of cyclotron orbits
of different energies. This energy discretization leads to quantum oscillations in longitudinal
magnetoresistance known as Shubhnikov de-Haas oscillations which provide a plethora of
properties, including the effective mass of charge carriers and topological properties like Berry
phase. Furthermore, the transverse resistance in the magnetic field is quantized, making it
useful for resistance metrology. The quantization effects have been realized in graphene in
the ohmic regime, i.e., with a small current density < 0.01 A/m passing through the channel.
Non-equilibrium magnetotransport studies in two-dimensional electron gas systems based on
GaAs-AlGaAs quantum wells have been intensively investigated under high current densities,
demonstrating the effect of carrier heating, magnetophonon-oscillations, and Hall field-induced
magneto-oscillations in longitudinal resistance. However, the effect of high current densities on
magnetotransport in graphene has not been thoroughly investigated. In this thesis, we have explored the magnetotransport in graphene Hall bar devices under non-equilibrium conditions by
introducing a high current density (> 1 A/m) through the channel, which produces a strong Hall
field across the channel and results in tilting of the Landau levels. For the experiments aimed at
realizing electron transitions between two cyclotron orbits in the presence of a magnetic field,
the width of the channel becomes crucial. We have fabricated large-width Hall bar devices,
which ensures the number of cyclotron orbits in the bulk is significant, and edge scattering will
have less contribution, making it more sensitive to magnetotransport in the bulk of the channel.
Making extra-large width devices becomes a significant step that requires a sizeable clean
xii
area of graphene to ensure high mobility. The dry pick-up and transfer method to fabricate
hexagonal boron nitride (hBN)- encapsulated graphene is a standard technique to achieve
high-quality devices. Sandwiching graphene between hBN often leads to folding, wrinkling,
and the formation of air pockets between hBN and graphene, which limit sample quality. There fore, it becomes essential to identify the geometrical extent of clean graphene. Here we have
developed a non-invasive sub-surface electrical scanning probe technique to identify a clean and
significant area of graphene encapsulated by 20-30 nm thick hBN. We have used Electrostatic
Force Microscopy (EFM) to identify the region of interest. This method reveals the effect of
substrate and ambient environment on the doping of graphene. We have conducted elaborate
measurements on various encapsulated layered materials and observed that the EFM phase acts
as a clear fingerprint of the constituent layered materials in complex heterostructures involving
graphene, hBN, and transition metal dichalcogenides. In addition to providing visually striking
images of buried layers, the technique is also useful in probing the electrical properties of the
constituent layers. We have extended the technique to other van der Waals heterostructures
of transition metal dichalcogenides such as MoS2 and WSe2 encapsulated in hBN. We expect
our findings to advance reliable and high throughput device architectures for various nano and
optoelectronics applications.
To explore the non-equilibrium transport properties in graphene we have exploited the EFM
technique to identify the homogeneous and residue-free region of graphene encapsulated by
hBN. Hall bars with device widths ranging between 12 µm to 18 µm were made to investigate
the non-equilibrium magnetotransport. In addition to expected carrier heating effects, we
observe two branches of novel magnetoresistance oscillations near the charge neutrality point
when plotted as a function of carrier density and dc current at magnetic field ranging between
1 T to 5 T. These oscillations show linear dispersion as a function of dc current and carrier
density. The drift velocity of carriers associated with dispersion matches well with the TA, and
LA phonon modes in graphene, indicating phonon-assisted intra-Landau level transitions aid in
these oscillations. The novelty of these results are expected to stimulate further studies that
can help unravel a unified picture of the various resonant processes in this regime, not only in
graphene but also in related Moiré heterostructure
Novel Neural Architectures based on Recurrent Connections and Symmetric Filters for Visual Processing
No full textArtificial Neural Networks (ANN) have been very successful due to their ability to extract meaningful information without any need for pre-processing raw data. First artificial neural networks were created in essence to understand how the human brain works. The expectations were that we would get a deeper understanding of the brain functions and human cognition, which we cannot explain just by biological experiments or intuitions. The field of ANN has grown so much now that the ANNs are not only limited for the purpose which they emerged for but are also being exploited for their unmatched pattern-matching and learning capabilities in addressing many complex problems, the problems which are difficult or impossible to solve by standard computational and statistical methods. The research has gone from ANN being used only for understanding brain functions to creating new types of ANN based on the neuronal pathways present in the brain. This thesis proposes two novel neural network layers based on studies on the human brain. First is a type of Recurrent Convolutional Neural Network layer called a Long-Short-Term-Convolutional-Neural-Network (LST_CNN) and the other is a Symmetric Convolutional Neural Network layer based on Symmetric Filters.
The current feedforward neural network models have been successful in visual processing. Due to this, the lateral and feedback processing has been under-explored. Existing visual processing networks (Convolutional Neural Networks) lack the recurrent neuronal dynamics which are present in ventral visual pathways of human and non-human primate brains. Ventral visual pathways contain similar densities of feedforward and feedback connections. Furthermore, the current convolutional models are limited in learning spatial information, but we should also focus on learning temporal visual information, considering that the world is dynamic in nature and not static. Thus motivating us to incorporate recurrence in the convolutional neural networks. The layer we propose (LST_CNN) is not just limited to spatial learning but is also capable of exploiting temporal knowledge from the data due to the implicit presence of recurrence in the structure. The capability of LST_CNN’s spatiotemporal learning is examined by testing it on Object Detection and Tracking. Due to the fact that LST_CNN is based on LSTM, we explicitly evaluate its spatial learning capabilities through experiments.
The visual cortex in the human brain has evolved to detect patterns and hence has specialized in detecting the pervasive symmetry in Nature. When filter weights from deep SOTA networks are visualized, several of them are symmetric similar to the features they represent. Hence inspiring the idea of constraining standard convolutional filter weights to symmetric weights. Given that the computational requirements for DNN training have doubled every few months, researchers have been trying to come up with NN architectural changes to combat this. In light of that, deploying symmetric filters reduces not only computational resources but also memory footprint. Therefore, using symmetric filters is beneficial for inference and also during training. Despite the reduction in trainable parameters, the accuracy is comparable to the standard version, thus allowing us to infer that they prevent over-fitting. We establish the quintessence of symmetric filters in NN models
Fast Methods for Modelling and Simulation of Fully Integrated Voltage Regulators in Microprocessors
No full textFully Integrated Voltage Regulators (FIVR) have been introduced in the latest generation of high-performance server microprocessors to improve the performance and power efficiency of the processors. FIVR is a switched inductor DC-to-DC step-down converter with on-chip power bridges, control circuits and on-chip Metal-Insulator-Metal (MIM) capacitors. The inductors for FIVR are designed on the package using package traces and Plated Through Hole vias (PTH). Each FIVR module generates the voltage needed for the functional units such as the CPU cores locally, and a typical microprocessor package has more than 100 such FIVR modules. The switching frequency of FIVR is kept high (~100 MHz) to minimize the inductor size. The different components of FIVR are modelled and simulated using circuit simulators to predict the output voltage, input voltage noise, switching ripple, efficiency, etc., Due to the multiscale nature of FIVR, a lot of challenges are faced in the modelling and simulation of FIVRs by the circuit simulation approach. This thesis discusses the various challenges involved in the modelling and simulation of FIVRs and proposes fast methods to solve these challenges.
At the high switching frequency of FIVR distributed effects dominate, and Full Wave Electromagnetic Extraction tools need to be used for extracting the models of the package inductors. Volume-based electromagnetic modelling tools such as the Finite Element Method (FEM) or the Finite Difference Time Domain (FDTD) method need more runtime to model the FIVR inductors as the entire volume needs to be meshed. The latest generation of FIVR inductors uses Magnetic materials for the magnetic cores with frequency-dependent permeability which further increases the runtime in volume-based methods. In this thesis, the fast surface integral equation method also known as the Method of Moments (MoM) is developed for modelling Perfect Electric Conductor (PEC) and PEC-dielectric/magnetic objects.
The multiple FIVR modules in the chip share a common input supply (Vccin) for cost reduction. However, the sharing of the input supply also introduces the problem of noise coupling between the FIVRs. The load current transients at the output of one FIVR can couple to the output of other FIVRs through the input network. This noise is referred to as the Vccin feedthrough noise. The modelling of noise coupling between multi-domain FIVR is a challenge, and one normally runs into long run times or convergence issues in circuit simulation. In this thesis, two fast methods are developed to model and simulate the noise coupling in multi-domain FIVRs. The first method is a frequency-domain method and is based on the g-parameter transfer functions of FIVR. The second method is a time-domain method and is based on the state-space models of FIVR inductors and MIM capacitors determined from the Vector-Fitting technique. The proposed methods are demonstrated to improve the runtime and simplify the modelling of multi-domain FIVRs.
The modelling of the input ripple noise due to the switching of the FIVR power trains is a challenge for multi-domain FIVR due to the need to include switching models of many FIVRs in circuit simulation. The input power supply (Vccin) is also a large, distributed network that prohibits the detailed switching simulation of multi-domain FIVRs. Convergence issues are faced in circuit simulations with such large models and the runtime is high. In this thesis, a Harmonic Domain (HD) method based on Linear Periodic Theory is developed to model the switching ripple voltage in multi-domain FIVRs. The various challenges faced in the circuit simulation approach can be avoided using the Harmonic Domain method.
In FIVR domains with high load current, many FIVRs are ganged in parallel to supply the high load current. The modelling and simulation of ganged FIVR is a challenge due to the large scale of the problem with many FIVR modules ganged together with a large output power plane. The tuning of the FIVR control loop to attain the stability of ganged FIVRs remains a challenge with circuit simulation-based methods. Convergence issues are frequent in circuit simulations, and the runtime is high in circuit simulations. In this thesis, transfer functions are derived for ganged FIVRs from the extracted models of FIVR inductors and on-chip MIM capacitors. The transfer functions are used to model the stability of the ganged FIVR and can be used for the transient analysis also.
The tuning of the FIVR control loop in a single domain FIVR is a challenge as there are a lot of Resistor-Capacitor (RC) combinations of the op-amp compensator circuit in the FIVR feedback control loop. The RC values need to be tuned to meet the control loop specifications such as the Unity Gain Bandwidth (UGB), Phase Margin (PM) and Gain Margin (GM). The practical op-amp compensator is non-ideal, and it is difficult to model its behaviour using analytical tuning methods such as the k-factor method. In this thesis, a machine learning method based on Bayesian Optimization is developed to tune the FIVR control loop and is demonstrated to reduce the number of circuit simulations significantly compared to traditional optimization methods. This method can be easily extended to post-Si measurements where the optimization is done on the fly and the next set of samples to be measured is selected based on the machine learning algorithm.Intel Technology India Pvt Lt
Design of Compact Antennas With Metasurface for Wideband and Wireless Applications
No full textAn antenna is one of the essential elements in a wireless system, that converts the guided waves in an electronic circuit to unguided waves in the air and vice versa. They are often designed according to the specifications of the underlying system. Compact antennas are required in miniaturised systems such as those used in an aircraft. They are designed by modifying or appending the antenna with additional structures or circuit elements without degrading its responses. In this thesis, the design of compact antennas is investigated with metasurface for two unique purposes - i. wideband applications for detection/sensing application, and ii. spatial modulation to communicate a multipath environment.
For wideband applications, a spiral antenna is considered a primary radiator due to its wideband impedance matching and circular polarization (CP) response with simple and planar geometry. It has a bidirectional radiation pattern on either side of the structure, along the axis of the antenna. But in many practical applications, a single-sided radiation pattern is extracted by placing it above a metallic body of a ship or aircraft, which disturbs the freestanding radiation response of the antenna. A conductor placed more than half a wavelength away from the spiral reduces the boresight gain significantly at high frequency, whereas the same placed too close to the antenna degrades the matching and polarization performance at low frequency. These issues have been addressed over years with different techniques, but the design of compact spiral still possesses significant challenges especially when a frequency band of 1-18~GHz is considered.
As this research work begins, the spiral is placed at different heights above a metallic conductor and the effects are observed over the considered frequency range. It is followed by an investigation with profiled metallic geometries to combine the benefits of varying antenna heights at different frequencies. Based on these observations, a compact spiral antenna is designed by placing it above a modified conical conductive backing to radiate a CP wave over a wide frequency band.
In the next part of this thesis, some of the challenges at low frequencies are addressed using different absorber techniques when the spiral is kept extremely close to a conductor. A hybrid technique consisting of absorbing material and resistors is proposed to design such a compact spiral antenna for wideband application. To improve the performance below 2~GHz, a wideband metasurface absorber is investigated with the spiral. The metasurface possesses significant electromagnetic absorption at low frequency and has been used to design a spiral antenna for 1-18~GHz with an extremely low profile.
Another work with a compact spiral antenna approaches to tilt its main beam over a wide frequency range. This investigation is required to compensate for the shift in the antenna main beam due to the supporting structure or to tilt the antenna main beam in a given direction for different purposes. A semicircular lens made of lossy dielectric material is placed above a compact spiral to fulfil this requirement. Effects of different material properties and lens profiles are investigated to arrive at the final design. Since placing the lens along the spiral affects the compactness of the antenna and disturbs the planar profile required in a flush mounting configuration, a sectoral metasurface is designed and printed on the backside of the antenna substrate. The metasurface possesses effective material properties to tilt the antenna main beam at a consistent angle.
For all cases, numerical investigations were carried out to optimize the antenna geometries followed by prototyping and characterization of some of these structures. The measured results are compared with the simulated outcomes and the numerical predictions have been verified. This required the design and realization of a wideband balun and appropriate fixtures to integrate various parts of this antenna in a flush-mount arrangement.
For a unique wireless application with a compact antenna, a digitally reconfigurable metasurface in the vicinity of a patch antenna is proposed, to realize for the first time a modulator for a spatial modulation technique known as media-based modulation (MBM). MBM facilitates a fast, secure, and multiuser wireless link in a multipath environment (e.g., indoor or office environments) by exploiting the multipath components of the channel. The metasurface works as an electromagnetic window as the power flowing through the unit cell can be electronically controlled by switching a PIN diode embedded within. A significant difference in transmission coefficient is observed between the two switching states of the unit cell. A meander geometry is used to make it compact and the diode is placed between the meander and one of the two contiguous strips that provides the necessary biasing to the diode. Numerical investigations are carried out to characterize the unit cell, and to optimize the array dimensions and the gap between metasurface and antenna. A prototype of the array is fabricated with the necessary control circuitry and a complete wireless link is set up to communicate in a real-time environment. Experiments are carried out in different scatter free and scattering environments in line of sight and non-line of sight configurations to validate the theoretical predictions of MBM. The effects of multipath as a factor that improves communication performance are also validated. In the end, data transmission over a wireless link is also demonstrated using this scheme.Ministry of Education, Govt. of India and Thales Defence Mission System