Ph.D. - Electrical Engineering
Permanent URI for this collection
Browse
Recent Submissions
Item Optimal Control of Distributed Energy Resources in Baseline-Based Demand Response Programs(2024) Ellman, Douglas; Xiao, Yuanzhang; Electrical EngineeringItem Microwave and Millimeter-Wave Ubiquitous Physiological Sensing Using Low-Cost Doppler Radar and Communication Systems(University of Hawaii at Manoa, 2023) Ishmael, Khaldoon Mardy; Borić-Lubecke, Olga; Electrical EngineeringMicrowave and millimeter-wave Doppler radar can remotely detect physiological parameters, such as respiration and heart signals. Recent work on physiological Doppler radar further sought to extend continuous radar monitoring beyond controlled settings and into unconstrained environments common to applications such as security, human-machine interface, at-home medical tests, intelligent buildings, and search and rescue operations. Challenges in unconstrained environments include separating signals from multiple individuals and ubiquitous coverage indoors and outdoors. This thesis addresses those challenges as follows: (1) phase correlation approach is proposed and demonstrated effective for signal separation from multiple, close spaces sources using a single antenna, single channel 2.4 GHz radar (2) physiological sensing using channel state information (CSI) for 28 GHz OFDM communication system is modeled and successfully demonstrated experimentally, and (3) remote life sensing, compact and fully integrated 24 GHz dual radar system is implemented to enable noise cancellation for moving platform like UAV.Item Investigation of Demand Response Potential in Smart Grids(University of Hawaii at Manoa, 2022) Najafi, Foad; Fripp, Matthias; Electrical Engineering21st-century power systems are being exposed to many new challenges and opportunities at the same time. Climate change and the increasing cost of fossil fuels force us to replace our primary source of energy with renewables. At the same time, new communication technologies such as 5G paves the way for two-way communication between the generation side and the demand side. The two-way communication between two sides provides numerous advantages for the modern power systems to tackle the environmental and economic challenges with new and creative innovations.One of the new ideas that are conceptualized after the emergence of two-way communication between generation and consumption is the demand response (DR). The DR can help us to overcome these challenges in parallel with other technologies such as energy storage systems. The DR program can help improve any objective that we pursue in the power systems. They can help to reduce the operation cost of power systems by redistributing their consumption plans to the times that less energy consumption is expected. They can help to incorporate more renewable energy sources into power systems. Since renewables are not dispatchable, the current power systems structure has difficulty incorporating a larger share of renewables into power systems due to their unpredictable nature. However, demand-responsive loads can alleviate this problem by adapting their consumption plans based on the generation side with a large share of renewables. The DR program can help reduce the capital cost of power system expansion. Each year, governments are forced to build new infrastructure so only to accommodate the annual peak increase that just happens for a couple of days each year. However, with the adaptive demand that could be used to curb the peak load, millions could be saved by reducing the need to build new infrastructures that are built to meet the peak demand. The DR program can help to improve power quality and operational costs. The unit commitment problems are armed with demand-responsive loads, much lower operational cost, and better power quality at the same time. While there are various types of demand control, such as direct load control (DLC), the focus of this research is on price responsive demand (PRD). The PRDs are more advantageous to DLC for various reasons. One reason is the possible discomfort that DLC approaches could bring. I.e., the service could cut off for the user when the user needs energy. However, with the PRD, the user simply pays the higher cost for using the service. Another main advantage that PRD control has over the DLC is the possibility of creating a market. With the two-way communication that is available, the loads can participate in the markets. E.g., currently, there is a surplus of solar production at midday. However, the PRDs could be used to increase their consumption in the middle of the day by responding to the low cost of electricity in those times. They can participate in ancillary services such as frequency and voltage provisioning based on the needs of the market (operator here) and reduce the cost for the user. Given these opportunities the PRDs have, this thesis is focused on exploring the new opportunities that this program can bring to reality. In the literature, appliances are categorized roughly into three main categories: non-shiftable such as TVs, shiftable such EVs, and thermostatically controller appliances TCAs) such as Electric Water Heaters (EWH). There have been many studies to evaluate the potential of shiftable appliances such as EVs. However, due to temperature drop of TCAs, and the relationship between working temperature and the ambient temperatures, the potentials of TCAs as a reliable PRD is not investigated. Among the TCAs, the EWH have the most potentials to be successfully used in DR programming. They have relatively good insulation and can keep the energy for hours. In contrast, HVAC appliances (heating, ventilation, and air conditioning) cannot maintain their temperature for a long time and need to be fed by energy in short periods of time. At the same time, EWH are a large consumer of energy (some sources report up to 40 percent of daily energy usage). These characteristics makes EWHs a good choice for DR programming. The first two section of this thesis is focused on the potential of EWH as a PRD. First capture investigates the effect unpredictability of energy usage for scheduling the EWHs. alongside that, a new approach to model the concept of comfort is introduced. Accurate Scheduling of appliances needs accurate estimation of environmental inputs. However, a 100 accurate prediction is not feasible. To tackle the uncertainty in prediction of variables, researchers proposed to main approaches to deal with uncertainly. One of them is robust optimization and the other stochastic optimization. In the first chapter of the thesis, a stochastic optimization model is proposed to deal with the randomness of user hot water withdrawal. Alongside that, a dual objective model where one of the objectives is the newly conceptualized discomfort cost. Together, the dual objective comfort-based approach combined with stochastic approach, improves the decision-making procedure by a large margin. The propose model proves that is more accurate at prediction user consumption and at the same time, it incurs less discomfort to the user. Chapter two investigates the potential of EWH as solar energy self-absorbing unit. As mentioned before, solar energy production is concentrated mostly on the middle of the day and no production on other times. However, until now, no effective mechanism is used to absorb this excess energy to the grid. An alternative method is the use of batteries as a storage. However, batteries are expensive as of today. An alternative method is increasing solar energy absorption by inducing demand in appliances. One of the best appliances for this matter is the EWHs. due to their large working temperature range, they can behave as a virtual energy storage unit. I.e., they can absorb energy when solar energy production is high and then gives back this energy to the user when they needed it. In chapter two, a new price-responsive EWH model with solar-absorption capacity is introduced. This model which is based on a modified version of the model that has been proposed in chapter one, can optimally schedule EWH. it can take advantage of using free solar energy and increase the PV self-absorption capacity. This price-responsive approach reduces the operational cost by reducing the dependency on the energy from the grid. At the same time, since it can absorb a large chunk for solar energy, it can reduce the need for a large capacity energy storage unit. Thirst chapter investigates the possibility of market bidding mechanism with the presence of PRDs. a market for energy where the energy generators and users bid for the amount of energy that are willing to buy for a given price would be an ideal mechanism than can bring the energy section close to a free market where ha numerous advantages. However, this ideal model did not happen due to numerous reason such as lack of communication infrastructure, inelastic loads and many other reasons. Aside from these physical obstacles, the lack of an effective method to coordinate the PRD is another challenge. One of the main challenges for coordination of PRDs is creating a balance or equilibrium. In chapter three, a new market-based coordination algorithm for PRDs based on Dantzig-Wolfe decomposition is proposed. The aforementioned method can find an optimal solution (Nash Equilibrium) where no appliance is willing to change their consumption plan. At the same time, as this algorithm progresses, the generation cost for the generation cost and correspondingly, the user payment is decreased. This procedure is continuing until to a point where no PRD is willing to change their behavior since they reached to their global minimum of their payments.Item Security Investigation Of Drone Control Algorithms(University of Hawaii at Manoa, 2021) Chen, Wenxin; Dong, Yingfei; Electrical EngineeringWhile more and more autonomous vehicles and devices are deployed in our society, the security of these systems has raised serious concerns. Although many efforts have focused on their performance and reliability, more systematic research on their security becomes urgent and critical. Therefore, we explore this direction and select consumer drones as our subjects because we can access open-source drone systems (i.e., ArduPilot systems) such that we are able to conduct in-depth investigations of their control algorithms in both theory and practice. As consumer drones have been abused in many incidents, protecting critical assets from consumer drone invasions has become increasingly challenging. While existing methods can interrupt an invading drone, none of them is able to accurately guide it to a desired location for safe handling. By exploiting the weaknesses identified in common state estimation methods and navigation algorithms of drones, and utilizing existing sensor attacking tools, in this research, we develop generic methods to compromise drone state estimation and position control in order to make malicious drones deviate from their targets. In general, an autonomous drone can be attacked at three levels: its onboard sensors, its state estimation, and its navigation algorithms. Our first focus is to accurately manipulate a drone’s state estimation by utilizing existing sensor attack tools. We propose several False Data Injection (FDI) attacks to quantitatively control the EKF-based estimation of 2-dimensional horizontal position, altitude, and magnetic states, and conduct comprehensive analyses on the proposed attacks. Our simulation results show the effectiveness of such attacks. We also propose countermeasures to deal with such attacks. Furthermore, we focus on the navigation algorithms and develop the Drone Position Manipulation (DPM) attack based on the ability of precisely attacking drone sensors and state estimation. DPM is able to accurately manipulate a drone’s physical position and help us guide an invading drone away from its target to a redirected destination. In addition, we formally analyze the feasible range of redirected destinations for a given target. The proposed attacks are validated on the popular ArduPilot flight control system to show its effectiveness. This unique method of exploiting the entire stack of sensing, state estimation, and navigation control together enables the quantitative manipulation of flight paths, different from all existing methods. We also discuss countermeasures to deal with such attacks and illustrate potential solutions. Because the weaknesses of common control algorithms investigated here are popular in many autonomous systems, the proposed attacks may also pose serious threats to the security of these systems. Utilizing different resources available on these autonomous systems, we are further investigating unique vulnerabilities and countermeasures in these environments while ensuring system performance.Item Online Learning and Prediction with Eventual Almost Sure Guarantees(University of Hawaii at Manoa, 2021) Wu, Changlong; Santhanam, Narayana Prasad; Electrical EngineeringIn this dissertation, we introduce a general prediction paradigm where a learner predicts properties of an underlying model and of future samples from the model using past observations from the model. The prediction game continues for infinite steps in an online fashion as the sample size grows with new observations (not necessarily i.i.d.). After each prediction, the predictor incurs a binary (0-1) loss. The probability model underlying a sample is otherwise unknown except that it belongs to a known class of models. The goal of a learner is to make only finitely many errors (i.e. loss 1) with probability 1 under the generating model, no matter what the underlying model may be in the known model class. Any model class and loss pair that admit predictors that make finitely many errors in the above fashion is called eventually almost surely (or e.a.s. for abbreviation) predictable. Our main contributions of this dissertation are general characterizations for the e.a.s.-predictable class and loss pairs. Using our general characterizations, we establish tight necessary and sufficient conditions for a wide range of prediction problems to be e.a.s.-predictable, which include hypothesis testing, online learning and risk management theory. Moreover, our results establish striking connections between the e.a.s.-predictability and the notion of regularization. While e.a.s.-predictable classes admit predictors with only finitely many errors, where we made the final error may yet remain unknown. In particular, we say a class and loss pair to be e.a.s.-learnable if it is e.a.s.-predictable and, in addition, we are able to identify the last error with any given confidence using a universal stopping rule. We provide general characterizations for the e.a.s.-learnability, which is tight in many natural settings. While the above results bring out broad principles, to bring about a more refined development of the framework, we study three broad categories of applications in our framework: hypothesis testing, online classification and learning, and risk prediction. Our characterization of hypothesis testing problems includes testing general properties of distributions using i.i.d. samples, including testing entropy properties of discrete distributions and testing properties of random matrices with Bernoulli entries. Our general results in the e.a.s.-prediction framework also strengthen and extend prior results in Dembo and Peres (1994) with simple and elementary proofs and provide a partial resolution to an open problem posed therein. In our approach to online classification, a classifier obtains the training data in an online fashion and predicts labels on the next instance, but is required to only make finitely many errors over an infinite horizon. Extending the classical bounded error scenario by Littlestone (1988), we show that a binary labeled hypothesis class can be learned online with finitely many errors almost surely using i.i.d. samples from a given distribution µ iff the class is effectively countable w.r.t. that distribution µ. We also characterize the setting where µ is unknown and show that corrupting the labels by independent Bernoulli(η) noise does not change learnability so long as η < 1/2. We extend our results to the case where class labels need not be binary. Going past prior results on learning recursive functions in Zeugmann and Zilles (2008), we show that the class of all binary valued computable functions on naturals can be online learned with finitely many errors almost surely by a computable predictor given samples from certain non-degenerated distributions. Next, we bound the computational complexity of the predictor, and study classes of functions that can be computed in exponential and polynomial time respectively. Lastly, we study the problem of predicting upper bounds on the next draw of an unknown probability distribution after observing a sample generated by it. We show that a prediction rule exists that violates the bounds only finitely many often almost surely if and only if a class can be decomposed into countable union of tight classes. This implies, e.g., the class of all monotone distributions can not be predicted in such a sense.Item Non-contact And Secure Radar-based Continuous Identity Authentication In Multiple-subject Environments(University of Hawaii at Manoa, 2020) Islam, Shekh Md Mahmudul; Lubecke, Victor M.; Electrical EngineeringAn unobtrusive, secure and non-contact continuous authentication system can potentially improve security throughout a login session. Traditional user authentication procedures such as fingerprint, password, or facial identification typically provide only an initial spot-check of identity at the start of a user session, potentially allowing undesired access to personal information (e. g. social security number) at some later point in an apparently continuous user session. The research goal of this PhD dissertation is to create a non-contact and secure continuous authentication system based on Doppler radar, which analyzes back-scattered RF signals which carry body motion information indicating a human subject’s vital signs (breathing rate, heart rate) and associated unique patterns. An additional advantage to this radar technique is that continuous authentication is achieved without intrusive video imaging. Reported prior results are focused solely on use of respiratory motion to identify a single isolated subject. Simultaneous measurement of multiple subjects is a critical challenge. In realistic environments (airport security, in-home sleep apnea test, etc.) the presence of multiple subjects in front of the radar system is likely. To make this technology effective for real world applications, isolation of one particular subject’s breathing pattern from the combined mixture of motion for multiple subjects is essential. Reported research has so far been limited to maintaining 1-m subject separation based on the radar antenna beam-width. This thesis proposes a hybrid method consisting of an SNR-based intelligent decision algorithm which integrates two different approaches to isolate respiratory signatures of two-subjects within the radar antenna beamwidth separated by less than 1-m. A 24-GHz phase comparison Monopulse radar module (K-MC4) has been used to estimate the Direction of Arrival (DOA) for the physiological motion signals of well-spaced subjects at the edge of the beamwidth of the transceiver. DOA is inherently limited to the main beamwidth of the transceiver so when the subjects get closer, crossing the edge of the beamwidth, an additional independent component analysis with the JADE algorithm (ICA-JADE) process is employed to isolate individual respiratory signatures. Experimental results demonstrated that, this proposed SNR-based decision algorithm works with an accuracy of above 93%. In addition, angular location of each subject is estimated by phase-comparison monopulse and an integrated beam switching capability is also demonstrated to optimally extract respiratory information. Additionally, we also conducted a medium scale experiment with twenty participants and collected Doppler radar signals containing the combined respiration mixtures of every pair of participants, over the course of about one month. We then used our proposed SNR-based decision algorithm to separate respiratory signatures from the combined mixtures. From the separated respiratory signatures, we extracted highly distinguishable breathing dynamics-related hyper-features from the respiratory signatures including breathing rate, heart rate, inhale/exhale rate and inhale/exhale area for identity verification. We evaluated the hyper-feature sets with two different classifiers, k-nearest neighbor (KNN), and support vector machine (SVM), and achieved an accuracy 97.5%. We also analyzed the empirical entropy of the hyper-feature set and found that intrinsic entropy of the hyper-feature set is approximately 3-bits which is insufficient for secure identity verification. To improve the security of the proposed system, we also combined fuzzy extractors with linear coding to transform the breathing dynamic related feature into strong biometric keys compatible with machine learning classifiers. We also integrated this proposed radio-based identity verification system with in-home sleep apnea test scenarios. A compliance tracking switching protocol has been developed to integrate the radio-based identity verification system with OSA test. To the best of our knowledge this is the first attempt to achieve secure radio-based multi-subject identity verification by combining the Doppler radar and Fuzzy extractor.Item Reconfigurable and Flexible Liquid-Metal Devices and Circuits(University of Hawaii at Manoa, 2020) Elassy, Kareem Salah Bayomi; Ohta, Aaron T.; Shiroma, Wayne A.; Electrical EngineeringLiquid-metal reconfigurable and flexible electronics attempt to address design constraints on materials, space, cost, and power. The work presented in this dissertation utilizes the unique properties of gallium-based liquid metals towards reconfigurable electronics and flexible electronics. The first part of this work discusses a novel approach to achieve versatile reconfigurable electronics using liquid-metal elements. Typically, reconfigurable electronics are realized using various kinds of switching elements such as MEMS switches, PIN diodes, varactors, and filter banks; these have limited versatility and scalability. These limitations are addressed by using LM in microfluidic channels to shape the structures of RF devices, such as antennas and transmission lines. Reconfigurable 1D, 2D, and 3D arrays of electrically conductive elements were implemented to demonstrate reconfigurable dipole and patch antennas that can change: radiation patterns, creating 87 patterns by altering the shape of a single antenna; polarization angle by exchanging the radiating edge and half-wavelength edge in a patch antenna; and maximum gain from 0 to 2 dBi at different angles of a dipole-antenna radiation pattern. Such antennas are useful in mitigating channel interference in a communication system with multiple transmitters and receivers. The second part of this dissertation discusses liquid-metal patterning using low-cost high-resolution techniques for flexible electronics. Usually flexible electronics are fabricated using nanoparticle or polymeric conductive inks. Conductive nano inks suffer from cracks after cyclic deformation, while polymeric inks have low conductivity. Gallium-based liquid-metal alloy is a promising alternative that addresses these issues. Low-cost fabrication techniques were developed to pattern printed and sprayed LM on polymeric substrates obtaining high-resolution 35-µm features with 25-µm spacing. The usefulness of these fabricating methods was demonstrated by realizing graphene transistors, RFID, patch antennas, and transmission lines. Many more possibilities for liquid-metal applications are still unrealized.Item Timing based source separation(University of Hawaii at Manoa, 2019) Young, Jeremy Brian Wai Mun; Høst-Madsen, Anders; Electrical EngineeringThe motivation of this work is to use statistical signal processing to help to separate mixtures of marine mammal vocalizations, in particular sperm whale click trains. It is observed that clicks from a single whale are spaced at regular intervals which we exploit to do the separation. To begin, we consider an idealized problem: each source emits at impulses at iid intervals according to some distribution. That is, the impulse times for each source constitute a renewal process. These timing distributions induce a likelihood function for the impulse times given some assignment. We present an algorithm inspired by the Viterbi algorithm to give assignments that maximizes this likelihood. Additionally, we provide more computationally feasible approximations of this algorithm. We verify these algorithms with a derived lower bound. Furthermore, we provide methods to estimate the timing distribution parameters and the total number of sources using alternating maximization and the minimum description length. We also attempted to use timing to help improve detection methods for impulses using a Bayesian framework. Unfortunately, the results were impractical, but the concepts were used to generate a classification using impulse shape information timing information. Finally we explore modifications to make the algorithm work on actual sperm whale click mixtures.Item Covariance Selection Quality and Approximation Algorithms(University of Hawaii at Manoa, 2018-12) Tafaghodi Khajavi, Navid; Kuh, Anthony; Electrical EngineeringThis dissertation conducts a study of graphical models, discusses the quality of statistical model selection approximation, and proposes algorithms for model approximation. Graphical models are useful tools for describing the geometric structure of networks in applications that deal with high dimensional data. Learning from these high dimensional data requires large computation power which is not always available due to hardware limitation for different applications. Thus, we need to compromise between model complexity and its accuracy by using the best possible approximation algorithm that chooses a simpler, yet informative model. The first problem we study in this work is the quality of statistical model selection. We consider the problem of quantifying the quality of approximation model. The statistical model selection often uses a distance measure such as the Kullback-Leibler (KL) divergence between the original distribution and the model distribution to quantify the quality of approximated model distribution. Although the KL divergence is minimized to obtain model approximation in many cases, there are other measures and divergences that can be used to do so. We extend the body of research by formulating the model approximation as a parametric detection problem between the original distribution and the model distribution. The proposed detection framework results in the computation of symmetric closeness measures such as receiver operating characteristic (ROC) and area under the curve (AUC). In particular, we focus on statistical model selection for Gaussian distributions, i.e. the covariance selection [1]. In the case of covariance selection, closeness measures such as KL divergence, reverse KL divergence, ROC, and AUC depend on the eigenvalues of the correlation approximation matrix (CAM). We find expressions for the KL divergence, the log-likelihood ratio, and the AUC as a function of the CAM. We present a simple integral to compute the AUC. In addition, easily computable upper and lower bounds are also found for the AUC to assess the quality of an approximated model. Through some examples and simulation for real and synthetic data, we investigate the quality of the covariance selection for both tree-structured models and non-tree structured models. The second problem we target in this work is to formulate a general framework and algorithms to perform covariance selection. We develop a multistage framework for graphical model approximation using a cascade of models such as trees. In particular, we look at the model approximation problem for Gaussian distributions as linear transformations of tree models. This is a new way to decompose the covariance matrix. Here, we propose an algorithm which incorporates the Cholesky factorization method to compute the decomposition matrix and thus can approximate a simple graphical model using a cascade of the Cholesky factorization of the tree approximation transformations. The Cholesky decomposition enables us to achieve a tree structure factor graph at each cascade stage of the algorithm which facilitates the use of the message passing algorithm since the approximated graph has fewer loops compared to the original graph. The overall graph is a cascade of factor graphs with each factor graph being a tree. This is a different perspective on the approximation model, and algorithms such as Gaussian belief propagation can be used on this overall graph. Here, we present theoretical results that guarantee the convergence of the proposed model approximation using the cascade of tree decompositions. In the simulations, we look at synthetic and real data and measure the performance of the proposed framework by comparing the KL divergences.Item Game Theoretic Approaches To Communication Over MIMO Interference Channels In The Presence Of A Malicious Jammer(University of Hawaii at Manoa, 2018-12) McKell, Kenneth Clayton; Arslan, Gurdal; Electrical EngineeringThis dissertation considers a system consisting of self-interested users of a common multiple-input multiple-output (MIMO) channel and a jammer wishing to reduce the total capacity of the channel. In this setting, two games are constructed that model different system-level objectives. In the first—called “utility games”—the users maximize the mutual information between their transmitter and their receiver subject to a power constraint. In the other (termed “cost games”), the users minimize power subject to an information rate floor. A duality is established between the equilibrium strategies in these two games, and it is shown that Nash equilibria always exist in utility games. Via an exact penalty approach, a modified version of the cost game also possesses an equilibrium. Additionally, multiple equilibria may exist in utility games, but with mild assumptions on users’ own channels and the jammer-user channels, systems with no user-user interference, there can be at most one Nash equilibrium where a user transmits on all of its subchannels. A similar but weaker result is also found for channels with limited amounts of user-user interference. Two distributed update processes are proposed: gradient-play and best-response. The performance of these algorithms are compared via software simulation. Finally, previous results on network-level improvement via stream control are shown to carry over when a jammer is introduced.