Department of Electrical Engineering
http://hdl.handle.net/10106/4968
2024-03-28T19:17:38ZEarables as Medical Devices: Opportunities and Challenges
http://hdl.handle.net/10106/31885
Earables as Medical Devices: Opportunities and Challenges
Abdul, Aziz; Karkar, Ravi; Ding, Kan; Harvey, Jay; Nguyen, Phuc
The recent advancements in remote healthcare monitoring have revolutionized clinical practices. They provide reliable and affordable
solutions to monitor physiological and neural signals in various
environments instead of having to rely solely on on-site clinical
monitoring in hospitals and medical centers. Wearable devices,particularly earables, computers placed in or around the ear, have
gained considerable attention for their convenience, social acceptability, comfort, and minimal disruption to daily activities.
The development of earable devices has been propelled through
the recent technological advancements in edge and fog computing,
battery technology, and energy harvesting methods [11]. Earables
allow us to capture important signals from the upper part of the
human body, especially the human head. Equipped with the right
sensors (e.g., accelerometers, IMU, PPG), these devices have the potential to monitor various biometrics such as heart and respiratory
rate [12], blood pressure [2] and muscle movements [8]. Electrochemical sensors in earables can even open up novel possibilities
to investigate alternative body fluids such as sweat [13]. Moreover, their proximity to the brain and eyes enables the detection
of complex modalities like EEG [6], and EOG [9] making earables
promising for health monitoring [7, 10]. They are also less affected
by motion noise compared to smartwatches, which are prone to
disruptions from wrist movements [4]. The portable, discreet, and
unobtrusive nature of these devices allows them to monitor patient
vitals for extended periods in any setting and environment [1].
However, there are inherent challenges associated with earable
technologies and the use of the ear as a sensor location. Interference
from motion and physiological artifacts, limitations in electrode
placement area, and the trade-off between signal resolution and
wearability pose obstacles to reliable and accurate monitoring [5].
Factors such as battery failures, computation latency, and RF interference can introduce errors in measurements leading to wrong
treatment. Additionally, as wearable medical technology becomes
339
UbiComp/ISWC ’23 Adjunct, October 08–12, 2023, Cancun, Quintana Roo, Mexico Abdul Aziz, Ravi Karkar, Kan Ding, Jay Harvey, and Phuc Nguyen
more interconnected, security and data privacy mechanisms to protect patients from theft or malicious actions must be considered
[3]. In addition to addressing technical complexities, sustainability
considerations play a vital role in the development of earable computers as a growing field. This involves reducing waste throughout
the entire value chain and contributing towards designing a more
circular device lifecycle. Overcoming these challenges is crucial to
fully leverage the potential of earables in medical applications.
Our Position: In this vision paper, we reflect on the growth of
earable computing and make an attempt to chart its near and longterm challenges and opportunities, particularly in a medical context.
Our vision covers three broad areas – technological capabilities,
user experience, and clinical implementation. Each of the three
areas is critical if we expect earables to move from an invention in
the lab to a medical device in the clinic to a tracker in the home.
Given the broad range of expertise in earables in the workshop,
we aim to present our vision and build on it as a community. We
welcome input from other earable researchers on the challenges,
opportunities, and the path to translation we imagine.
2023-10-01T00:00:00ZHardware in the Loop Test Stand for Battery Management System Validation in Shipboard Controls
http://hdl.handle.net/10106/31798
Hardware in the Loop Test Stand for Battery Management System Validation in Shipboard Controls
**Please note that the full text is embargoed until 8/1/2024** ABSTRACT: Traditionally, large-scale power systems have relied on power generation always exceeding the total load demand. Though this is most often the case, there are rare occurrences where failures in the generation, distribution, or even planning, that have led to situations where load demand quickly surpasses the generation capacity. In such cases, uncontrolled shedding of loads occurs, leading to brownouts or blackouts that can have catastrophic impact to both people and property. Advanced architectures, such as microgrids, offer a variety of improvements that are aimed at mitigating or even preventing these failures. Improvements include distributed power generation, active source and load monitoring, manageable load control, enhanced resilience against distribution failures, and in some instances the integration of energy storage to buffer transient loads. Despite all these benefits, system complexity is massively increased and heightened system monitoring is required, which is costly and difficult to implement.
Shipboard power system architectures resemble microgrids and the US Navy has proposed zonal shipboard power systems that employ low voltage (LV) AC, medium voltage (MV) AC, and even MVDC nodes within the same architecture. To both evaluate and validate such architectures, the Intelligent Distributed Energy Analysis Laboratory (IDEAL) testbed was established at the University of Texas at Arlington (UTA). IDEAL is intended to emulate one zone of a shipboard power system architecture that introduces various generation sources, power electronic converters, loads, and LFP energy storage, all of which are designed and assembled at naval relevant voltages. The platform is set up to enable Hardware in the Loop (HIL) model emulation allowing for the development, testing, and validation of control algorithms in a flexible emulated environment. Such an environment also allows for collaborative opportunities, thus facilitating a partnership with institutions like Florida State University (FSU), University of South Carolina (USC), and Clarkson University (CU), further enhancing research and innovation in the field.
Realizing the full potential of zonal architectures and the associated algorithms needed to reliably operate them, may require effective utilization of energy storage. However, safety concerns surrounding lithium-ion energy storage has many weighing the risks against the benefits it introduces. Lithium-ion batteries are most often managed using battery management systems (BMS) that monitor and manage the state of charge (SoC) of the multitude of cells that make up the battery. As battery configurations expand and as new BMSs are designed and introduced into the market, the need to study and validate their operation before they introduced into real batteries is critical. Furthermore, the ability to interface the BMS with the overarching system level controller is essential, and its operation must be validated across all potential use cases where intervention may be required. Power HIL (PHIL) platforms offer unique capabilities for emulating batteries comprised of multiple lithium-ion cells. Such capabilities include increased flexibility for rapidly studying the BMS’s reaction to normal and abnormal operating conditions, as well as a safe controlled environment for these types of scenarios. Thus, leading to increased user confidence and hopefully lead to the wider scale deployment of energy storage in shipboard power systems.
The work performed in this dissertation comprises of a few different, but interrelated thrusts. In the first, it discusses the design, rational, assembly and results obtained from a PHIL testbed used to validate BMS performance and software integration operating on batteries with up to 264 cells in series, ~1 kVDC. In the second thrust, a collaborative effort performed by UTA, FSU and USC is discussed in which the UTA IDEAL testbed was interfaced with remote HIL platforms being operated at FSU and USC, respectively. The remote HIL co-simulation effort demonstrated the employed physical energy storage at UTA and it was used to provide ramp rate buffering of a 12 kVDC bus emulated in the co-simulated HIL. In the third and final thrust, the IDEAL testbed is used to demonstrate the effectiveness of predictive high ramp rate (PHRR) and advanced load shed (ALS) algorithms developed by Clarkson University (CU) for maintaining operability and power quality within shipboard power systems deploying continuous and transient loads. Each of these thrusts will be discussed in detail.
2023-08-24T00:00:00ZMODEL OPTIMIZATION AND APPLICATIONS IN DEEP LEARNING
http://hdl.handle.net/10106/31781
MODEL OPTIMIZATION AND APPLICATIONS IN DEEP LEARNING
Machine learning refers to a machine or an algorithm that draws experience from data. A certain pattern is found to build a model, which is used to solve real problems.
Deep learning, an important branch and extension of machine learning, employs a neural network structure containing multiple hidden layers. It learns critical features of the data by combining lower-level features to form more abstract higher-level representations of attribute categories or features.
In this dissertation, deep learning network models were applied to sense-through-foliage target detection and extended with Rake structure. The deep learning network models had a large number of redundant parameters from the convolutional layer to the fully-connected layer, and a large number of neuron activation values converged to zero. The challenging task was to reduce parameter redundancy while maintaining model accuracy.
In Chapter 2, an approach based on stacked autoencoders (SAE) was proposed for ultra wide band radar for sense-through-foliage target detection. SAE, as one of the widely used deep learning structures, could learn representations of data with multiple levels of abstraction automatically. The SAE-based target detection approach performed well in processing poor signal collections in some positions. In other positions, a single radar target detection performed under satisfaction. Rake structure was applied in radar sensor networks with maximum ratio combining and equal combining to combine radar echoes from different radar cluster-members.
In Chapter 3, pruning in deep learning network models was investigated. Pruning presented significant opportunities for compression and acceleration in deep neural networks by eliminating redundant parameters. Structured pruning gained popularity in the edge computing research area, especially with more terminal chips integrated with AI accelerators for Internet of Things (IoT) devices. Stripe-wise pruning (SWP), which conducted pruning at the level of stripes in each filter, was different from filter pruning and group-wise pruning. The existing SWP method introduced filter skeleton (FS) to each stripe, setting an absolute threshold for the values in FS, and removing stripes whose corresponding values in FS could not meet the threshold. The research involved investigating the process of stripe-wise convolution and using the statistical properties of the weights located on each stripe to learn the importance between those stripes in a filter and remove stripes with low importance.
In Chapter 4, the conception of a deep energy autoencoder (EA) for a noncoherent multicarrier single-input and multiple-output (SIMO) system operating amidst multipath channels was explored. The multicarrier SIMO structure involved a single-antenna sender and a multi-antenna receiver, both depicted via neural networks. The encoder generated a real-valued vector for each subcarrier, while the decoder received the combination of energy from all the receiving antennas. To address the major challenge of mitigating intersymbol interference (ISI) caused by multipath channels without relying on delicate designs common in traditional communication systems, two different types of neural networks, namely DNN (Deep Neural Network) and RNN (Recurrent Neural Network), were adopted for the demodulation rule at the receiver. Simulation results demonstrated that, with adequate training, RNN efficiently recovered the transmitted data even in the absence of channel state information, which was often required in traditional communication systems.
2023-08-14T00:00:00ZEnvironment, communication and decision for multiagent systems
http://hdl.handle.net/10106/31761
Environment, communication and decision for multiagent systems
**Please note that the full text is embargoed until 8/1/2025** ABSTRACT: Multiagent systems (MAS) are ubiquitous in modern systems and have found broad applications, such as in intelligent transportation systems (ITS). Environment, communication and decision are among the essential components of MAS. The realistic environment in which MAS operates is usually stochastic, and its modeling, identification and estimation are important to consider. The communication in MAS is also critical for decisions. For example, in ITS, to improve travel efficiency and reduce traffic accidents, scheduling schemes for Vehicle-to-Everything (V2X) communication need to be developed. MAS decisions also need to be robust to uncertainties. For example, in mixed-traffic autonomous driving, the decisions for autonomous vehicles need to take into consideration human drivers’ uncertain behaviors to avoid crash and ensue safe driving. This dissertation contributes to the MAS research in the aforementioned three aspects: environment, communication and decision.
In the first thrust of the dissertation, we capture the stochastic spatiotemporal environment in which the MAS operates using a discrete-time stochastic model, namely the influence model (IM). The identifiability and estimation of IMs with reduced computation for real MAS applications are thoroughly studied in this dissertation, considering, first, a specific network topology (i.e., the uniform completely connected homogeneous networks), second, general homogeneous and heterogeneous networks, and finally, partially observed IMs (POIMs). Compared with using the standard master Markov chain approach for estimation, our proposed approaches are much more computationally efficient. In addition, per the authors’ knowledge, our work is the first in the literature that studies the identifiability of heterogeneous IMs and heterogeneous POIMs.
In the second thrust of the dissertation, we study sub-6 GHz assisted mmWave scheduling and design a distributed V2X communication scheduling scheme with multiples head nodes for long highway traffic. The long highway is divided into contiguous and non-overlapping sections, and a head node within each section collects mmWave link requests, runs the scheduler and coordinates with each other to achieve conflict-free schedules. A decomposition-based approximate solution is developed to address the intra-section computational scalability. Two coordination schemes are designed to address the inter-section communication scalability, and to achieve an overall conflict-free transmission schedule with low control overhead.
In the third thrust, we propose a stochastic hierarchical game (SHG) to support safe and efficient autonomous driving decision under uncertain intentions in mixed-traffic scenarios. First, a random mobility model (RMM) is developed to capture the uncertain intentions of MAS, including the random switching behavior. Then, an efficient sampling-based uncertainty evaluation technique, named the multivariate probabilistic collocation method integrated with an orthogonal fractional factorial design (MPCM-OFFD) is leveraged to solve the SHG with reduced computation by using a limited number of sample scenarios while guaranteeing the safety.
2023-07-13T00:00:00Z