High-performance computing platforms have become critical in meeting the demands of modern computing applications. Rising performance requirements in a broad range of platforms from mobile devices to server systems combined with the proliferation of these high-performance computing platforms has increased the energy costs incurred and lead to an exigent need for improvement in platform energy efficiency. This requires infrastructure for monitoring of energy consumption and methods to reduce the platform energy costs. In this dissertation, we present a new measurement infrastructure to provide real-time event-synchronized platform energy measurements, demonstration of these energy measurement capabilities through application to network data transport and an operating system task scheduler that utilizes these energy measurements to greatly improve energy efficiency for multi-core computing platforms.

The energy measurement infrastructure is integrated at the platform level and provides event-synchronized energy measurements for the complete platform along with important components such as the CPU, memory modules, secondary storage, peripherals and others. Furthermore, since modern secondary storage devices have buffering mechanisms that defer data write operations, the energy consumption of these operations is modeled and the model is integrated into the platform to characterize the impact of deferred operations.

The energy measurement capabilities are demonstrated through application to network data transport where a data file is transported over a network link. The data compression scheme is dynamically selected using real-time energy measurements during transport of the data file to enable adaptation to the dynamic system and network conditions. The energy cost of transporting the data file is significantly reduced through the use of this energy aware compression algorithm.

A novel task scheduler is presented and is designed to improve energy efficiency of multiprocessing platforms. It utilizes real-time energy measurements along with CPU performance monitoring units to identify inefficient tasks that suffer from co-run degradation due to resource contention. These inefficient tasks have their scheduling priority modified to avoid contention. Evaluation of the scheduler demonstrates large energy and execution time benefits on a quad-core platform.

Cost effective, wireless, wearable and robust systems are imperative for allowing us to shift our focus from the traditional event-driven healthcare model to a pervasive health awareness model where we are continuously monitoring a person's well-being. This capability will not only prove essential for timely diagnostics and early treatments, but will enable an array of tools for furthering scientific knowledge. Bringing these benefits to the orthopedic applications is of great important, since degeneration of joins is known to be a leading cause of chronic disability in the United States [1].

Monitoring acoustic emissions from joints has been previously studied to demonstrate that via quantization, clinically healthy and osteoarthritic joints can be distinguished. In the trials, piezoelectric transducers were used to detect sound waves emitted from knee joints during sit-stand-sit movements [5]. Such systems have previous been limited by the size and complexity of the acquisition unit, thus only a very limited range of activities have been monitored.

In this thesis, we present the novel Wireless Health Orthopedics (WHO) system, the developed hardware, as well as the accompanying software architecture. The specifics of the acoustic emissions (AE) system are shown in detail. We then explain how it interacts with the joint angle monitoring (JAM) measurements. We discuss the implementation of the holistic WHO system in healthy subject trials, which include sit-stand-sit, leg-extensions, walking and climbing stairs as movement sets. We further describe data post processing algorithms and techniques for synchronizing and visualizing the acquired measurements. Preliminary results indicate a quantitative difference between healthy young knee joints and healthy older knee joints. Further analysis allows for the discovery of distinct wave characteristics unique to certain movements and certain participant profiles. Finally, synchronized and superimposed acoustic emission and joint angle monitoring data is presented.

An orthopedic plaster or fiberglass cast is often applied after a bone fracture to hold the broken bone in place and allow the fracture to heal and the bone to set. If this cast becomes loose, due to a reduction in swelling for example, the bone may not heal properly and a misalignment of the bone may occur. If this problem goes unnoticed, surgery may need to be performed to re-break and reset the bone. This is particularly a problem with patients who are unable to accurately express what they are feeling, e.g. children, and who may not be able to identify that their cast is loose. The SmartCast system has been designed to avoid this complication and to identify early if a cast is not fitting well. SmartCast consists of a sleeve with novel textile sensors capable of measuring the pressure the cast is applying to an appendage as well as electronics to sample and log the sensor data. Several types of textile piezoresistive sensor sleeves have been designed and tested to determine which type would best be able to measure the cast pressure. A full prototype system, including firmware, sensor sleeve, and printed circuit board, has been designed, implemented and tested. The SmartCast system will allow doctors and orthopedic professionals to obtain a better understanding of how well a cast fits over time and to avoid the painful procedure of resetting a bone.

Deployments of sensors for real-world applications face critical challenges that can make large-scale adoption difficult or impossible to achieve due to prohibitive cost. Much of this cost can be reduced by enabling the adoption of more cost-effective behaviors by users and by guiding users on proper usage of technology solutions. Proper guidance for both subject behaviors and sensor usage training can reduce barriers to adoption by improving compliance and providing guarantees.

The aim of this research is to design and develop tractable and scalable monitoring system architectures for large-scale, real world usage that can guide users to adopt more beneficial behaviors. To achieve this aim, system designs were developed in direct support of monitoring applications to address problems in healthcare and energy conservation.

For energy conservation, the research used low-cost energy monitoring as a building block to test the effectiveness of various types of information feedback on real consumers in controlled experiments. In healthcare, we addressed critical unmet needs in post-operative patient care using interactive mobile technology solutions, integrated with physiological sensing to implement sensor usage guidance. Scalability was enabled by the development of hardware and software systems that allow subjects to easily perform data collection with guarantees on reliability and validity. To provide the complete system, end-to-end software and hardware architectures were developed to combine the core research with the other necessary components and field validations have been conducted.

Minder is a wearable device capable of continuously monitoring a user’s physiology and behaviors while delivering individualized guidance to keep users healthy. Unlike many wearable technologies currently on the market, Minder can be used to track athletic performance by providing accurate heart rate data during strenuous activity, as well posture quality, and context specific data such as the number of steps taken over the course of a day. This health information is transmitted over Bluetooth to a smart phone device where an app allows users to track their stats and receive guidance. This thesis discusses the process through which the Minder system was designed, and how the requirements of the system were addressed.

Three circuit boards were designed over the progression of this project starting with the first development board. Development board 1 allowed us to test two different design approaches and conduct preliminary testing of the parts selected for the system. The first approach used a central microcontroller to implement the core functionality of our system, while the second approach used the microcontroller inside the Bluetooth module to implement the core functionality as well as the Bluetooth stack. After testing, it was decided that the first approach would be best because it reduced the complexity of firmware development that would be required to implement the Minder functionality while not affecting the Bluetooth stack as well as made the system more flexible with the additional pins available on the standalone microcontroller. A second development board was designed to implement this approach while ensuring communication interfaces and GPIO were easily accessible via headers, pads and 0 ohm resistors to make the board easily tested and debugged. The final circuit board is currently being designed as a miniaturized fieldable prototype that will allow for system level testing of our design.

In addition to the circuit board design, firmware for a stand-alone ARM microcontroller and a Bluetooth Low Energy module were developed to implement the system functionality as well as over the air firmware update and serial bootloader capabilities.

Minder has been developed through the hard work of the entire Minder team. My focus has primarily been on the design of the first development board and the firmware development of the Bluetooth module, over the air update and serial bootloader.

The Minder system has been successfully tested on development boards, and is in the process of being miniaturized for further prototyping.

Human motion monitoring and activity classification, specifically in the free-living environment, are becoming increasingly important as preventative, diagnostic and rehabilitative measures in health and wellness applications.

In contrast to gait analysis, wearable sensor-based evaluation of upper body activities is not well studied. The work in this thesis tends to explore a novel system for upper limb activity monitoring and classification. The system focuses specifically on the application of motion classification to a complex task of automating rehabilitation evaluation, such as the Wolf Motor Function Test. The presented system consists of a novel wearable motion sensor platform that integrates accelerometers, gyroscopes and flex-sensors, and classification algorithms that convert motion data into an alphabet representation and form a string of primitives. String expressions are then derived for each test item and a regular expression based searching method is developed. We present results from the successful application of the proposed system to upper limb activity characterization in the context of the Wolf Motor Function Test.

Assurance of outcome is critical for patients that are afflicted with neurological disorders (for example stroke that introduces hemiparetic gait) and motion is a critical indicator of outcome. However outcome in the community environment is very difficult to determine. Wearable sensing system, due to its low cost, energy efficient, non-intrusive and applicable for large-scale development, is feasible for remote, continuous monitoring of motion and activities and thus provide outcome measures in the community. Towards this goal, this thesis explores the following problems:

1. A precise motion classification system is introduced.

It is applicable for diverse subjects whose motion ability ranges from very disabled to very capable and suitable for complicated community environment. Critical kinematic parameters of motion is modeled, characterized and validated, which provides direct outcome measures in the community. Those methods are applied to a large-scale trial with stroke patients. The problems encountered in large-scale trial and the corresponding engineering solutions are summarized.

2. For effective feedback provision, a PCA based visualization method is proposed, where gait quality evolution path is shown.

It also introduces a gait quality vector and in particular, a metric to measure uncertainty in gait.

3. The individualized motion classification model is extended into a population based analysis. First an indexing method for data dimension reduction is proposed. The indexing method expedites the processing speed and maintains least critical information loss for motion analysis. I also hypothesize and validate a power law exist between the diversity of motion (for example gait pattern) and the largest approximate cluster. Second, hierarchical clustering method is employed to derive the primitive gait patterns from the stroke population. The developed techniques help scale the motion classification system for big data.

Proper fulfillment of prescribed exercise has shown to greatly reduce the rate of patient re-admission and speed recovery of patients experiencing an array of health conditions. Currently, technology that allows doctors to monitor the fulfillment of exercise regiments requires patients to submit to expensive physical therapy and long term monitoring. An urgent need exists for low cost, rapidly deployable, and networked systems that enable monitoring of exercise activity for patients both in the clinic and at home.

The mFit exercise system utilizes modern sensor fusion techniques in conjunction with open source smartphone technologies to monitor: cadence, applied force, generated power, and dissipated energy while providing guidance and real time data to the user. The system also transmits patient data to a central server, which allows service providers to oversee patients, ensuring adherence to prescribed exercise regiments, and provide feedback as needed.

To accommodate patients in different stages of recovery, two variants of the mFit exercise system have been developed. The mFit restorator outfits a standard restorator bike with polymer force and Hall Effect sensors, enabling direct measurement of a user's force and cadence. Data is transmitted via Bluetooth to a smartphone that displays the information in an intuitive graphical user interface. The smartphone further utilizes WiFi or a mobile telecommunication network to transmit data to a central server which can be accessed by the user or a third party.

The mFit stationary cycle has been developed for patients that are prescribed more strenuous exercise routines. An inexpensive, mass market stationary cycle was modified and a strain gauge, common in most weight scales, and a photo-interrupter were added to enable force and cadence measurements. Similar to the mFit restorator, the data is transmitted to a smartphone that displays the results and simultaneously pushes the data to a central server.

In this thesis, we present the novel mFit system concept, the hardware used to develop the mFit restorator and mFit stationary cycle, as well as the software architecture which presents the required metrics. We further describe applications which interface with the mFit systems to provide guidance and measure a user's fitness level.

Biomedical devices in the past provided limited capability for the data acquisition and presented the data in the form of user interface for a care provider to observe. Now, what is required for biomedical devices has fundamentally changed. Many devices must now support secure networking and include a network of sensors to enable machine learning-based sensor fusion for accurate inference of the subject’s state.

This thesis introduces an Internet-of-Things (IoT) body area network (BAN) platform for medical devices that will provide rapid development capability with the assurance of security, networking, and the ability to host computationally intensive processes that are now required by medical devices. The BAN platform consists of seven wearable sensor nodes on the chest, wrists, upper legs, and ankles. Each sensor node includes sixteen general-purpose input/output (GPIO) pins, an analog-to-digital converter (ADC), two inter-integrated circuit (I2C) controllers, a serial peripheral interface (SPI), two universal asynchronous receiver transmitters (UART), and a universal serial bus (USB) on-the-go (OTG) to interface with sensors. The platform base model includes 9 degree-of-freedom inertial measurement unit (9DOF IMU) motion sensors, an electrocardiogram (ECG) sensor, a microphone, and a heart rate sensor. With its flexible interfaces, the platform is highly customizable and more sensors can be easily added.

Each sensor node features an IoT computer-on-module called the Intel Edison. The device can connect to expansion boards for rapid development. Although it has two official expansion options, the BAN platform uses boards from a third party manufacturer due to their small size. Intel provides a library to access the external interfaces. The library is fully compatible only if the Arduino breakout is used. A C library that abstracts /sys/class/gpio interface was developed to access the GPIO. The ADC device used in the platform is an I2C device. A C library was developed that abstracts the I2C communication between the Intel Edison and the ADC to provide an intuitive application program interface (API). The UART interface is accessible via /dev/ttyMFD2. A Python package called PySerial is used to interface the serial port. These interfaces in addition to the Intel’s official breakouts and library enable many more applications.

One of the most powerful features of the Intel Edison is its integrated WiFi module, enabling connection within the BAN and to the Internet. Since the BAN platform collects the user’s private health and activity data, the connection is secured by transport layer security (TLS). The networking among sensor nodes allows time synchronization with network time protocol (NTP) to have accurate sensor fusion.

Powered by its Intel Atom dual-core processors, the BAN platform can host neural network-based classifiers to monitor users’ states. From experiments, the performance of the neural network hosted on the platform was found to be on par with that of neural network implemented in Matlab.

The BAN platform was successfully distributed to freshmen, senior, and graduate IoT courses with exceptional assessment records. The IoT courses have shown that the students were able to rapidly develop fully functioning biomedical devices on the BAN platform.

ABSTRACT OF THE THESIS

Internet-of-Things

Instructional Platform

for Electrical and Computer Engineering

by

Xu Zhang

Master of Science in Electrical and Computer Engineering

University of California, Los Angeles, 2019

Professor William J. Kaiser, Chair

The motivation of this thesis is to design an interactive and scalable educational platform that delivers latest industrial technology and related electrical and computer engineering fundamentals to engineering students with inspiring and intuitive project examples. Current platform mainly demonstrates merged topics including Internet-of-Things (IoT) network, embedded system, data processing and computing, and machine learning.

The platform is composed of hardware system, software system, and curriculum system. The hardware system is consisted of STMicroelectronics SensorTile kits, STMicroelectronics Nucleo board, and Beaglebone, which provide different levels of IoT hardware components with reasonable costs. The software system is constituted with open-source integrated development environment (IDE), System WorkBench, which is a professional tool to support bare-metal microcontroller programming. The curriculum system has series of tutorials and reference designs, which provide various intuitive projects with detailed instructions and programming guidance.

There are nine tutorials covering fundamental topics including firmware level programming, sensor system signal acquisition, motion sensing, audio sampling and signal processing, Bluetooth low energy (BLE), and inertial sensing. Reference design provides complete, end-to-end experience in development of a system. This experience prepares developers for innovation and implementation of new systems.

This platform is initiated through UCLA Engineering 96C course, which becomes one of the engineering course requirement, and UCLA ECE 180D course, which is the senior capstone design course. UCLA students have collaborated together and developed novel systems for motion classification with SensorTile data sources and machine learning methods developed. Those extraordinary projects are tidily compiled as student project reference designs to inspire other future developments. In addition, the platform is widely adopted by various universities including Columbia University, California State University – North Bridge, Georgia Tech, University of Maine, and so on.