Department of Electrical and Computer Engineering Graduate Theses

Permanent URI for this collection


Recent Submissions

Now showing 1 - 5 of 449
  • Item
    Objective Methods for Speech Intelligibility Prediction
    Alghamdi, Ahmed; Electrical and Computer Engineering; Chan, Wai Yip Geoffrey
    The focus of this thesis is on reference-based speech intelligibility predictors (SIPs) which compute their outputs based on a similarity metric between the auditory representations of the clean and degraded signals. Existing SIPs do not work consistently in conditions involving modulated noise (non-stationary noise commonly encountered in real-world settings). In addition, the computational cost of many SIPs are high which makes them unsuitable for use in an iterative design scenario. The main goal of this thesis is to design intelligibility predictors that perform well across a wide range of degradations with a focus on keeping low execution time. To tackle this task, we proposed three SIPs that employ a low complexity auditory model and use an improved similarity metric to quantify the effect of degradation. The parameters of the proposed SIPs were either optimized to maximize correlation with subjective intelligibility scores or determined based on knowledge drawn from psychoacoustical studies about the acoustic ques relevant to speech intelligibility. For performance evaluation we used 16 subjective datasets involving speech corrupted by modulated noise, nonlinear processing, and reverberation. The evaluation results show that the proposed SIPs works well across all examined conditions and on average outperforms most of the baseline measures. Moreover, the proposed SIPs have the lowest execution time among all baseline measures.
  • Item
    A Novel Multilevel Current-Driven Full Bridge Converter for Wide Input Voltage Range Applications
    Woelfle, Justin S.; Electrical and Computer Engineering; Pahlevani, Majid
    Climate change has driven the implementation of renewable energy resources, such as solar, to new heights in the past decade. Power converters utilized in solar microinverters should be highly efficient over a wide range of operating conditions. Industry-standard converters, such as the flyback converter and the LLC resonant converter, suffer in performance as the operating point deviates from the nominal design point. The deviation from the nominal operating point is caused by variations in panel manufacturing processes and changing ambient conditions, such as irradiance and temperature. In this thesis, a novel multilevel current-driven full-bridge (MLCDFB) converter alongside a new pulse voltage modulation (PVM) scheme is proposed to address the limitations of existing converters. The multilevel structure of the converter enables great flexibility in the operation over a wide input voltage range. The proposed converter and modulation scheme are presented and analyzed at steady state to illustrate this flexibility. Design considerations for the converter are then derived using analytical, numerical, and heuristic analysis. Extensive experimental results are then presented to show the efficacy of the proposed converter and modulation scheme under a wide range of operating conditions. The peak efficiency of the converter is 98.01%. At the nominal input voltage, the converter achieves a CEC efficiency of 97.29%. Over an input voltage range of 34-56 V and a power of 500 W, the converter achieves efficiencies between 97.16% and 97.54%. Over the same voltage range and a power range of 300-500 W, the proposed converter efficiency varies from 97.02% to 97.91%. Metrics to compare the proposed converter to other full-bridge-derived converters, such as the LLC converter, are then proposed. The LLC converter is chosen because it is an industry-standard converter known for having a high efficiency. The metrics compare various loss factors, such as the RMS current, the peak current, and switching currents to the power transferred by the converter. The variation of switching frequency is also considered. These metrics describe how the losses of a converter grow as the operating point changes. An experimental comparison between the MLCDFB converter and the LLC converter is then made. The LLC converter achieves a peak efficiency of 97.60% at 500 W; however, its efficiency drastically decreases at high input voltages, down to 94.38%. This makes it clear that the proposed MLCDFB converter outperforms the LLC converter over a wide input voltage range.
  • Item
    Physical Layer Security in Millimeter-Wave Communication
    He, Miao; Electrical and Computer Engineering; Ni, Jianbing
    Due to operating at high carrier frequencies (30-300 GHz), the millimeter-wave (mmWave) communication has many advantages. To preserve the secrecy for the mmWave wireless communication, in addition to the data encryption with cryptographic keys, physical layer security (PLS) is another effective technique to achieve secret communication. The inherent randomness of wireless channels are exploited to ensure the confidential information carried by the signals can only be obtained by the target users, but not by the unauthorized users. The implementation of the PLS mmWave communication systems should consider the characteristics of the mmWave channels, as well as some requirements (computational complexity, hardware cost and multicast demand) and practical limitations (imperfect hardware) in engineering. First, we consider to reduce the computational complexities in the PLS mmWave communication systems. The high computational complexities are the barriers for practical implementations. Second, we consider the demand on multicast transmission in practical implementations. Currently, a few PLS schemes support secret multicast transmissions by solving the optimization problems with the conditions of abundant computational resources and known eavesdroppers' channel state information. Such conditions may not be met in practical implementations. In addition, we consider the phase quantization errors of the phase shifters in the phased-array antennas. The phase quantization errors have adverse effects on the PLS mmWave communication performances. Lastly, we consider to reduce the hardware costs and simplify the structure complexities of the PLS mmWave communication systems in practical implementations. In order to reduce the heavy computational burdens, a light-weight algorithm for the phased-array PLS mmWave communication system is proposed. For the multicast transmission, we propose a scheme based on an integrated oblique projection approach. The proposed scheme achieves multicast PLS mmWave communication without the knowledge of eavesdroppers' channel state information, at low computational complexities. Based on the analyses on the transmission gain with phase shifter quantization errors, we develop a compensation scheme to address the phase deviation caused by phase quantization errors. To reduce the hardware costs, we propose a scheme with an analog dual-shifter-structure, which achieves multicast PLS mmWave communication at affordable hardware costs.
  • Item
    Uncoupled Stability of Haptic Simulation of Viscoelastic Virtual Environments
    Oliver, Seanna M.; Electrical and Computer Engineering; Hashtrudi-Zaad, Keyvan
    Haptic simulation systems enable users to kinesthetically interact with virtual environments through interaction with a robotic mechanism, known as a haptic device. Application of these systems are in medical simulators, robotic rehabilitation, and entertainment. Viscoelastic medium is a common environment that is simulated as a virtual environment (VE) in haptic simulation systems. Kelvin-Voigt (KV) and Hunt-Crossley (HC) are models commonly used to simulate viscoelastic environments in haptic simulation systems. Due to the sample-and-hold process, the range of dynamics - viscosity and elasticity, that can be rendered in a stable way is limited. While uncoupled stability, as a stringent stability condition, has been analyzed for KV VEs, it has not been evaluated for HC environments. In this thesis, we experimentally evaluate the range of dynamic parameters for each model that result in uncoupled stability. To compare the results, we map the HC parameters to the KV parameter space. Results show that higher values of n_HC are more suitable for interactions with stiffer environments, at the expense of damping, and compared to the dynamic range obtained from using KV VE, the HC model offers lower stiffness but higher damping range. To confirm the mapping, we conduct a user study to compare the viscosity and elasticity effects perceived by the users. Results show that different penetrations may not affect the stable HC elastic and viscous ranges, but it affects the feel of the HC modeled environment, as is reflected by the identified KV ranges, which confirms that the HC model is penetration dependent. In addition to determining the stable implementable range of HC and KV parameters for non-deformable environments, the stable implementable range of physical deformable objects implemented as a VE is also found. Physical gels with known Young’s Moduli are experimentally tested to determine their point of contact apparent stiffness and damping for varying penetration. These parameters are utilized to implement a spring-mass-damper mesh VE model for each gel. These values are also interpolated and extrapolated to implement a larger range of deformable VEs for uncoupled stability tests. The results show a viscoelastic dynamic range that depends on the depth of penetration.
  • Item
    Multi-Muscle High-Density Electromyography for Improved End-Point Force Prediction
    Johns, Greggory A.; Electrical and Computer Engineering; Morin, Evelyn
    Muscles play a fundamental role in human interaction with the physical world. Coordinated control of multiple muscles enables the execution of complex motor tasks with varying force, dexterity, and postures. Understanding neuromuscular function holds significance in various fields, including kinesiological research, sports, rehabilitation medicine, and robotics. The electromyographic (EMG) signal is widely utilized for muscle force estimation, but challenges persist due to physiological and non-physiological factors. High-density electromyography (HD-EMG) shows promise in improving accuracy, but its high dimensionality poses challenges. This research investigates the use of HD-EMG coupled with an ensemble learning algorithm and shows promising results for improved end-point force estimation performance. An ensemble fast orthogonal search (EFOS) algorithm was explored and it exhibited a monotonic increase in performance as channel density increased in both isotonic and non-isotonic isometric contractions in the horizontal plane. Specifically, an inter-session pRMSE of 3.03 (1.63) was obtained for isotonic, and 9.01 (4.07) for non-isotonic contractions, for the highest channel density. This research also showed that it is important to consider the temporal mismatch between EMG and force data induced by both electromechanical delay as well as filtering steps in pre-processing, where signal variability must be balanced with induced delay.