2019 ECE PhD Graduates
Alvaro Ulloa Cerna
Alvaro Ulloa Cerna defended his PhD thesis on Wednesday, July 3 at 8 am in room 118 of the ECE building. Dr. Marios Pattichis served as his committee chair. The title of Mr. Cerna's dissertation is, "Large Scale Electronic Health Record Data and Echocardiography Video Analysis for Mortality Risk Prediction."
Electronic health records contain the clinical history of patients. The enormous potential for discovery in such a rich dataset is hampered by their complexity. We hypothesize that machine learning models trained on EHR data can predict future clinical events significantly better than current models. We analyze an EHR database of 594,862 Echocardiography studies from 272,280 unique patients with both unsupervised and supervised machine learning techniques.
In the unsupervised approach, we first develop a simulation framework to evaluate a family of different clustering pipelines. We apply the optimized approach to 41,645 patients with heart failure without providing any survival information to the underlying clustering approach. The model separates patients with significantly different survival characteristics. For example, in a 10-cluster model, the minimum and maximum risk clusters had a median survival of 22 and 53 months respectively.
In the supervised approach, with 723,754 videos available from 27,028 unique patients, we assess the predictive capacity of Echocardiography video data for one- year mortality. Also, we hold out a balanced dataset of 600 patients to compare the model performance against cardiologists. We found that the best model, among four candidate architectures, is a 3D dyadic CNN model with an average AUC of 0.78 for a single parasternal long axis view. The model yields an accuracy of 75% (AUC of 0.8) on the held-out dataset while the cardiologists achieve 56% and 61%. The model performance was significantly higher than that of the cardiologists.
Finally, we develop a multi-modal supervised approach that enables interpretability. The model provides interpretations through polynomial transformations that describe the individual feature contribution and weights the transformed features to determine their importance. We validate our proposed approach using 31,278 videos from 26,793 patients. We test our proposed approach against logistic regression and non-linear and non-interpretable models based on Random Forests and XGBoost. Our results show that the proposed neural network architecture always outperforms logistic regression models while its performance approximates the other non-linear models. Overall, our multi-modal classifier based on 3D dyadic CNN and the interpretable neural network outperforms all other classifiers (AUC=0.83).
Brock Roberts defended his PhD thesis on Monday, April 15, 2019 at 1 pm in room 237 of the ECE Building. Dr. Edl Schamiloglu served as his committee chair. The title of Mr. Brock's dissertation is, "Noise and Gain Characterization of Interband Cascade Infrared Photodetectors."
A cavity designed to have multiple harmonic TM0N0 modes can be used to accurately measure the longitudinal profile of a bunched charged particle beam passing through its bore, non-invasively, and in real time.
Multi-harmonic TM0N0 cavities were designed, constructed, and beamline tested in a variety of experiments at the Thomas Jefferson National Accelerator Facility (TJNAF or Jlab). Measurements with a sampling oscilloscope provided signals that resemble the profile of electron bunches passing through the cavity’s bore. Straightforward signal processing techniques reduce distortion in the measurement and provide real time profiles of electron bunches with picosecond accuracy. Subharmonic beams having bunch repetition rates of 1/3rd and 1/6th of Jlab’s 1497 MHz bunch frequency, and interleaved sub-harmonic beams were also measured. Comparison between measurements made using a harmonic cavity were corroborated with an established invasive measurement method and with computer models. A harmonic cavity from this effort has been installed within the CEBAF injector, allowing accelerator operators to view, in real time, the shape and duration of electron bunches entering the accelerator. Another harmonic cavity has been installed within Jlab’s Upgraded Injector Test Facility (UITF), and two more are planned for installation there. This effort was awarded the 2016 International Beam Instrumentation Conference’s Faraday Cup Award.
Francisco German Perez Venegas
Francisco German Perez Venegas defended his PhD thesis on Friday, April 12 at 2:30 pm in room 118 of the ECE Building. Dr. Balu Santhanam served as his committee chair. The title of Mr. Venegas' dissertation is, "Detection and classification of vibrating objects in SAR images."
The vibratory response of buildings and machines contains key information that can be exploited to infer their operating conditions and to diagnose failures. Furthermore, since vibration signatures observed from the exterior surfaces of structures are intrinsically linked to the type of machinery operating inside of them, the ability to monitor vibrations remotely can enable the detection and identification of concealed machinery.
This dissertation focuses on developing novel detection schemes for performing detection and classification of vibrating objects in SAR images. Three are the central claims of this dissertation. First, the non-linear transformation that the microdoppler return of a vibrating object suffers through SAR sensing does not destroy its information. Second, the instantaneous frequency (IF) of the SAR signal has sufficient information to characterize vibrating objects. Third, it is possible to develop a detection model that encompasses multiple scenarios including both monocomponent and multi-component vibrating objects immersed in noise and clutter.
For answering these claims, two different schemes are studied for both the detection and classification of vibrating objects in SAR images. The first scheme is data-driven and utilizes features extracted with the discrete fractional Fourier transform (DFRF T) to feed machine-learning algorithms (MLAs). Specifically, the DFRFT is applied to the IF of the slow-time SAR data, which is reconstructed using techniques of time-frequency analysis. The second scheme is model-based and employs a probabilistic model of the SAR slow-time signal, the Karhunen-Loeve transform (KLT), and a likelihood-based decision function. The performance of the proposed schemes is characterized using simulated data as well as real SAR data collected with the Lynx SAR. The suitability of SAR for sensing vibrations is demonstrated by showing that the separability of different classes of vibrating objects is preserved even after non-linear SAR processing.
Finally, the proposed algorithms are studied in the presence of signal noise and terrain clutter. The results show that the proposed schemes produce high-precision classifiers capable of dealing with noise and clutter of moderate intensity. In order to loosen these requirements, the Hankel rank reduction (HRR) method, previously used for suppressing ocean clutter in ground-wave radar, is adapted to suppress clutter-noise in SAR images.
Lilian Casias defended her PhD thesis on Wednesday, March 20 at 10 am at the CHTM building. Dr. Ganesh Balakrishnan served as her committee chair. The title of Mr. Cerna's dissertation is, "Transport in Mid-Wavelength Infrared (MWIR) p- and n- type InAsSb and InAs/InAsSb Type-II Strained Layer Superlattices (T2SLs) for infrared detection."
III-V materials such as InAsSb ternaries and InAs/InAsSb Type-II Strained Layer Superlattices (T2SLs) have significant potential for infrared (IR) detector applications, including space-based detection, when utilized in a unipolar barrier detector architecture (nBn). However, recent studies revealed the quantum efficiency in nBn detectors degrades significantly faster from proton-irradiation induced displacement damage as compared to HgCdTe photodiodes. Improving the quantum efficiency radiation-tolerance is theoretically possible by enhancing vertical hole mobility and thereby the vertical hole diffusion length. The vertical hole mobility of T2SLs materials differs significantly from the lateral mobility and measuring it is much less straightforward.
In order to tackle vertical transport, in-plane or lateral transport must be better understood. There are added challenges to determining the in-plane bulk carrier concentration in narrow bandgap materials due to the potential for electron accumulation at the surface of the material and at its interface with the layer grown directly below it. Electron accumulation layers form high conductance electron channels that can dominate both resistivity and Hall-effect transport measurements. Therefore, to correctly determine the in-plane bulk concentration and mobility, temperature- and magnetic-ﬁeld-dependent transport measurements in conjunction with Multi-Carrier Fit (MCF) analysis were utilized on a series of p-doped InAs0.91Sb0.09 samples on GaSb substrates. The samples are etched to different thicknesses and variable-field measurements are utilized to assist in confirming whether a carrier species represents bulk, interface or surface conduction.
Secondly, n-type temperature- and magnetic-field dependent measurements on InAsSb and InAs/InAsSb T2SLs materials were performed to extract the in-plane transport properties for all the carriers present in each sample under two different doping concentrations (undoped and Silicon-doped). Lastly, substrate-removed, metal-semiconductor-metal (MSM) devices were fabricated to attempt vertical measurements, while standard van der Pauw structures were used for in-plane measurements. The MSM processing serves as a potential fabrication technique to measure vertical transport, that can be improved in the future. The goal of this dissertation is to accurately determine the lateral and vertical transport properties in the presence of multiple carrier species, Multi-Carrier Fit (MCF) and High-Resolution Mobility Spectrum Analysis (HR-MSA) were employed.
Eli Garduno defended his PhD thesis on Thursday Sept. 6, 2018 at 10 am in the CHTM Building. Dr. Ganesh Balakrishnan served as his committee chair. The title of Mr. Venegas' dissertation is, "Noise and Gain Characterization of Interband Cascade Infrared Photodetectors."
Infrared (IR) detectors are an enabling technology for a broad and growing list of applications including gas detection, night vision, and space-based missile warning. There are ongoing efforts in IR detector research to explore the potential of new material systems and energy band structures in addition to continuously improving their sensitivity through increasing their quantum efficiency and lowering their dark current and noise. This dissertation examines an emerging class of IR detectors known as Interband Cascade Infrared Photodetectors (ICIPs).
ICIPs contain multiple regions to facilitate the collection of photogenerated electrons and to limit unwanted dark current. Theory regarding their performance also indicates that multi-stage ICIPs may have lower noise than single-stage ICIPs and may provide improved detectivity in cases where the absorption coefficient of a material system is small and/ or where the diffusion length in the material is short or degraded.
In this work, four long-wavelength infrared ICIP devices with one, four, six, and eight stages were characterized at varying temperatures from 80 to 300 K and at biases up to one volt in both forward and reverse polarities. Noise spectra were collected on the four devices and show significant 1 / f noise that prevented direct measurement of the ICIP noise gain. The 1/ f noise in the ICIPs was linked to generation-recombination current. The devices were found to cause circuit instability when operated in bias regions with negative differential conductance (NDC) due to bias-dependent resonant tunneling. Additionally, bias-dependent photocurrent gain was observed using illumination of the devices with 632 nm and 1550 nm lasers which peaked near the NDC regions. This photocurrent gain was experimentally shown to be caused by current-mismatch between device stages, verifying theories regarding its origin.
Nicholas Tarasenko defended his PhD thesis on Thursday, February 21, 2019 at 1 pm in room 118 of the ECE Building. Dr. Christos Christodoulou served as his committee chair. The title of Mr. Brock's dissertation is, "DESIGN AND IMPLEMENTATION OF A 72 & 84 GHZ TERRESTRIAL PROPAGATION EXPERIMENT; EXPLOITATION OF NEXRAD DATA TO STATISTICALLY ESTIMATE RAIN ATTENUATION AT 72 GHZ."
The wireless communication sector is rapidly approaching network capacities as a direct result of increasing mobile broadband data demands. In response, the Federal Communications Commission allocated 71-76 GHz “V-band” and 81-86 GHz “W-band” for terrestrial and satellite broadcasting services. Movement by the telecommunication industry towards W/V-band operations is encumbered by a lack of validated and verified propagation models, specifically models to predict attenuation due to rain. Additionally, there is insufficient data available at W/V-bands to develop or test propagation models. The first aim of this study was the successful installation and operation of a terrestrial link to collect propagation data at W/V-band frequencies. In September 2015, the University of New Mexico, in collaboration with the Air Force Research Laboratory’s Space Vehicle Directorate, NASA’s Glenn Research Center and industry partners including (ACME, Applied Technology Associates, and Quinstar Technologies, Inc.) established the W/V-band Terrestrial Link Experiment (WTLE). WTLE was installed in the Albuquerque metro area with persistent tonal transmissions at 72 GHz and 84 GHz on a 23.5 km slanted path.
The second aim of this study was the utilization of the National Weather Service’s Next Generation Weather Radar (NEXRAD) system data to statistically estimate attenuation due to rain at 72 GHz. NEXRAD data provides a distributed sense of rain rates along WTLE’s path and alleviates challenges associated with instrumenting the 23.5 km link. Furthermore, NEXRAD data alleviates the need to develop complicated routines using in-situ meteorological measurements to estimate the size of the rain cell affecting the link. Non-linear regression techniques were applied on 2017 monsoon season data to obtain rain rate power law model coefficients. Testing of these coefficients was conducted on 2018 monsoon season data with satisfactory results. The techniques employed in this analysis represent a significant advancement in the ability to predict attenuation due to rain at 72 GHz for terrestrial links by enabling the use of historical archives of publicly available National Weather Service NEXRAD data. The technique has promising potential for estimation of path attenuation due to rain for links other than WTLE because of the vast nationwide coverage provided by NEXRAD systems.