Defense Notices

Future wireless networks will face a compound challenge of supporting large traffic volumes, providing ultra-reliable and low latency connections to ultra-dense mobile devices. To meet this challenge, various new technologies have been introduced among which mutual-information accumulation (MIA), an advanced physical (PHY) layer coding technique, has been shown to significantly improve the network performance. Since the PHY layer is the fundamental layer, MIA could potentially impact various network layers of a wireless network. Accordingly, the understanding of improving network design based on MIA is far from being fully developed. The purpose of this dissertation is to study the fundamental performance improvement of MIA over wireless networks and to apply these fundamental results to guide the design of practical systems, such as cognitive radio networks and massive machine type communication networks. 
This dissertation includes three parts. The first part of this dissertation presents the fundamental analysis of the performance of MIA over wireless networks. To begin with, we first analyze the asymptotic performance of MIA in an infinite 2-dimensional(2-D) grid network. Then, we investigate the optimal routing in cognitive radio networks with MIA and derive the closed-form cooperative gain obtained by applying MIA in cognitive radio networks. Finally, we characterize the performance of MIA in random networks using tools from stochastic geometry. 
The second and third part of this dissertation focuses on the application of MIA in cognitive radio networks and massive machine type communication networks. An optimization problem is formulated to identify the cooperative routing and optimal resources allocation to minimize the transmission delay in underlay cognitive radio networks with MIA. Efficient centralized as well as distributed algorithms are developed to solve this cross-layer optimization problem using the fundamental properties obtained in the first part of this dissertation. A new cooperative retransmission strategy is developed for massive MTC networks with MIA. Theoretical analysis of the new developed retransmission strategy is conducted using the same methodology developed in the fundamental part of this dissertation. Monte Carlo simulation results and numerical results are presented to verify our analysis as well as to show the performance improvement of our developed strategy. 

Advanced switching algorithms and modulation methods for power electronics converters controlled with model predictive control (MPC) strategies have been proposed in this work. The methods under study retain the advantage of conventional MPC methods in programing the nonlinear effects of the converter into the design calculations to improve the overall dynamic and steady state performance of the system and builds upon that by offering new modulation technique for MPC to minimize the voltage and current ripples through using a fixed switching frequency. The proposed method is easy to implement and provides flexibility to prioritize different objectives of the system against each other using the objective weighting factor. To demonstrate the effectiveness of the proposed method, it has been used to overcome the stability problems caused by a constant power load (CPL) in a multi converter system as a case study. 
In addition, to further evaluate the merits of the proposed method, it has been used to control modular multilevel converters (MMCs) in voltage source converter-high voltage DC (VSC-HVDC) systems. The proposed method considers the nonlinear properties of the MMC into the design calculations while minimizing the line total harmonic distortion (THD), circulating current ripple and steady-state error by generating modulated switching signals with a fixed switching frequency. In this work, the predictive modeling of the MMC is provided. Next, the proposed control method is described. Then, the application of the proposed method to a MMC system is detailed. Experimental results from the systems under study illustrate the effectiveness of proposed strategies. 

Through silicon via (TSV) based three-dimensional (3D) integrated circuit (IC) aims to stack and interconnect dies or wafers vertically. This forefront technology offers a promising near-term solution for further miniaturization and the performance improvement of electronic systems and follows a more than Moore strategy. 
Along with the need for low-cost and high-yield process technology, the successful application of TSV technology requires further optimization of the TSV electrical modeling and design. In the millimeter wave (mmW) frequency range, the root mean square (rms) height of the TSV sidewall roughness is comparable to the skin depth and hence becomes a critical factor for TSV modeling and analysis. The impact of TSV sidewall roughness on electrical performance, such as the loss and impedance alteration in the mmW frequency range, is examined and analyzed following the second order small perturbation method. Then, an accurate and efficient electrical model for TSVs has been proposed considering the TSV sidewall roughness effect, the skin effect, and the metal oxide semiconductor (MOS) effect. 
However, the emerging application of 3D integration involves an advanced bio-inspired computing system which is currently experiencing an explosion of interest. In neuromorphic computing, the high density membrane capacitor plays a key role in the synaptic signaling process, especially in the spike firing analog implementation of neurons. We proposed a novel 3D neuromorphic design architecture in which the redundant and dummy TSVs are reconfigured as membrane capacitors. This modification has been achieved by taking advantage of the metal insulator semiconductor (MIS) structure along the sidewall, strategically engineering the fixed oxide charges in depletion region surrounding the TSVs, and the addition of oxide layer around the bump without changing any process technology. Without increasing the circuit area, this reconfiguration of TSVs can result in substantial power consumption reduction and a significant boost to chip performance and efficiency. Also, depending on the availability of the TSVs, we proposed a novel CAD framework for TSV assignments based on the force-directed optimization and linear perturbation. 

Ice basal reflectivity is much needed for the determination of ice basal conditions and for the accurate modeling of ice sheet to estimate the future global mean sea level rise. Reflectivity values can be determined from the received radio echo sounding data if the power loss caused by different components along the two-way transmission of EM wave are accurately compensated. 

For the large volume of received radio echo sounding data collected over Byrd glacier in 2011-2012 with multichannel radar, the spherical spreading loss caused due to two-way propagation, power reduction due to roughness and relative englacial attenuation are compensated to estimate the relative reflectivity values of the Byrd glacier. 
In order to estimate the scattered incoherent power component due to roughness, the distributions of echo amplitudes returned from air-firn interface and from ice – bed interface are modeled to estimate RMS height variations. The englacial attenuation rate of wave for two-way propagation along the ice depth is modeled using the observed data. The estimated air-firn interface roughness parameters are relatively cross verified using the Neal’s method and with the correlations from the Landsat image mosaic of Antarctica. Estimated relative basal reflectivity values are validated using the cross-over analysis and abruptness index measurements. From the Byrd relative reflectivity map, the corresponding echograms at the locations of potential subglacial water systems are checked for the observable lake features. 
The obtained results are checked for correlations with previously predicted lake locations and subglacial flow paths. While the results doesn’t exactly match with the previously identified locations with elevation changes, high relative reflectivity values are observed close to those locations, aligning exactly or close to previously predicted flow paths providing a new window into the hydrological network of the glacial. Relative reflectivity values are clustered to indicate the different potential basal conditions beneath the Byrd glacier 

Rule induction is an important technique of data mining or machine learning. Knowledge is frequently expressed by rules in many areas of AI, including rule based expert systems. The machine learning/ data mining system LERS (Learning from Examples based on Rough Sets) induces a set of rules from examples and classifies new examples using the set of rules induced previously by LERS. LERS induces rules based on supervised learning. The MLEM2 algorithm is a rule induction algorithm in which rule induction, discretization, and handling missing attribute values are all conducted simultaneously. A rule checker is implemented to classify new cases using the rules induced by MLEM2 algorithm. MLEM2 algorithm induces certain and possible rule sets. Bucket Brigade algorithm is implemented to 
classify new examples. K-fold cross-validation technique is implemented to measure the performance of MLEM2 algorithm. The objective of this project is to find out the efficiency of the MLEM2 rule induction method for incomplete data set. 

Rough set theory is a popular approach for decision rule induction. However, it requires the objects in the information system to be completely described. Many real life data sets are incomplete, so we cannot directly apply rough set theory for rule induction. A characteristic relation is used to deal with incomplete information systems in which ‘do not care’ data coexist with lost data. There are scenarios in which two objects that do not have the same known value are indiscernible and on the other hand the two objects which have a lot of equivalent known values are very likely to be in different classes. To rectify such situations, a modification of the characteristic relation was introduced. This project implements rule induction from the modification of the characteristic relation for incomplete data sets.

In this project, an idea of the maximal consistent block is applied to formulate a new approximation to a concept in incomplete data sets. The maximal consistent blocks have smaller cardinality compared to characteristic sets. Because of this, the generated upper approximations will be smaller in size. Two interpretations of missing attribute values are discussed: lost values and “do not care” conditions. Four incomplete data sets are used for experiments with varying levels of missing information. Maximal Consistent Blocks and Characteristics Sets are compared in terms of cardinality of lower and upper approximations. The next objective is to compare the decision rules induced and cases covered by both techniques. The experiments show that both techniques provide the same lower approximations for all the datasets with “do not care” conditions. The best results are achieved by maximal consistent blocks for upper approximations for three datasets.

Data classification is a methodology of data mining used to organize data by relevant categories to obtain meaningful information. A model is generated from the input training set which is used to classify the test data into predetermined groups or classes. One of the most widely used models is a decision tree which uses a tree like structure to list all possible outcomes. Decision tree is an important predictive analysis method in Data Mining as it requires minimum effort from the users for data interpretation. 

This project implements ID3, an algorithm for building decision tree using information gain metric. Furthermore, through illustrating the basic ideas of ID3, this project also addresses the inefficiency of ID3 in handling continuous numerical attributes. A cut point approach is presented to discretize the numeric attributes into discrete intervals and enable ID3 functionality for them. Experiments show that such decision trees contain fewer number of nodes and branches in contrast to a tree obtained by basic ID3 algorithm. This modified algorithm can be used to classify real valued domains containing symbolic and numeric attributes with multiple discrete outcomes. 

Data sets may have instances where multiple values are possible which are described as set-value attributes. The established LEM2 algorithm does not handle data sets with set-value attributes. To solve this problem, a parallel approach was used during LEM2's execution to avoid preprocessing data. Changing the creation of characteristic sets and attribute-value blocks to include all values for each case allows LEM2 to induce rules on data sets with set-value attributes. The ability to create a single local covering for set-value data sets increases the variety of data LEM2 can process.

Machine learning is being applied in many domains of research. One such research area is the automation of gender prediction. The goal of this project is to determine a person’s gender based on his/her voice. Although it may seem like a simple task for any human to recognize this, the difficulty lies in the process of training a computer to do this job for us. This project is implemented by training models based on input data of voice samples from both male and female voices. The voice samples considered were from different datasets, with varying frequencies, noise ratios etc. This input data is passed through various machine learning models, with/without parameter tuning, to compare results. A comparative analysis of multiple machine learning algorithms was conducted, and the prediction with the highest accuracy is displayed as output for the given input voice sample.