Computer Engineering Commons^™

Gen-Acceleration: Pioneering Work For Hardware Accelerator Generation Using Large Language Models, Durga Lakshmi Venkata Deepak Vungarala

Theses

Optimizing computational power is critical in the age of data-intensive applications and Artificial Intelligence (AI)/Machine Learning (ML). While facing challenging bottlenecks, conventional Von-Neumann architecture with implementing such huge tasks looks seemingly impossible. Hardware Accelerators are critical in efficiently deploying these technologies and have been vastly explored in edge devices. This study explores a state-of-the-art hardware accelerator; Gemmini is studied; we leveraged the open-sourced tool. Furthermore, we developed a Hardware Accelerator in the study we compared with the Non-Von-Neumann architecture. Gemmini is renowned for efficient matrix multiplication, but configuring it for specific tasks requires manual effort and expertise. We propose implementing …

Modeling Of Quad-Station Module Cluster Tools Using Petri Nets, Aung Nay

Theses

The semiconductor industry is highly competitive, and with the recent chip shortage, the throughput of wafers has become more important than ever. One of the tools that the industry has deployed is to use of quad-station modules instead of the traditional single-station modules that allow for higher throughput and better wafer consistency by processing multiple wafers at the same time and distributing work. The industry trend is to use multiple transfer chamber robots to stack the quad-station modules in a series, particularly for etch products. In this work, the quad-station cluster tool wafer movement is modeled by using Petri net …

Un-Fair Trojan: Targeted Backdoor Attacks Against Model Fairness, Nicholas Furth

Theses

Machine learning models have been shown to be vulnerable against various backdoor and data poisoning attacks that adversely affect model behavior. Additionally, these attacks have been shown to make unfair predictions with respect to certain protected features. In federated learning, multiple local models contribute to a single global model communicating only using local gradients, the issue of attacks become more prevalent and complex. Previously published works revolve around solving these issues both individually and jointly. However, there has been little study on the effects of attacks against model fairness. Demonstrated in this work, a flexible attack, which we call Un-Fair …

Design And Implementation Of Photovoltaic Energy Harvesting Automaton, Iskandar Askarov

Theses

Global domestic electricity consumption has been rapidly increasing in the past three decades. In fact, from 1990 to 2020, consumption has more than doubled from 10,120 TWh to 23,177 TWh [1]. Moreover, consumers have been turning more towards clean, renewable energy sources such as Photovoltaic. According to International Energy Agency, global Solar power generation alone in 2019 has reached almost 3% [4] of the electricity supply. Even though the efficiency of photovoltaic panels has been growing, presently, the highest efficiency solar panels available to an average consumer range only from 20%-22% [14]. Many research papers have been published to increase …

Treated Hfo2 Based Rram Devices With Ru, Tan, Tin As Top Electrode For In-Memory Computing Hardware, Yuvraj Dineshkumar Patel

Theses

The scalability and power efficiency of the conventional CMOS technology is steadily coming to a halt due to increasing problems and challenges in fabrication technology. Many non-volatile memory devices have emerged recently to meet the scaling challenges. Memory devices such as RRAMs or ReRAM (Resistive Random-Access Memory) have proved to be a promising candidate for analog in memory computing applications related to inference and learning in artificial intelligence. A RRAM cell has a MIM (Metal insulator metal) structure that exhibits reversible resistive switching on application of positive or negative voltage. But detailed studies on the power consumption, repeatability and retention …

Human Activity Recognition Using Wearable Sensors: A Deep Learning Approach, Jialun Xue

Theses

In the past decades, Human Activity Recognition (HAR) grabbed considerable research attentions from a wide range of pattern recognition and human–computer interaction researchers due to its prominent applications such as smart home health care. The wealth of information requires efficient classification and analysis methods. Deep learning represents a promising technique for large-scale data analytics. There are various ways of using different sensors for human activity recognition in a smartly controlled environment. Among them, physical human activity recognition through wearable sensors provides valuable information about an individual’s degree of functional ability and lifestyle. There is abundant research that works upon real …

Multiframe Coded Computation For Distributed Uplink Channel Decoding, Brinell F. Monteiro

Theses

The latest 5G technology in wireless communication has led to an increasing demand for higher data rates and low latencies. The overall latency of the system in a cloud radio access network is greatly affected by the decoding latency in the uplink channel. Various proposed solutions suggest using network function virtualization (NFV). NFV is the process of decoupling the network functions from hardware appliances. This provides the exibility to implement distributed computing and network coding to effectively reduce the decoding latency and improve the reliability of the system. To ensure the system is cost effective, commercial off the shelf (COTS) …

Multi-Population-Based Differential Evolution Algorithm For Optimization Problems, Ishani Chatterjee

Theses

A differential evolution (DE) algorithm is an evolutionary algorithm for optimization problems over a continuous domain. To solve high dimensional global optimization problems, this work investigates the performance of differential evolution algorithms under a multi-population strategy. The original DE algorithm generates an initial set of suitable solutions. The multi-population strategy divides the set into several subsets. These subsets evolve independently and connect with each other according to the DE algorithm. This helps in preserving the diversity of the initial set. Furthermore, a comparison of combination of different mutation techniques on several optimization algorithms is studied to verify their performance. Finally, …

Preemptive Efficient Queueing For Scheduling Deadline-Bound Data Center Flows, Vinaykrishna Gopalakrishna

Theses

This thesis introduces a new deadline aware flow scheduling scheme called Preemptive Efficient Queuing (PEQ). Unlike other schemes which use policies like EDF, SJF or fair share, this scheme aims to procrastinate flows as much as they can afford to, while keeping with their deadlines. Thereby, PEQ prioritizes more urgent flows that have shorter time to procrastinate, or cushion time. Results of PEQ are compared with state-of-the-art schemes for the transport of data-center flows, such as Preemptive Distributed Quick (PDQ) and Deadline-Driven Delivery (D³). The results show that PEQ outperforms D³ and PDQ. We identify results …

Decision Tree Rule-Based Feature Selection For Imbalanced Data, Haoyue Liu

Theses

A class imbalance problem appears in many real world applications, e.g., fault diagnosis, text categorization and fraud detection. When dealing with an imbalanced dataset, feature selection becomes an important issue. To address it, this work proposes a feature selection method that is based on a decision tree rule and weighted Gini index. The effectiveness of the proposed methods is verified by classifying a dataset from Santander Bank and two datasets from UCI machine learning repository. The results show that our methods can achieve higher Area Under the Curve (AUC) and F-measure. We also compare them with filter-based feature selection approaches, …

Source-Channel Coding For Coordination Over A Noisy Two-Node Network, Sarah A. Obead

Theses

Recently, the concept of coordinating actions between distributed agents has emerged in the information theory literature. It was first introduced by Cuff in 2008 for the point-to-point case of coordination. However, Cuff’s work and the vast majority of the follow-up research are based on establishing coordination over noise-free communication links. In contrast, this thesis investigates the open problem of coordination over noisy point-to-point links. The aim of this study is to examine Shannon’s source-channel separation theorem in the context of coordination. To that end, a general joint scheme to achieve the strong notion of coordination over a discrete memoryless channel …

High-Performance Matrix Multiplication On Intel And Fgpa Platforms, Gang Li

Theses

Matrix multiplication is at the core of high-performance numerical computation. Software methods of accelerating matrix multiplication fall into two categories. One is based on calculation simplification. The other one is based on increasing the memory access efficiency. Also matrix multiplication can be accelerated using vector processors. In this investigation, various matrix multiplication algorithms and the vector-based hardware acceleration method are analyzed and compared in terms of performance and memory requirements. Results are shown for Intel and Xilinx FPGA platforms. They show that when the CPU is fast, Goto's algorithm runs faster than Strassen's algorithm because the data access speed is …

Modeling Next Generation Air Traffic Control System With Petri Net, Hang Wu

Theses

The Federal Aviation Administration (FAA) is one of the largest Air Navigation Service Providers, managing air traffic for more than 15% of the world's airspace. Today's Air Traffic Control (ATC) system cannot meet the growth of the air traffic activities, which brings with more unprecedented delays. At the same time, Air Traffic Controllers are facing higher workload than ever before. The FAA has declared that the existing ATC system will transition to a new system known as “Free Flight”. “Free Flight” will change today's ATC system by giving pilots increased flexibility to choose and modify their routes in real time, …

Energy Scavenging System For Indoor Wireless Sensor Network, David Andrew Crea

Theses

As wireless communication evolves wireless sensors have begun to be integrated into society more and more. As these sensors are used to a greater extent newer and better ways to keep them working optimally have begun to surface. One such method aims to further the sensors energy independence on humans. This technique is known as energy scavenging. The logic behind energy scavenging is to allow the device to have its own reliable source of energy that does not require upkeep, has a long life expectancy, and does not completely rely on an internal power source. The aim of this thesis …

Kerberos Phone Secure Messenger, Nabeel Al-Saber

Theses

Security is becoming vital in today's open insecure Internet. While popular Internet enabled mobile devices are spreading widely, the security of such platforms is not maturely addressed. This research extends the popular Kerberos authentication protocol to run on mobile phones and builds a novel Kerberos Secure Phone Messenger (KSPM) on top of the protocol. Moreover, the Kerberos network authentication protocol provides user authentication and message privacy with the convenience of secret key cryptography. Such an advantage in mobile phones helps reduce the computational burden and power consumption if compared with public key cryptography. KSPM achieves high standards in terms of …

Retransmission Strategies For Moving Networks Over Ricean Fading Channels, Alex Ruiz Lee

Theses

The problem of designing retransmission protocols based on Hybrid Automatic Repeat reQuest (HARQ) for moving networks over Ricean fading channels has been considered. The scenario of interest consists of one receiver traveling at a given speed while connected to a wireless network based on OFDM (e.g., WiMax) through several base stations located along the path. Five different retransmission protocols that combine HARQ with handover between adjacent base stations are considered. Their performance in terms of number of retransmissions per packet via numerical simulations, considering both frequency selective and nonselective channels are presented. The results obtained show the relevant advantage that …

Proactive Methods For Measurement Of Available Bandwidth And Link Capacity, Khondaker Musfakus Salehin

Theses

With the continuous expansion of network infrastructure and deployment of applications sensitive to Quality-of-Service, network measurement plays a major role in both network planning and management. Accurate measurement of various network parameters, e.g., available bandwidth, link capacity, delay, packet loss and jitter, provides a positive impact for effective traffic engineering, Quality-of-Service (QoS) routing, optimization of end-to-end transport performance, and link capacity planning. For network measurement, there exists several proactive estimation tools based on either probe-gap model or probe-rate model that estimate path related attributes. Most of these tools that have been implemented can measure tight-link capacity (smallest available bandwidth) and/or …

Two Factor Authentication And Authorization In Ubiquitous Mobile Computing, Qing Zhu

Theses

Handheld devices, such as mobile phones, PDAs and others, have become an integrated part of our lives. They are perfectly suitable to become the first real-life platforms for mobile computing applications. A lightweight holistic authentication and authorization scheme is presented for this typical wireless scenario. It is shown that either SIM or Kerberos authentication has its limitations. A two-factor authentication method SIM-based Kerberos authentication is developed to implement the Secure Single Sign-on function with LDAP protocol realization for authorization and mobile directory services. Performance analysis and power consumption analyses for this protocol are presented as well. Our results show that …

Route Recovery Schemes For Link And Node Failure And Link Congestion, Ibrahim Takouna

Theses

Link/Node failure occurs frequently causing service disruption in computer networks. Hardware techniques have been developed to protect the network from Link/Node failure. These techniques work in physical layer, therefore their convergence time is very small. On the other hand, many schemes have been proposed to mitigate the failure influence on the network. These schemes work in upper layers such as the network layer. However, hardware solutions faster than other schemes, but they are expensive. Link/Node failure causes all flows which were using the failed link/node are temporarily interrupted till a new path reestablished.

Three recovery algorithms have been proposed that …

Development And Evaluation Of A Simultaneous Multithreading Processor Simulator, Carla Verena S. Nunez

Theses

Modem processors are designed to achieve greater amounts of instruction level parallelism (ILP) and thread level parallelism (TLP). Simultaneous multithreading (SMT) is an architecture that exploits both LLP and TLP. It improves the utilization of the processor resources by allowing multiple independent threads to reside in the pipeline and dynamically scheduling the available resources among the threads.

The first part of this thesis presents the development of a simultaneous multithreading processor simulator. The SMT simulator is derived from SimpleScalar, a superscalar processor simulator widely used in the computer architecture research field. The basic pipeline is expanded to allow multiple threads …

Non-Repudiation Secure File Transfer Protocol (Nrsftp), Jerry Chen

Theses

Non Repudiation Secure File Transfer Protocol (NRSFTP) is designed to resolve three main concerns for today's electronic file transfer methodology. The three main concerns are Non-Repudiation, Secure, and Non-Real Time file transfer. Non-repudiation is to assure the receiver that the sender of the document is not an imposter. Secure document transfer is to assure the sender that only the intended receiver will be able to read the document. Non-real-time file transfer is to provide convenient and low cost transportability of the encrypted data from one party to another. With the above three concerns addressed, the NRSFTP protocol can be widely …

A Wireless Method For Monitoring Medication Compliance, Jeffrey Scott Jonas

Theses

There are many devices on the market to help remind patients to take their pills, but most require observation by a caregiver to assure medication compliance. This project demonstrates three modes to detect pill removal from a pillbox: a switch under the pills, a reflective type photointerrupter and a transmissive "electric eye" photosensor. Each mode exhibited blind spots or other failures to detect pill presence, but by combining modes with complementary characteristics, the accuracy of pill detection is greatly increased.

Two methods of caregiver notification are demonstrated: text messages transmitted via an attached cellular phone, or the status is collected …

Fpga-Based Implementation Of Parallel Graph Partitioning, Mohammad Kharashgeh

Theses

Graph partitioning is a very important application that can be found in numerous areas, from finite element methods to data processing and VLSI circuit design. Many algorithms have been developed to solve this problem. Of special interest is multilevel graph partitioning that provides a very efficient solution. This method can also be parallelized and implemented on various multiprocessor architectures. Unfortunately, the target of such implementations is often unavailable high-end multiprocessor systems. Here a parallel version of this method for an in-house developed multiprocessor system is implemented on an FPGA. The system designed provides a cost-effective solution.

The design is based …

Classification, Testing And Optimization Of Intrusion Detection Systems, Javier Leon

Theses

Modem network security products vary greatly in their underlying technology and architecture. Since the introduction of intrusion detection decades ago, intrusion detection technologies have continued to evolve rapidly. This rapid change has led to the introduction of a wealth of security devices, technologies and algorithms that perform functions originally associated with intrusion detection systems.

This thesis offers an analysis of intrusion detection technologies, proposing a new classification system for intrusion detection systems. Working closely with the development of a new intrusion detection product, this thesis introduces a method of testing related technologies in a production environment by outlining and executing …

Location Estimation In A 3d Environment Using Radio Frequency Identification Tags, Adwitiya Akash Jain

Theses

RFID tag location estimation in a 3D environment is investigated. The location of the tag with unknown coordinates can be estimated with certain accuracy. However, accuracy can be improved using the knowledge based on measurement of additional reference tags with known location. This thesis studies the mathematical formulation and practical realization of location sensing using RFID tags.

Deviating from the standard use of RFID technology which employs one tag reader to identify the presence of tag, here multiple tag readers with known location are used to estimate the physical location of an individual tag, with/without the help of few reference …

New Hybrid Automatic Repeat Request (Harq) Scheme For 4x4 Mimo System, Based On The Extended Alamouti Quasi-Orthogonal Space-Time Bloc Coding (Q-Stbc), In Invariant And Variant Fading Channel, Jordi Ferrer Torras

Theses

A new Hybrid Automatic Repeat reQuest (HARQ) combining scheme for a 4x4 Multiple Input Multiple Output (MIMO) system in invariant and variant fading channel conditions is proposed and analized. Based on the Extended Alamouti Quasi-orthogonal Space-Time Block Coding (Q-STBC), the use of the so-called Alternative Matrices for transmission, depending on the Channel State Information (CSI) received as feedback, is compared to other existing solutions.

Sign changes and permutations in the retransmission sequences allow reducing the interference while exploiting the spatial diversity to introduce some gain in the signal power. The best transmission order is selected by the Determinant Criterion, which …

New Hybrid Automatic Repeat Request (Harq) Combining With Space Time Block Coding (Stbc) In Invariant And Variant Fading Channel, Guillem Ernest Malagarriga Vilella

Theses

Hybrid Automatic Repeat reQuest (HARQ) combining for space time block coding Multiple Input Multiple Output (MIMO) system consisting of N transmit antennas and M receive antennas (NxM) in a time invariant channel and in a time varying channel is proposed and its performance is analyzed. Based on the measured channel matrix, the receiver chooses the best retransmission order and communicated it to the transmitter.

The scheme, where this algorithm is performed, is a combination of a precombining HARQ scheme and the Multiple Alamouti Space-Time Block Coding (MASTBC), which is suitable for more than 2 transmit antennas. The combination of this …

Online Clustering With Single-Pass Topology Based Fuzzy Clustering Algorithm, Abhishek Jaiantilal

Theses

Online clustering is of significant interest for real-time data analysis. Generic offline clustering methods such as K-Means, C-Means and others are computationally expensive. The computational burden of these methods increases non-linearly with the size of the data set. In addition these methods usually require a good amount of supervised knowledge yielding a non-unique solution. For real-time data analysis, there is an important tradeoff between accuracy and computational efficiency. An unsupervised one-pass clustering method that efficiently adapts to data distribution and evaluation is proposed. This method, Topology-Based Fuzzy Clustering (TFC), uses the topology of data to discover clusters. TFC uses the …

An Analytical Model Of Mac Protocol Dependant Power Consumption In Multi-Hop Ad Hoc Wireless Sensor Networks, Komlan Egoh

Theses

Power efficiency is the most constraining requirement for viable operation of battery-powered networked sensors. Conventionally, dynamic power management (DPM) is used to put sensor nodes into different states such as active, idle, and sleep, each consuming a certain level of power. Within the active state, communication operational states, such as receive and transmit consume different levels of nodal power. This thesis shows how DPM states and protocol operational states can be combined into a single stochastic model to finely evaluate the power consumption performance of a medium access control (MAC) protocol. The model is formulated as a semi-Markov decision process …

Evaluation Of Coding Scheme For Mimo Radar, Suresh Kumar Harikrishnan

Theses

Multiple Input Multiple Output (MIMO) antenna systems have shown a great potential for wireless communication. These systems support high capacity, increased diversity and interference suppression. Recently it has been proposed MIMO constellations for Radar. MIMO Radar is not only a new research field, but also a very promising approach in terms of overcoming Radar Cross Section (RCS) fluctuations with diversity. This thesis explores the potential of coding schemes for MIMO Radar.

The ambiguity functions measures related to MIMO Radar are used to evaluate how much diversity gain can be coherently achieved with certain coding schemes. The results of this analysis …