Carleton University Institutional Repository
Not a member yet
20816 research outputs found
Sort by
Toward Data-Driven Spectrum Management: Machine Learning Techniques for Localized Spectrum Demand Estimation
No full textWith the expansion of 5G and the development of 6G networks, local mobile spectrum demand is expected to grow significantly. In response, spectrum regulators seek to better understand current demand to ensure spectrum-related decisions maximize the socioeconomic benefits of this finite resource and continue to foster innovation within the wireless industry. This thesis presents a data-driven approach to estimate localized mobile spectrum demand for regulatory applications. A new demand proxy, derived from crowdsourced measurements and validated using proprietary traffic data, is proposed to address limitations of traditional proxies. Next, spectrum demand modeling is framed as a regression task, various classical machine learning models are explored with geospatial data used as input features, and an interpretability technique is applied to demonstrate how these models can inform regulatory decision-making. Finally, advanced deep learning models are designed to improve performance and transfer learning is leveraged to showcase their applicability across diverse regulatory scenarios
Automated Fine-tuning CNN Using Firefly Algorithm for Bearing Fault Diagnostics
Full text linkAutomated fine-tuning of Convolutional Neural Networks (CNNs) is essential for improving diagnostic accuracy in bearing fault detection. Traditional methods often require manual tuning of hyperparameters, or exhaustively searches through all combinations of hyperparameters, which can be time-consuming and suboptimal, especially in complex fault scenarios. In this work, a novel approach is presented that integrates the Firefly Algorithm (FA) with CNNs to automate the fine-tuning process, optimizing key hyperparameters such as batch size, units, epochs and learning rates. The Firefly Algorithm, inspired by the natural behavior of fireflies, excels in exploring the search space for global optima, making it well-suited for optimizing CNN architectures. Applied to MFPT data, the proposed method demonstrates extraordinary adaptability in various of CNN models and also presented improvements in test accuracy and computational efficiency comparing to main-stream automated finetuning approaches. This framework provides a scalable solution for deploying CNN-based diagnostic systems across various industrial applications
Enhancing Privacy in Peer-to-Peer Energy-Sharing Networks Using a Compliance Management Platform
No full textEnergy-sharing systems utilize decentralized infrastructures and peer-to-peer networks to reduce transmission and distribution losses of renewable energy. However, the personal data of prosumers, acting as both energy consumers and producers, may be shared with third parties unknowingly. Despite using tamper-proof technologies to enhance traceability, secure management of personal data remains inadequate. This includes safe data storage and transfer and the lack of control over data processing. Moreover, privacy regulations mandate compliance for all platforms to protect personal data and provide user control over their data. This thesis introduces a Compliance Management Platform (CMP) to secure prosumers' data that improves transparency by automatically recording data access. The proposed CMP complies with PIPEDA, the primary privacy regulation in Canada, by addressing its principles and offering automatic verification mechanisms to evaluate the systems' compliance. The performance of the CMP is evaluated through multiple prototype implementations tested in a simulated environment, demonstrating its feasibility
MEASURING W BOSON DRELL-YAN ANGULAR COEFFICIENTS AND DIFFERENTIAL CROSS-SECTION WITH THE ATLAS DETECTOR USING A LOW-PILEUP DATA SET
No full textThe W and Z bosons, mediating the fundamental weak force, are some of the particles produced from the proton--proton collisions in the Large Hadron Collider (LHC), located at CERN. A proton--proton interaction producing a W boson that decays immediately into leptonic particles, such as an electron and a neutrino, is known as the Drell--Yan process. The differential cross-section of this process expresses how likely it is to occur depending on the kinematics of the W boson -- such as the transverse momentum, pₜᵂ -- and the kinematics of the leptons. This differential cross-section can be decomposed into eight spin--related ratios and an unpolarized term. The eight spin-related ratios are known as the Drell-Yan angular coefficients. The angular coefficients and the unpolarized term are coupled to angular polynomials of cosθ and ϕ, which describe the angles between the leptons and the W boson polarization. By measuring the angular distributions of the decay products, the angular coefficients that describe the production of the W boson can be extracted. These angular coefficients are important as they are a direct probe of perturbative Quantum Chromodynamics production of W bosons from proton--proton interactions, making these coefficients an important input to other precision measurements, such as the W boson mass. This thesis presents a measurement of the four dominant angular coefficients -- A₀, A₂, A₃, and A₄ -- and the production differential cross-section as a function of pₜᵂ by reconstructing W boson events to measure the cosθ and ϕ distributions. The measurements were made possible by utilizing a data set of special runs with fewer interactions per bunch crossing, measured with the ATLAS detector at a centre-of-mass energy √s = 13 TeV. Included are the first measurements at the LHC of the W boson coefficients A₂ and A₃ as a function of pₜᵂ and the first measurements ever of A₀ and A₄ as function of pₜᵂ
Enhanced FSO (Free Space Optical) Data Backhaul for Satellite IoT Networks in the Presence of Adverse Weather Conditions
Full text linkFor decades, satellites have facilitated remote Internet of Things (IoT) services. However, the recent proliferation of increasingly capable and numerous sensors has led to a substantial growth in the volume of data. But legacy Radio frequency-based systems have limited capacity. Hence, free space optical (FSO) communication systems have been proposed, as they allow for high data rates. However, FSO communications are vulnerable to adverse weather. We investigated the potential benefits of using high-altitude platform systems (HAPS) to improve the performance of these networks. In addition, we also developed strategies that use predicted weather conditions to improve the HAPS-enabled networks' performance. Subsequently, we developed a reinforcement learning-based (RL) solution to improve the energy efficiency of satellite-to-ground FSO downlink operations in the presence of adverse weather. We compared the RL solution to a set of simple threshold solutions and found notable performance increases for some network configurations
Terrestrial Locomotion of Bats: Associations of Gait Characteristics with Body Size and Roosting Behaviour
Full text linkSome bat species exhibit morphologies and behaviours that are specialized for terrestrial locomotion. Many species’ terrestrial abilities have not been evaluated, despite the importance of locomotion in bats’ survival when grounded or when using terrestrial resources. I conducted treadmill-facilitated kinematics trials to quantify and compare the terrestrial gait patterns of eastern small-footed bats (Myotis leibii), little brown bats (Myotis lucifugus), big brown bats (Eptesicus fuscus), and tricolored bats (Perimyotis subflavus); four species with differing roosting preferences. Species’ gait patterns reflected their terrestrial resource use, with the two ground-roosting Myotis species exhibiting higher stride frequencies and velocities than species that prefer raised roosts, and similar gait patterns and normalized velocities as walking common vampire bats (Desmodus rotundus) and New Zealand short-tailed bats (Mystacina tuberculata). My comparative approach expands our understanding of variation in the terrestrial locomotive behaviours of bats and demonstrates the utility of standardized quantitative analyses for interspecies comparisons
Cross Domain Model Adaptation and Generalization
No full textDeep learning models often experience performance degradation when applied to data that differs from their training distribution. This thesis addresses the challenge of data distribution discrepancies through two main aspects: adapting source models to target domains without data annotation and improving model generalization to unseen domains. For source-free domain adaptation (SFDA), where source data is inaccessible, this thesis proposes two innovative methods to address key challenges. The first method generates labeled surrogate source training data by optimizing inputs while keeping the source model fixed. Gradient-based global fitting constraints are introduced to ensure the surrogate data accurately reconstructs the complete source distribution. These surrogate data can then be utilized by existing unsupervised domain adaptation methods. The second method focuses on source-free domain adaptation under the inductive setting. A semi-supervised fine-tuning approach is introduced, which partitions the unlabeled target training set into a confident pseudo-labeled subset and a less-confident unlabeled subset based on prediction confidence from the source model. A moving-average prototypical classifier updates soft labels for the unlabeled subset, enabling incremental adaptation of the source model to the target domain. Complementary to SFDA, domain generalization focuses on training models that are capable of generalizing to unseen testing domains. This thesis introduces two methods to address standard and federated domain generalization. The first method employs a two-stage training approach for standard domain generalization that simulates an approximate meta-generalization scenario and incorporates a self-adaptation module to adjust pretrained meta-source models to the meta-target domains. The core concept of self-adaptation involves leveraging contextual information as domain knowledge to automatically adapt a model trained in the first stage to new contexts in the second stage. The second method tackles federated domain generalization, where isolated client data is used to train a unified and generalizable model. The proposed approach incorporates local and global flatness regularizations to avoid sharp minima and encourage convergence to the global optimum. These regularizations leverage adversarial parameter perturbations, with two perturbation methods proposed at the level of weights and singular values. All proposed methods are evaluated on standard benchmarks, demonstrating their effectiveness in addressing data distribution discrepancies in deep learning
Numerical Analysis of Electrical Parameter Effects on a Nanosecond Pulsed Dielectric Barrier Discharge Actuator
Full text linkDielectric barrier discharge (DBD) plasma actuators are active flow control devices that have shown promising capabilities of enhancing aerodynamic performance. To study the underling mechanisms occurring within the plasma produced by a nanosecond pulsed DBD actuator and the effect the input electrical parameters have on plasma generation, a numerical model has been developed. Because this work is focused on the plasma behaviour, the impact on the bulk flow has been neglected. The plasma properties analysed here include the charged particle densities, surface charge along the dielectric, and the ion sheath propagation. As the applied voltage amplitude and pulse duration was increased, a larger plasma region formed due to the electric field expanding further throughout the domain. The results highlight a significant relationship between the residual surface charge on the dielectric and the propagation of the ion sheath
Knowledge, Power, and Migration : Contesting the North/South Divide
No full textAs the field of migration studies has grown, the asymmetrical relationship between researchers in the Global North and in the South has produced a body of work that centres the concerns of the former. Those from the Global North and wealthier countries continue to produce the greater portion of this research, while research from Global South scholars with lived experiences as migrants is received as anecdotal or too niche to have universal application. Knowledge, Power, and Migration assembles researchers from across the divide to question the ways in which research practices can change the conversation on immigration. It encourages a necessary curiosity about how scholarship in the field can shape global, social, and epistemic justice. Migration is a constant in human history, but the sharp decline in permanent resettlement options, increasingly selective criteria, and violent enforcement measures of the twenty-first century constitute a crisis of immigration policy. Only by redressing the inequalities it shares with global governance structures can the discipline confront this historic challenge. Research on immigration can occasion reflections and practices that challenge epistemic injustices. Knowledge, Power, and Migration contributes to this ongoing project while offering insights on the practical organization of new forms of dialogue on migration in a largely unequal world.Free access to this e-book is available to readers, scholars, and students located in the Global South whose institutions lack the resources to purchase access to these books as well as to those in other regions who are part of non-profit or community organizations concerned with displacement and who lack alternate forms of access to the book or the resources needed to purchase these publications. Please see full access conditions below
A Critical Discourse Analysis of Ontario’s 2019 Sex Education Curriculum’s Risk-Centred Framework’s Impact on Students’ Access to Contraceptives
Full text linkThis project analyzed Ontario’s 2019 Health and Physical Education (HPE) curriculum for grades seven and eight to assess how its risk-centered framework impacts the effectiveness of sex education in facilitating students’ access to contraception. The curriculum and interview transcript data were triangulated and analyzed with a critical discourse analysis (CDA). The CDA examined how the risk-centered framework is operationalized within the curriculum and if it is reinforced through teachers’ accounts of the hidden curriculum. The CDA also analyzed how the curriculum’s risk-centered framework impacts the information and resources student receive to support their access to contraception. This project used a critical feminist analysis to determine the impact of the risk-centered framework on contraception access and the health of people assigned female at birth (AFAB). This study found that the curriculum’s risk-centered framework restricts the information and resources students receive to facilitate their access to contraception, which perpetuates gender health inequities