Description: This thesis will concentrate on integrating sophisticated energy models into a simulation framework designed for joint communication and edge computing load balancing in satellite networks. This includes modeling dynamic energy harvesting (solar panel efficiency based on orientation, eclipse periods) and detailed energy consumption (per-component power draw for CPU, radio, sensors, thermal effects). Building on this, the student will design and evaluate energy-aware adaptations to existing Hierarchical Reinforcement Learning (HRL) algorithms (e.g., MA-HRL) by modifying reward functions or state representations to optimize task allocation for extended satellite operational lifetime or peak performance under stringent energy budgets.
Expected Background: Strong Python programming skills, basic understanding of satellite power subsystems, optimization concepts. Familiarity with Machine Learning (specifically Reinforcement Learning) is a plus.
Thesis Methodologies:
Literature review (satellite energy modeling, energy-aware computing & RL).
Simulation framework enhancement (Python module development for energy dynamics).
Algorithm design (modifying HRL agents for energy awareness).
Extensive computer simulations to compare against energy-agnostic HRL approaches and heuristics like MCA.
Data analysis and visualization (Pandas, Matplotlib) of energy profiles, task success rates, and satellite lifetime.
Description: The student will implement advanced Inter-Satellite Link (ISL) models within a simulation framework for joint communication and edge computing load balancing in satellite networks. This includes incorporating factors like variable bandwidth/latency due to orbital mechanics, antenna pointing inaccuracies, atmospheric effects (if considering optical ISLs), signal interference, and link congestion. The core of the thesis will be to analyze the impact of these realistic ISL dynamics on the performance, convergence, and robustness of distributed MA-HRL algorithms. The student may propose and evaluate adaptive communication strategies or modifications to MA-HRL to mitigate the effects of unreliable links.
Expected Background: Strong Python programming skills, good understanding of computer networking principles (OSI layers, routing, link characteristics), wireless communication fundamentals. Familiarity with Reinforcement Learning is beneficial.
Thesis Methodologies:
Literature review (ISL characteristics, modeling techniques for space networks).
Simulation framework enhancement (Python module for advanced ISL dynamics).
Performance analysis of existing MA-HRL under the new, realistic link models.
Design and implementation of adaptive communication or robust MA-HRL strategies.
Computer simulations to quantify impact and effectiveness of proposed solutions.
Description: This thesis will design, implement, and simulate a federated learning (FL) system within a simulation framework for joint communication and edge computing load balancing in satellite networks, tailored for a specific Earth Observation (EO) task (e.g., cloud detection, simple feature extraction). Satellite nodes will collaboratively train a shared model using their locally acquired EO data snippets without transmitting raw data. Key challenges to address include communication efficiency (model updates vs. raw data), managing non-IID data distributions across satellites, and integrating FL cycles with resource allocation mechanisms (e.g., MA-HRL deciding when to initiate training based on energy and link availability).
Expected Background: Strong Python programming skills, solid understanding of Machine Learning (Federated Learning, Convolutional Neural Networks for image data), data processing. Basic knowledge of EO data is helpful.
Thesis Methodologies:
Literature review (Federated Learning, on-board AI for EO).
Design of an FL architecture suitable for the simulation framework (aggregation strategies, communication protocols).
Implementation of the FL system within the Python simulator.
Simulation using representative (synthetic or public) EO datasets.
Evaluation based on model accuracy, convergence speed, communication overhead, and impact on satellite resources.
Description: This thesis will focus on a specific application: real-time maritime surveillance using a simulated satellite constellation employing joint communication and edge computing load balancing. Tasks would involve processing simulated AIS data, EO/SAR snippets for vessel detection, and fusing this information at the edge. The student will adapt HRL agents to prioritize time-critical anomaly detection (e.g., "dark" vessels, illegal fishing) and optimize data dissemination to relevant (simulated) authorities, considering communication constraints and satellite resource availability.
Expected Background: Strong Python programming, basic understanding of remote sensing data (AIS, EO/SAR concepts), data fusion concepts, (optional) GIS.
Thesis Methodologies:
Detailed scenario modeling for maritime surveillance (vessel types, sensor data characteristics, alert priorities).
Adaptation/refinement of HRL agent reward functions and state spaces for the specific application.
Implementation within the simulation framework.
Computer simulations to evaluate key performance indicators like detection latency, accuracy of information fusion (simplified), and efficiency of alert dissemination.
Description: Moving beyond pure simulation, this thesis aims to develop a small-scale hardware-in-the-loop (HIL) testbed to validate core algorithms for joint communication and edge computing load balancing in satellite networks. This could involve using microcontrollers or single-board computers (e.g., Raspberry Pi, Jetson Nano) to represent satellite nodes. The student would port a simplified version of a load balancing algorithm (e.g., MCA or a lightweight SA-HRL agent) to these nodes and interface them with the existing Python simulator (which would simulate the space environment and ISLs). The focus is on demonstrating basic algorithm execution on real hardware and identifying challenges in hardware deployment.
Expected Background: Strong Python programming, C/C++ programming for embedded systems, experience with microcontrollers/SBCs, basic understanding of computer networks.
Thesis Methodologies:
Literature review (HIL simulation, embedded systems for space).
Design of the HIL testbed architecture.
Selection and setup of hardware components.
Development of software interfaces between hardware nodes and the Python simulator.
Porting and testing of selected algorithms on the hardware.
Comparative analysis of algorithm behavior in pure simulation vs. HIL testbed.
Description: This thesis will investigate and develop energy-efficient Federated Learning (FL) techniques for 6G-enabled vehicular networks, explicitly considering the role of aerial edge computing (UAVs). The student will design algorithms and protocols that minimize energy consumption at vehicular terminals (VTs) during the FL process, considering factors such as communication overhead, local computation, and UAV-assisted data aggregation. The research will also explore adaptive techniques where VTs dynamically adjust their FL participation based on their energy budget and channel conditions.
Expected Background: Strong Python programming skills, understanding of wireless communication principles, machine learning fundamentals (especially Federated Learning), and optimization concepts.
Thesis Methodologies:
Literature review on energy-efficient FL and UAV-assisted edge computing.
Development of an energy consumption model for VTs in FL.
Design and implementation of energy-aware FL algorithms.
Computer simulations to evaluate the proposed algorithms in various vehicular network scenarios.
Performance analysis using metrics like energy consumption, learning accuracy, and convergence time.
Description: This thesis will focus on designing and evaluating network slicing strategies to support Distributed Learning-as-a-Service (DLaaS) in 6G-enabled Internet of Vehicles (IoV). The student will investigate how network slicing can be used to allocate resources and prioritize different distributed learning methods (e.g., Federated Learning, Split Learning) based on the specific requirements of various IoV services (e.g., autonomous driving, infotainment). The thesis will also explore dynamic slice adaptation mechanisms to handle the changing demands of IoV applications and network conditions.
Expected Background: Strong Python programming skills, good understanding of computer networking concepts (especially network slicing and SDN), machine learning principles, and experience with network simulation tools.
Thesis Methodologies:
Literature review on network slicing in 6G and distributed learning for IoV.
Design of network slicing architectures for DLaaS in IoV.
Implementation of a simulation environment to model network slicing and distributed learning.
Evaluation of different slicing strategies using metrics like latency, throughput, resource utilization, and learning performance.
Analysis of the trade-offs between various slicing configurations and their impact on IoV services.
Description: This thesis will address the joint optimization of communication and Federated Learning (FL) in vehicular edge computing environments. The student will develop algorithms that concurrently optimize resource allocation (e.g., bandwidth, computing resources) and FL parameters (e.g., local training iterations, aggregation frequency) to improve the overall performance of vehicular applications. The research will consider the dynamic nature of vehicular networks and the diverse requirements of different vehicular services.
Expected Background: Strong Python programming skills, solid understanding of wireless communication, machine learning (Federated Learning), optimization theory, and experience with network simulation.
Thesis Methodologies:
Literature review on joint optimization of communication and FL.
Development of a system model for vehicular edge computing with FL.
Design of joint optimization algorithms.
Implementation of the proposed algorithms in a simulation environment.
Performance evaluation using metrics such as latency, throughput, energy efficiency, and learning accuracy.