Projects

Intelligent Edge (iEDGE) Middleware: iEDGE, by design, will incorporate heterogeneous devices over heterogeneous communication platforms.  Seamless integration of such devices through middleware ensuring performance, reliability, security, and other constraints is a challenging endeavor.  The objective of this project is to research a middleware that can (a) respond quickly to the workload (transactional (Online Transaction Processing (OLTP)), streaming, machine learning (ML), or security-related (authentication, access control, etc.)) requirements and (b) be easily deployable.

Environment-Aware Mobile Intelligent Edge (iEDGE): Mobile edge nodes (e.g., UAVs) often operate in clear, sunny conditions. The graphics processing unit (GPU) temperature can rise significantly when executing compute-intensive tasks, reducing computational capacity and increasing latency. The maximum computing capacity of mobile edge devices is not constant and is greatly affected by environmental factors, such as temperature and wind speed. Enhanced ecological awareness in edge computing will improve the computing capacity of mobile edges in dynamic environments. This project aims to evaluate how environmental factors (e.g., temperature and wind speed) affect the computing capacity of mobile edge nodes and to design an environment-aware iEDGE that will dynamically adjust the computing power of nodes according to the environment to minimize inference latency.

Sparse Intelligent Edge (iEDGE) Computing: UAV and ground platform-based iEDGE have been increasingly utilized in energy-constrained applications, such as precision agriculture and surveillance. Multiple sets of data from radio frequency (RF), multispectral, and platform control sensors are collected and need to be jointly processed to infer target (e.g., soil moisture, temperature, chemistry) parameters. Energy efficiency and latency of deep learning-based models utilized for inference on resource-constrained iEDGE nodes are critical. Sparsity and attention-based design of deep neural networks (DNN) can improve latency and energy efficiency while preserving inference performance. This project evaluates how sparsity-based DNN approaches, attention, and knowledge distillation mechanisms with teacher-student networks affect the iEDGE latency, energy, and inference performance.

Intelligent Edge (iEDGE) Open Radio Access Network (O-RAN) Modeling, Codesign, and Management: Future low-latency communications must be processed at the edge. The implementation and dynamic deployment of coexisting network services must be intelligently managed based on the number of users, service type, etc. The O-RAN architecture is a promising solution that supports intelligent controllers. It defines available interfaces, enabling real-time processing and practical base station disaggregation across the antenna sites, edge resources, and the cloud. This project aims to code and analyze iEDGE O-RAN deployments to enable research on the interplay between edge intelligence and next-generation RAN intelligence.

Device Sampling in Intelligent Edge (iEDGE): Within iEDGE, an IoT cluster may consist of various mobile devices, such as cellular phones, sensors, vehicles, etc., each with heterogeneous computing and transmitting capabilities. These heterogeneous devices need to share a limited radio resource (spectrum) when conducting the processes of federated learning (FL) and sending their local learning results to an edge server. The performance of FL highly depends on how the local learning results can be efficiently transmitted using limited radio resources. The project aims to investigate how to efficiently sample iEDGE devices to share local FL results with limited radio resources.

Electromagnetic Pulse-Protection for Intelligent Edge (iEDGE): The iEDGE may incorporate many high-sensitivity sensory nodes/systems. Intense electromagnetic pulses (e.g., lasers) may induce a strong current on the iEDGE sensory system antennae, causing a security threat.  This project aims to research and develop a low-cost, highly efficient electromagnetic pulse protection system to protect the iEDGE sensory systems from cyberattacks.

Simulation-enhanced Testbed Development for Intelligent Edge (iEDGE): Autonomous ground vehicle (AGV)-based iEDGE have been increasingly utilized in unstructured off-road scenarios (navigation, logistics, and infrastructure-less communications). However, environmental factors such as vegetation density and soil conditions affect the AGV’s understanding of traversability. While deep neural networks (DNNs) can provide high-quality results, testing in all situations is impossible. Consequently, utilizing simulations is critical. The Mississippi State Autonomous Vehicle Simulator (MAVS) generates high-fidelity Light Detection and Ranging (LIDAR) and camera imagery. It allows users to control environmental aspects such as rain, fog, dust, etc.  As such, MAVS can be used to provide realistic training data to enhance DNN performance. This project evaluates how simulations can enhance DNN estimation of vegetation density and soil estimation in unstructured environments to assist with AGV-based iEDGE.