Edge Intelligence: 5G Integrated Testbed

Implementation work in “Wireless and Mobile Network Laboratory”, 2020

In this collaobrative project, I’m in charge of implementing the edge computing system and addressing related API of each component in our testbed.

System overview:

Deep Learning based Joint Optimization of UAV position and Resource Allocation in Vehicular Network

Research work in “Wireless and Mobile Network Laboratory”, 2020

In this work, I developed a 2-stage method to jointly optimize UAV deployment and resource allocation for vehicular network. By modeling it as a multi-task learning problem, I implemented a DNN that can dynamically tune its weights to adapt to wireless environment and can jointly determine offloading decision, allocation of spectrum and computing resources for vehicular tasks in real time.

Demo of key concept:

Collaborative Social-Aware Video Caching in Edge Network

Research work in “Wireless and Mobile Network Laboratory”, 2019

In this work, I first adopt a time series model to quantify users’ interaction patterns, and then mathematically formulated information dissemination among users in the online social network. Afterward, I developed a intact video caching and adaptation framework that utilizes users’ viewing history, users’ channel condition and video dissemination state in the community to determine 1) collaborative caching decision among multiple edge servers, and 2) real-time video transcoding decision with an aim to optimize users’ QoE.

Demo of key concept:

In the figure, the black line means the message exchange between users and the green/red nodes represent the users have/haven’t seen a certain video. All of the user data are collected in real-world traces.

MEC-Assisted FoV-Aware Adaptive 360° Video Streaming for VR

Side project, 2020

In this work, I enhanced QoE and bandwidth utilization of 360° video streaming service by adaptively determining encoding bitrate of each tile based on prediction of user’s Field of View (FoV). In addition, edge servers are considered to provide caching and computing capacity to reduce the experienced latency of VR devices.

Demo of key concept:

The left figure shows the actual/predicted FoV of a user during watching a 360° video. The right figure is the resolution decision made by the proposed scheme in a 360 video frame.

Analysis of D2D Caching Schemes in Heterogeneous Network

Course project in “Introduction to Wireless and Mobile Network”, 2019

In this work, by modeling user mobility, inter-cell interference from D2D and BS2D communication, and wireless fading channels, I implement the simulation for a few D2D caching schemes, where each device can prefetch popular contents from the base stations (BS) and share them with proximate peers.

Demo of key concept:

The red/blue lines represent BS2D and D2D transmission respectively. Also, the flash green point denotes a content request, obeying the Poisson process.

FPGA High Level Synthesis: A Survey and Implementation on Task Scheduling Algorithms

Course project in “Introduction to Electronic Design Automation”, 2020

In this work, I studied and compared state-of-the-art researches in resource-constrained scheduling algorithms, which aims to minimize total completion time of dependant tasks.

Demo of key concept:

The right figure is the task topology of the adopted benchmark, where the arrows in the directed graph denote dependencies between tasks. The left figure shows the scheduling decision of the implemented Deep Reinforcement Learning (DRL) model. Green/blue nodes signify unscheduled/finished tasks respectively.

Parallel Task Offloading for Augmented Realty in Dynamic MEC Slice

Research work in “Wireless and Mobile Network Laboratory”, 2020

This is a simple AR application I made for demonstrating our 5G testbed. The computation tasks of AR apps (e.g. object detection, object projection and rendering) are offloaded to our proximate edge servers for execution. I also realize a dynamic network slicing system that can adjust Docker-based isolated resources based on predicted workload.

Demo:

Interest-based QoE-Driven Video Caching and Adaptation in Edge Network

Research work in “Wireless and Mobile Network Laboratory”, 2019

In this work, I developed an interest-based QoE-driven partial video caching scheme in the edge network, where the caching decisions are made depended on users’ viewing history and current backhaul bandwidth. In addition, edge servers are deployed to perform real-time transcoding on cached videos to serve users with time-varying downlink capacity.

Performance Demo:

The experiment are conducted based on real-world datasets (i.e. Youtube trace collected in UMass campus).

Traffic Simulator for Public Transportation System of Taipei

Side project, 2015

Demo: