Task-oriented joint design of communication and computing for Internet of Skills

Kizilkaya, Burak (2023) Task-oriented joint design of communication and computing for Internet of Skills. PhD thesis, University of Glasgow.

Abstract

Nowadays, the internet is taking a revolutionary step forward, which is known as Internet of Skills. The Internet of Skills is a concept that refers to a network of sensors, actuators, and machines that enable knowledge, skills, and expertise delivery between people and machines, regardless of their geographical locations. This concept allows an immersive remote operation and access to expertise through virtual and augmented reality, haptic communications, robotics, and other cutting-edge technologies with various applications, including remote surgery and diagnosis in healthcare, remote laboratory and training in education, remote driving in transportation, and advanced manufacturing in Industry 4.0.

In this thesis, we investigate three fundamental communication requirements of Internet of Skills applications, namely ultra-low latency, ultra-high reliability, and wireless resource utilization efficiency. Although 5G communications provide cutting-edge solutions for achieving ultra-low latency and ultra-high reliability with good resource utilization efficiency, meeting these requirements is difficult, particularly in long-distance communications where the distance between source and destination is more than 300 km, considering delays and reliability issues in networking components as well as physical limits of the speed of light. Furthermore, resource utilization efficiency must be improved further to accommodate the rapidly increasing number of mobile devices. Therefore, new design techniques that take into account both communication and computing systems with the task-oriented approach are urgently needed to satisfy conflicting latency and reliability requirements while improving resource utilization efficiency.

First, we design and implement a 5G-based teleoperation prototype for Internet of Skills applications. We presented two emerging Internet of Skills use cases in healthcare and education. We conducted extensive experiments evaluating local and long-distance communication latency and reliability to gain insights into the current capabilities and limitations. From our local experiments in laboratory environment where both operator and robot in the same room, we observed that communication latency is around 15 ms with a 99.9% packet reception rate (communication reliability). However, communication latency increases up to 2 seconds in long-distance scenarios (between the UK and China), while it is around 50-300 ms within the UK experiments. In addition, our observations revealed that communication reliability and overall system performance do not exhibit a direct correlation. Instead, the number of consecutive packet drops emerged as the decisive factor influencing the overall system performance and user quality of experience. In light of these findings, we proposed a two-way timeout approach. We discarded stale packets to mitigate waiting times effectively and, in turn, reduce the latency. Nevertheless, we observed that the proposed approach reduced latency at the expense of reliability, thus verifying the challenge of the conflicting latency and reliability requirements.

Next, we propose a task-oriented prediction and communication co-design framework to meet conflicting latency and reliability requirements. The proposed framework demonstrates the task-oriented joint design of communication and computing systems, where we considered packet losses in communications and prediction errors in prediction algorithms to derive the upper bound for overall system reliability. We revealed the tradeoff between overall system reliability and resource utilization efficiency, where we consider 5G NR as an example communication system. The proposed framework is evaluated with real-data samples and generated synthetic data samples. From the results, the proposed framework achieves better latency and reliability tradeoff with a 77.80% resource utilization efficiency improvement compared to a task-agnostic benchmark. In addition, we demonstrate that deploying a predictor at the receiver side achieves better overall reliability compared to a system that predictor at the transmitter.

Finally, we propose an intelligent mode-switching framework to address the resource utilization challenge. We jointly design the communication, user intention recognition, and modeswitching systems to reduce communication load subject to joint task completion probability. We reveal the tradeoff between task prediction accuracy and task observation length, showing that higher prediction accuracy can be achieved when the task observation length increases. The proposed framework achieves more than 90% task prediction accuracy with 60% observation length. We train a DRL agent with real-world data from our teleoperation prototype for modeswitching between teleoperation and autonomous modes. Our results show that the proposed framework achieves up to 50% communication load reduction with similar task completion probability compared to conventional teleoperation.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science; T Technology > T Technology (General)
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Zhao, Professor Guodong, Imran, Professor Muhammad Ali and Heidari, Professor Hadi
• Date of Award: 2023
• Depositing User: Theses Team
• Unique ID: glathesis:2023-83742
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 25 Jul 2023 13:48
• Last Modified: 25 Jul 2023 16:00
• Thesis DOI: 10.5525/gla.thesis.83742
• URI: https://theses.gla.ac.uk/id/eprint/83742

Related URLs

• Article DOI
• Article DOI
• Enlighten Publications Record
• Article DOI
• Article DOI
• Enlighten Publications Record
• Article DOI
• Article DOI
• Data DOI