Uplink Access Scheme Design and Machine Learning based Optimization for Industrial Internet of Things

Student thesis: Doctoral ThesisDoctor of Philosophy

Abstract

Industrial Internet of Things (IIoT) revolutionizes the future manufacturing facilities by integrating the Internet of Things technologies into industrial settings. With the deployment of massive IIoT devices, it is difficult for the traditional downlink traffic-driven wireless network to support the ubiquitous uplink connections with diverse quality-of-service (QoS) requirements. Therefore, the main focus of this dissertation is on the analysis, design and optimization of uplink access schemes for IIoT. In particular, this dissertation presents the analysis of novel random access (RA) schemes over the licensed spectrum, and the design and optimization of novel access schemes over the unlicensed spectrum, respectively.

First, a cellular-connected UAV testbed is developed and evaluated to provide insights for the general cellular-connected IIoT system design, and the rest research of this dissertation. Based on the OpenAirInterface (OAI), a cellular-connected UAV testbed with uplink video transmission and downlink control&command (C&C) transmission is developed, where the End-to-End (E2E) delay, reliability of both uplink video and downlink C&C transmissions, along with throughput of uplink video transmission are evaluated. More importantly, the algorithms developed later can be implemented based on the testbed in the future for performance
evaluation.

Second, inspired by the C&C transmission experimental results in cellularconnected UAV testbed, a spatio-temporal analytical framework is derived to evaluate the potential RA schemes for uplink small data transmission (SDT) in IIoT. Based on stochastic geometry, the reliability, throughput, and energy consumption for uplink SDT over the licensed spectrum are derived.

Third, inspired by the video transmission experimental results in cellularconnected UAV testbed, a novel Cat4 Listen-before-Talk (LBT) access scheme over unlicensed spectrum is proposed. Based on the existing LBT access scheme, instantaneous interference level quantification, instantaneous interference level sharing, and Backoff speed determination are proposed to enhance the uplink throughput over unlicensed spectrum.

Fourth, deep reinforcement learning (DRL) algorithms are developed to further optimize the uplink throughput over unlicensed spectrum. Based on double Deep QNetwork (DDQN), centralized and federated algorithms are proposed to maximize the uplink throughput over unlicensed spectrum via optimizing the energy detection (ED) thresholds.



Date of Award1 Dec 2022
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorAbdol-Hamid Aghvami (Supervisor) & Yansha Deng (Supervisor)

Cite this

'