Please use this identifier to cite or link to this item:
Title: Intelligent resource allocation for joint radar-communication
Authors: Lee, Joash
Keywords: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Engineering::Electrical and electronic engineering::Wireless communication systems
Engineering::Mechanical engineering::Motor vehicles
Issue Date: 2022
Publisher: Nanyang Technological University
Source: Lee, J. (2022). Intelligent resource allocation for joint radar-communication. Doctoral thesis, Nanyang Technological University, Singapore.
Abstract: Autonomous vehicles (AVs) produce high data rates of sensory information from sensing systems. In a cooperative driving setting, efficient sharing and processing of this sensory information has the potential to achieve collective sensor fusion, among other advanced functionalities that can enhance safe and efficient mobility. For such sharing to occur, high data-rate communication becomes essential. Unfortunately, the forthcoming proliferation of faster mmWave communication networks has raised concerns on interference with automotive radar sensors, which has led to a body of research on Joint Radar-Communication (JRC). Current strategies for JRC often rely on specialised hardware, prior knowledge of the system model, and entail diminished capability in either radar or communication functions. This thesis focuses on how wireless resources for JRC can be intelligently allocated through coordinated signals transmissions, with the ultimate aim of supporting vehicle safety while respecting limits on the availability of wireless resources. We formulate frameworks on how AVs might intelligently schedule JRC, then identify contemporary techniques in deep reinforcement learning to allow intelligent vehicles to learn, in a distributed manner, how best to conduct JRC. In formulating frameworks for intelligent scheduling of JRC, we consider the question of how radar sensing and communication can best support road safety. Based on a review of contemporary literature on automotive sensing and a modern understanding of road safety models, we propose a metric for radar performance that incentivises more frequent sensing under risky environmental conditions for which radar excels. We also propose options for communications performance metrics that go beyond the traditional quantities of throughput signal-to-interference-plus-noise ratio (SINR). The first option considers the Age of Information (AoI), defined as the duration between data generation and its receipt, in conjunction with data classes that are organised by urgency level. The second option takes a more generalised view of data classes by considering instead what we term as the spatial signature of data: the positional coordinates for which the data contains information on, and how useful this information is to the receiving vehicle's safety. Next, we identify Deep Reinforcement Learning (DRL) algorithms to solve our JRC problem frameworks, with the central consideration being: how can vehicles in a vehicular network learn, in a distributed manner, to coordinate wireless resources? For vehicle networks with multiple intelligent vehicles, we introduce a Graph Neural Network (GNN) framework via a control channel. The key advantage of this framework is that it permits modular combinations of algorithmic extensions, thus allowing for a fair comparison of different combinations for our JRC problem. We also investigate other non-graphical algorithmic features: the use of a Convolutional Neural Network (CNN) architecture to extract features from raw representations of the data queue, and incorporating an entropy bonus to encourage varied solutions. Our results reveal several findings. Firstly, experiments show that DRL achieves superior performance to benchmark algorithms in our JRC problems, even with minimal prior knowledge about the environment. In a multi-agent setting, learning how to coordinate access to the shared wireless channel via the GNN framework further improves both learning speed and overall performance. Additional algorithmic features found to improve JRC performance include the use of CNNs and the inclusion of an entropy bonus in the DRL objective function.
DOI: 10.32657/10356/160990
DOI (Related Dataset): 10.5281/zenodo.6979385
Schools: Interdisciplinary Graduate School (IGS) 
Research Centres: Energy Research Institute @ NTU (ERI@N) 
Rights: This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:IGS Theses

Files in This Item:
File Description SizeFormat 
Joash Thesis 10 Aug 2022_.pdf15.25 MBAdobe PDFThumbnail

Page view(s)

Updated on Sep 30, 2023

Download(s) 50

Updated on Sep 30, 2023

Google ScholarTM




Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.