Applied Deep Learning for Artificial Intelligence-enabled Wireless Communication

Applied Deep Learning for Artificial Intelligence-enabled Wireless Communication

Wireless communications systems are ubiquitous capabilities underlying the technologies which are transforming our society, business, government, and industry today. Ever evolving applications are driving the requirements for the design of such wireless communication systems. Demanding increased speed and volumes in data transfer, the reliability and availability, and important protections such as the privacy as well as the essential security of these communications systems. Future wireless communications systems must approach these design challenges by addressing these emerging requirements and look for opportunities to leverage emerging technologies to complement the conventional design methods. Exponential growth in computing hardware and processing capabilities have supported the application of machine learning, and in particular deep learning, to extract value from large scale data assets. Researchers have also recognised the potential for the application of deep learning to the data-driven design of wireless communications systems, complex channel environments and emerging applications. This doctoral research thesis develops artificial intelligence-enabled deep learning models and training algorithms to specifically focus on four key objectives in application to wireless communications systems design. These objectives are Synchronisation (Objective 1), Adaptation (Objective 2) and Over-the-air learning (Objectives 3 & 4). Synchronisation is an important signal processing step in the receiver that aims to correct the perturbed signals to retrieve the original message. This thesis proposes a new method of parameter estimation for Synchronisation supporting multiple modulations, including chaotic modulations for secure communications. This objective aims to maximise the data-rate and security of wireless communications by focusing on short random preamble sequences which severely limit the accuracy of conventional methods. Adaptation is a desirable property supporting the reliability and availability of wireless communications systems in changing channel conditions. In this objective, a custom deep learning architecture is developed under a multi-task learning framework to learn multiple code-rates and is demonstrated to produce gains under fading channel environments in comparison to conventional coding methods. Over-the-air learning, as part of objectives 3 & 4, makes a novel contribution in the area of adaptation while enabling the global optimisation of both transmitter and receiver for a true channel environment. This objective addresses the challenging tasks of complex training and modelling regimes found in rapidly evolving wireless communication systems. Therefore, the thesis proposes two novel methods simplifying the training procedure and deep learning models for Over-the-air learning and addresses the reliability and availability in changing channel conditions as well as security during the training procedure. The newly proposed methods in this doctoral thesis clearly demonstrate the success of the proposed approaches for simulation, generalisation, training techniques and custom deep learning architectures. The research project outcomes are useful for establishing practical pathways for future applications of artificial intelligence-enabled wireless communications systems.

File reproduced in accordance with the copyright policy of the publisher/author/creator.