Improving the planning and deployment of wireless networks via the application of deep learning

Abstract

Fifth-generation (5G) and beyond (B5G) wireless communication networks are poised to transform the landscape of modern communication systems. In addition to providing conventional services offered by legacy wireless networks, 5G/B5G systems are expected to support an increasingly diverse set of applications, handling significantly larger volumes of mobile traffic with varying requirements in terms of bandwidth, latency, throughput, and reliability. As the rise of smart cities and the Internet of Everything (IoE) unfolds, wireless connectivity will play a central role in facilitating intelligent urban infrastructure. However, as wireless networks become more complex, the allocation of resources within these systems remains a key challenge, particularly in energy-constrained environments. In this context, effective planning and optimization of wireless networks are crucial for avoiding resource under-utilization, lowering operational costs, and improving overall system performance. Addressing these challenges, this thesis investigates the development of advanced tools for radio propagation modeling and resource allocation in next-generation wireless networks. It introduces an enhanced empirical model that utilizes extracted path loss components to expedite outdoor path loss predictions. Additionally, it presents the Path Loss Prediction Network (PPNet), a novel deep learning framework based on SegNet architecture aimed at refining radio propagation predictions. A generative model for radio map interpolation is also proposed to assist in path loss prediction when measurements are limited. Furthermore, the thesis introduces an innovative optimizer based on a large language model for access point (AP) placement. Our work significantly enhances both the accuracy and efficiency of path loss predictions in complex indoor and outdoor environments while optimizing AP deployment. Aiming at tackling the open problems in the wireless communication field, this thesis investigates the integration of artificial intelligence (AI) and machine learning (ML) techniques to develop robust, data-driven methods for network planning and radio map interpolation. These methods not only provide more accurate resource allocation strategies but also streamline the deployment of wireless networks in diverse urban landscapes. Ultimately, the findings of this research contribute to the advancement of intelligent, scalable, and energy-efficient wireless communication systems that are essential for the growth of 5G/B5G technologies and beyond.