2026-09-18T18:47:04Zhttps://api.repository.cam.ac.uk/server/oai/request

oai:www.repository.cam.ac.uk:1810/3758252024-11-09T01:43:15Zcom_1810_219481com_1810_256065col_1810_219482

Applications of Artificial Intelligence to Electromagnetism and Indoor Wireless Networks https://doi.org/10.17863/CAM.113346 Bakirtzis, Stefanos Sotirios Wassell, Ian Zhang, Jie Fifth-generation (5G) and beyond (B5G) mobile networks are anticipated to revolutionise the structure of the wireless ecosystem and support a set of highly heterogeneous applications, handing a substantially larger volume of mobile traffic with diverse requirements in terms of bandwidth, throughput, latency, quality of service and reliability. Remarkably, although the preponderance of this traffic is generated in indoor environments, the largest part of the legacy communication systems radio access network is deployed outdoors. As a result, next-generation networks are anticipated to align with the emerging need to serve indoor users, and therefore, major vendors and mobile network operators expect indoor cellular networks (ICNs) to be extensively deployed in 5G/B5G systems. Consequently, the installation of new components within the heart of the radio access network, in conjunction with the increasing network complexity and user requirements, calls for expedient electromagnetic wave propagation modelling tools and optimisation frameworks that can assist the design of ICNs, ensuring the orchestrated and prosperous operation of the wireless ecosystem. Furthermore, to enable their optimal functionality it is necessary to comprehend the characteristics of the mobile service demands generated by ICNs. Aiming at tackling these open problems, this thesis probes how artificial intelligence (AI) can be leveraged in the context of computational electromagnetics and indoor wireless networking. To this end, first, an innovative solution to Maxwell’s equations through graph neural network (GNN) message passing is introduced by establishing a causal connection between GNNs and the finite-difference time-domain method. Consequently, an expedient and credible data-driven radio propagation modelling tool, named EM DeepRay, is developed by coupling a high-performance ray-tracer with a convolutional encoder-decoder. Then, the computational efficiency and robustness of EM DeepRay are exploited to develop an AI-assisted indoor wireless network planning framework, and to assess the uncertainty in network’s performance. Finally, a characterisation of the mobile service usage across a nationwide ICN deployment is provided, indicating that ICNs inherently manifest a limited set of mobile application utilisation profiles, which are not present in conventional outdoor macro base stations. University of Cambridge 2024-07-30 Thesis Doctoral Doctor of Philosophy (PhD) eng European Commission, Horizon 2020 Framework], H2020-MSCA-ITN-2019, under Grant 860239, BANYAN. Onassis Foundation Foundation for Education and European Culture https://www.repository.cam.ac.uk/handle/1810/375825 https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/b9da415c-6448-4a40-8eca-301822777325/download 87eda9de84448d1f82354d60eee3eb5f https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/6c4c1451-8d2d-403a-a71a-428f8e575d1c/download 9a6d4c43311d9177817ecb66b596221b https://creativecommons.org/licenses/by-nc-nd/4.0/ Artificial Intellgience Deep Learning Electromagnetism Radio Propagation Traffic Analysis Wireless Communications