5G-Advanced Network Optimization: AI/ML and Sensing-Driven Beam Management with Energy Optimization
Thesis event information
Date and time of the thesis defence
Place of the thesis defence
Tellus Backstage
Topic of the dissertation
5G-Advanced Network Optimization: AI/ML and Sensing-Driven Beam Management with Energy Optimization
Doctoral candidate
Master of Science (Technology) Dhanushka Nalin Jayaweera Rajapakshalage
Faculty and unit
University of Oulu Graduate School, Faculty of Information Technology and Electrical Engineering, Centre for Wireless Communications - Radio Technologies
Subject of study
Communications Engineering
Opponent
Professor Jyri Hämäläinen, Aalto University
Custos
Professor Nandana Rajatheva, University of Oulu
5G-Advanced Network Optimization: AI/ML and Sensing-Driven Beam Management with Energy Optimization
The fifth generation (5G) communication networks introduced high data rates, low latency, extensive connectivity, and large capacity compared to previous generations. Considering such advancements as a baseline, future networks are exploring massive multiple input multiple output (MIMO) technologies and emerging tools like artificial intelligence (AI), machine learning (ML), integrated sensing and ultra-dense network deployments, such as cell-free massive MIMO.
AI/ML is a potential candidate for reducing the beam measurement and reporting overhead, which helps to enhance the energy efficiency of user equipment (UEs) and the network capacity. However, existing studies in AI/ML-based beam management (BM) mainly focus on improving the beam prediction accuracy, but system-level performance assessment is also essential. This research evaluates AI/ML-based BM, integrating trained models into a system-level simulator to analyze performance in terms of spatial and time-domain beam prediction. Results show that UEs with strong channel conditions can reduce the measurement overhead by up to 75% without a substantial throughput loss. In contrast, cell-edge users experience limited reductions due to prediction inaccuracies.
For enhanced situational awareness in future networks, sensors such as cameras and LiDAR (Light Detection and Ranging) provide crucial information such as user and blockage position, allowing the network to operate proactively. The next part of the thesis presents a multi-LiDAR approach for precise user positioning alongside a method for predicting line-of-sight (LOS) transitions, anticipating blockages up to 400 ms in advance. Such information from a LiDAR is used for AI/ML model activation/deactivation in AI/ML-based BM to avoid the performance loss of cell-edge UEs due to inaccurate beam predictions.
On the other hand, cell-free massive MIMO systems use distributed access points (APs) to serve multiple UEs on shared frequency-time resources. However, serving all users via all the APs can lead to inefficient spectral and energy usage under low traffic. This research introduces algorithms for optimizing power in cell-free networks by switching off low-contribution APs and minimizing power usage according to spectral efficiency needs and UE mobility. These methods reduce energy costs, outperforming systems where all APs remain active.
AI/ML is a potential candidate for reducing the beam measurement and reporting overhead, which helps to enhance the energy efficiency of user equipment (UEs) and the network capacity. However, existing studies in AI/ML-based beam management (BM) mainly focus on improving the beam prediction accuracy, but system-level performance assessment is also essential. This research evaluates AI/ML-based BM, integrating trained models into a system-level simulator to analyze performance in terms of spatial and time-domain beam prediction. Results show that UEs with strong channel conditions can reduce the measurement overhead by up to 75% without a substantial throughput loss. In contrast, cell-edge users experience limited reductions due to prediction inaccuracies.
For enhanced situational awareness in future networks, sensors such as cameras and LiDAR (Light Detection and Ranging) provide crucial information such as user and blockage position, allowing the network to operate proactively. The next part of the thesis presents a multi-LiDAR approach for precise user positioning alongside a method for predicting line-of-sight (LOS) transitions, anticipating blockages up to 400 ms in advance. Such information from a LiDAR is used for AI/ML model activation/deactivation in AI/ML-based BM to avoid the performance loss of cell-edge UEs due to inaccurate beam predictions.
On the other hand, cell-free massive MIMO systems use distributed access points (APs) to serve multiple UEs on shared frequency-time resources. However, serving all users via all the APs can lead to inefficient spectral and energy usage under low traffic. This research introduces algorithms for optimizing power in cell-free networks by switching off low-contribution APs and minimizing power usage according to spectral efficiency needs and UE mobility. These methods reduce energy costs, outperforming systems where all APs remain active.
Last updated: 8.5.2025