Leveraging machine learning and computer vision for advanced UAV communications

Ahmad, Iftikhar (2025) Leveraging machine learning and computer vision for advanced UAV communications. PhD thesis, University of Glasgow.

Abstract

In the rapidly developing field of wireless communication, there is a growing demand for technologies that can provide flexible deployment, extended coverage, and enhanced performance in next-generation networks. Traditional networks often struggle with high mobility and environmental blockages, highlighting the need for innovative solutions like Unmanned Aerial Vehicles based (UAV-based) dynamic base stations. UAVs offer a promising solution by functioning as dynamic base stations in 5G and 6G networks, with the potential to address these challenges and improve communication reliability and efficiency.

However, the integration of UAVs into wireless communication presents significant challenges. Ensuring reliable communication in high-mobility environments, optimizing beam management techniques, predicting blockages in real time, and managing the latency inherent in UAV-assisted networks all require innovative solutions. These challenges are combined by the need to balance power consumption and processing capacity, especially when performing complex tasks such as on-device machine learning and computer vision-based beamforming.

The first study of this dissertation focuses on the challenge of beam management in milimeter wave (mmWave) 5G and beyond networks, where speedy environmental changes in highmobility scenarios degrade signal quality. Previous studies have highlighted the limitations of traditional beamforming approaches, especially in their ability to adjust to dynamic environments. To enhance this, a novel technique is proposed that integrates computer vision (CV) with ensemble learning, employing the "you look only once" (YOLO-v5) for precise UAV detection and positioning. By stacking two neural networks to refine a meta-learner, this method achieves approximately 90% top-1 accuracy in K-beam predictions, significantly enhancing the signal-to-noise ratio and improving network performance in high-mobility scenarios.
The second study focuses on the problem of proactive blockage prediction and management in mmWave communications, where maintaining line-of-sight connectivity is necessary. Previous studies have stated that traditional reactive handover methods often result in service disruptions due to unexpected blockages. Computer vision techniques used previously resulted to a 40% improvement in user connectivity by predicting and managing blockages. Extending this concept, the study addresses proactive blockage prediction and management in mmWave communications, employing UAVs not only as base stations but also as proactive agents in handover processes. By leveraging CV to detect potential blockages and monitor user movement, the system facilitates proactive handovers to maintain line-of-sight connectivity. This approach, evaluated using a publicly available dataset and incorporating advanced antenna modeling techniques, has demonstrated a 20% enhancement in network performance.

The third study reveals a new approach that utilizes vision-aided machine learning for efficiently and precisely predicting the optimum beam orientations for UAVs using mmWave and terahertz (THz) technologies. Previous research has shown that, while utilizing visual data from UAVs can increase beam prediction accuracy, there are still issues in reducing beam training overhead and managing real-time mobility. Employing data from UAV cameras, the proposed method achieves approximately 90% accuracy in predicting the best beam direction for the top-1, and nearly 100% for the top-3. Performing these computations directly on the UAV (on-device inference) reduces communication delays by 15% and lowers the cost of communication by 50% in terms of power consumption in comparison with ground-based processing, greatly increasing the efficiency of real-time UAV communication. Collectively, these studies underline the potential of using UAVs to improve wireless communication providing innovative solutions for network expansion, precise beam management, and proactive blockage prediction.

This thesis emphasizes UAVs as a cornerstone technology for advancing future wireless communication, setting the stage for more reliable, efficient, and comprehensive communication systems.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: T Technology > TA Engineering (General). Civil engineering (General)
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Hussain, Professor Sajjad and Zoha, Dr. Ahmed
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-84850
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 31 Jan 2025 14:27
• Last Modified: 31 Jan 2025 14:28
• Thesis DOI: 10.5525/gla.thesis.84850
• URI: https://theses.gla.ac.uk/id/eprint/84850

Related URLs

• Article DOI
• Article DOI
• Article DOI
• Conference proceeding
• Conference proceeding
• Book section