Aristotle University Of Thessaloniki (AUTH)
Antenna array digital beamforming using AI
Abstract
dc:descriptionThis PhD dissertation investigates deep learning (DL)-based approaches for beamforming (BF) and direction of arrival (DOA) estimation in antenna array systems, aiming to overcome the limitations of traditional BF techniques that depend on accurate channel state information and static propagation models, which often fail in dynamic and interference-rich environments. The thesis begins with an extensive review of the state of the art in DL-based BF, examining its connections with fields such as ultrasound imaging, massive MIMO systems, and intelligent reflecting surfaces (IRS), in order to identify research gaps that shaped the subsequent practical phases. In the first experimental phase, the focus was on predicting BF weights using angle-of-arrival information, following the conventional pipeline-based methodology. Gated recurrent unit (GRU)-based neural networks (NNs) were evaluated alongside other architectures, demonstrating high prediction accuracy in scenarios with up to ten interference signals while significantly reducing inference time compared with the traditional null steering beamforming (NSB). The second phase shifted attention to DOA estimation, establishing a multi-class classification framework that uses correlation matrix information to estimate angles of arrival with high precision. A deep residual NN (ResNet)-based classifier was developed and shown to outperform alternative approaches across a wide range of noise levels and angular resolutions. This phase also introduced an efficient input dimensionality reduction technique that exploits the Hermitian symmetry of the correlation matrix, reducing computational complexity without compromising performance. The final and most novel phase redefines the BF problem by proposing a transformer-based architecture that directly maps correlation matrix inputs to BF weight vectors, thereby eliminating the need for explicit DOA estimation. This transformer-based beamforming model, referred to as TBF-CM, demonstrates strong accuracy, robustness, and scalability in scenarios involving up to six interference signals, maintaining effective main-lobe steering, side-lobe suppression, and null placement. Overall, the results confirm the potential of DL-based architectures to advance digital BF, while future research directions include exploring more compact and efficient models, reducing computational overhead, and enabling real-time deployment in resource-constrained environments.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Haya, Al Kassir
- Χάγια, Αλκαζίρ
Subjects
dc:subject × 20- Διαμόρφωση δέσμης
- Κατεύθυνση άφιξης
- Νευρωνικά δίκτυα
- Βαθιά μάθηση
- Συστοιχία κεραιών
- Ασύρματες επικοινωνίες
- Τεχνητή νοημοσύνη
- Beamforming
- Direction of arrival
- Neural networks
- Deep learning
- Antenna array
- Wireless communications
- Artificial intelligence
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Ηλεκτρική και Ηλεκτρονική μηχανική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Electrical and Electronic Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60503