Abstract
Artificial intelligence (AI) has been widely used in antenna design. For complex parameters of antenna configuration, computational intelligence effectively accelerates the design process. Sometimes, simple antenna structures are not able to meet the specifications, and unconventional antenna structures with many sensitive parameters may be needed. Direct simulations of unconventional antenna structures can be very time-consuming, especially when they rely on the trial-and-error approach of human experience. However, only the use of optimization algorithms adopted in the antenna parameters decision is still time-consuming. By assisting with the AI model, the efficiency of antenna design can be significantly enhanced.Chapter 1 makes a comprehensive literature review of existing techniques for computational intelligence in electromagnetics.
Chapter 2 proposes a novel directional dielectric resonator antenna design method. This method uses a heterogeneous graph convolutional network (Het-GCN). Unlike traditional optimization techniques that require extensive simulations and manual adjustments, the proposed approach models the antenna structure as a graph to capture the non-Euclidean relationships between different components. The Het-GCN is trained to predict both the reflection coefficient and radiation pattern simultaneously, enabling efficient forward and inverse antenna designs. The method allows for rapid generation of antenna characteristics, significantly reducing design time. To demonstrate its effectiveness, four cylindrical dielectric resonator antennas (DRAs) with different main-beam directions (0°, 30°, 60°, and 90°) are synthesized by adjusting the dielectric constant distribution. Comparative analysis shows that the Het-GCN outperforms conventional machine learning models in accuracy and generalization. This work establishes a new AI-driven framework that is scalable, flexible, and highly efficient for advanced antenna design tasks.
Chapter 3 proposes an all-dielectric transmitarray (TA) in the 26.5–28.5 GHz frequency band. The proposed TA is optically transparent and can be flexibly attached to planar window glass. It is obtained using novel transparent transmit unit cells. Each transparent cell consists of a sapphire cylinder, padding material, matching glass and attached window glass. Having no metallic structures, the phase shift provided by this unit cell covers a range of 362° with transmission magnitude loss lower than 2.25 dB at 28 GHz. A deep-learning approach using a generative adversarial network (GAN) is used to synthesize the TA. By using the intelligent algorithm, the number of demanded sapphire cylinders can be significantly reduced. To verify the theory, a transparent TA prototype is designed, fabricated and tested. It has a measured realized gain of 24.93 dBi. The measured and simulated results are in good agreement. Our proposed transparent TA can be conveniently attached to different windows in our daily life, such as building or car windows.
Overall, the novel AI-assisted antenna design methodology is investigated in this thesis. Using the proposed methodology, different directional LP and CP DRA and an all-dielectric transparent transmitarray are designed. The improved design method makes our proposed designs suitable for various 5G wireless communication systems.
| Date of Award | 31 Jul 2026 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Kwok Wa LEUNG (Supervisor) |
Cite this
- Standard