Scalable Models for Microwave Antennas Using Neural Networks and Space Mapping**

点击翻译按钮获取中文摘要

Antenna design is important for realizing wireless networks in many applications such as 5G, IoT and autonomous vehicles networks. With drastic advances in electromagnetic simulations, electromagnetic based antenna design has become widely adopted in industry. However, electromagnetic based optimization remains computationally expensive. The problem faced by our industry partner is even more challenging. It is constantly required to design different antennas for different frequency bands. This means that the expensive electromagnetic optimization has to be performed again and again whenever antenna specifications are changed. In this project, we develop scalable antenna models such that previous antenna design results can be re-used to help produce new antenna design solutions for different frequency bands. Technical challenges to be addressed includes the establishment of unknown nonlinear relationships between design parameters and antenna performance factors, the expense of generating many data samples from electromagnetic data generation and more. We will develop new approaches to combine space mapping and neural network techniques to solve this modeling problem. The project will be carried out with strong collaboration with our industry partner. The success of this work can help significantly increase the efficiency of antenna design, and strengthen the Canadian competitiveness in this globally competitive field.******