A Novel Antenna Pattern Design Using Generative Adversarial Network

Abstract

In this paper, we propose a novel method of commercial product antenna pattern design using the deep neural network. Our design method is based on the hybrid scheme of both the generative and discriminative model. The advantage of the proposed method is that we can easily create an antenna of a commercial product without in-depth knowledge of the RF theory or antenna design methodology. The designer needs to provide only the designed radiation pattern and reflection coefficient. The neural network used is trained based on the unsupervised learning, which outputs the antenna pattern with the desired radiation characteristics. With the generative model, we generate the pattern and the discriminative network evaluates and outputs the desired antenna pattern. We use the same 8-layer LSTM (long short-term memory) for generator and discriminator to construct a neural network. We use numerical methods to better validate the discriminator's work by high frequency structure simulator (HFSS). © 2019 IEEE.