Generative Method of Antenna Optimization Based on Dual Discriminator Generative Adversarial Nets

An antenna optimization method based on dual discriminator generative adversarial nets (D2GAN) is proposed, and an ultra-wideband (UWB) antenna is successfully optimized by this method. At the beginning, an antenna data set containing 100 samples is established randomly and divided into two parts according to their electromagnetic performances. Then, select the part of data with better performance for network training. And the training data generated by the network are also divided into two categories, then combined with the previous data and used as training data for the new network. After each iteration, the new network will generate antennas with better performance than the previous one. As a training example, a UWB antenna covering $3 \mathrm{GHz}-10 \mathrm{GHz}$ is optimized and the required performance is achieved in the third iteration. This method not only reduces the training set, but also enables the network to generate multiple desired antennas at once.