A reinforcement learning-driven automated antenna topology optimization method by using enhanced hybrid multilayer perceptron

Abstract A reinforcement learning (RL)-based automated antenna topology optimization method is proposed. The proposed framework can be divided into three phases, which are high-quality dataset construction, electromagnetic (EM) simulation acceleration, and RL-driven automated antenna topology optimization. Based on the high-quality dataset, a fully trained enhanced hybrid multilayer perceptron is proposed to replace time-consuming EM simulations. This approach allows the RL to acquire knowledge from the interaction between antenna topology and the environment quickly, reducing the optimization time cost caused by the large number of EM simulations. Additionally, two crucial components, topology bidirectional mapping strategy (TBM) and topology hierarchical analyzation strategy (THA), are introduced in this work to address the compatibility problems between ML and high-dimension antenna topology data. To verify the effectiveness of the proposed method, a microstrip patch antenna operating at 2.45 GHz is optimized. According to the measurement results, the antenna performance of gain and impedance bandwidth is improved greatly at the same time through the proposed method.