A new method for designing rectangular microstrip patch antennas using a dual neural network model has been proposed. First, it is proposed to use LSTM + Attention model to predict the amplitude and phase spectrum of the rectangular microstrip patch antenna according to the structural parameter, breaking through the limitations of predicting S11 amplitude spectrum curve solely based on structural parameters in the past. Second, a transformer model was proposed to perform high-precision back prediction of structural parameters using S-parameter points and phase points. Finally, transfer learning was applied to the mutual prediction of structural parameters and gain points, as well as the design of U-shaped slot microstrip patch antennas, all of which achieved good experimental results.
Read full abstract