Filter Design Method Using Small Data Set Trained CNN Models for Parameter Extraction

This research proposes a microwave filter design method using convolutional neural network (CNN) based models trained on small data sets for parameter extraction in microwave filter optimization. The proposed method and CNN models are described with examples for 3-pole and 5-pole parallel coupled line microstrip filters. The proposed method uses CNN models trained on a pair of small data sets to design microwave filters with various center frequencies and bandwidths. As a result, the proposed method saves on computation time during parameter extraction at each space mapping iteration. Furthermore, the models provide fast, reliable, and robust parameter extraction across varying filter requirements.