Improvement Techniques for the EM-Based Neural Network Approach in RF Components Modeling

Electromagnetic (EM)---based neural network (NN) approaches have recently gained recognition as unconventional and useful methods for radio frequency (RF) components modeling. In this paper, several improvement techniques including a new data preprocessing technique and an improved training algorithm are presented. Comprehensive cases are compared in this paper. The experimental results indicate that with these techniques, the modified model has better performance.