Neural Network With Fourier Series-Based Transfer Functions for Filter Modeling

The Fourier series is introduced as a transfer function (TF) in the artificial neural network (ANN) for parametric modeling of microwave filters in this letter. The reported pole-residue-based TF leads to an order-changing problem of input samples from vector fitting, which is usually solved with an order-tracking technique or data classification. The proposed Fourier series-based TF does not have to carry out the time-consuming operation because the only coefficient order can be determined for all input samples in an iterative process. Compared with the pole-residue-based TF, moreover, the ANN training involves a small number of TF coefficients in the proposed method. The predicted electromagnetic (EM) response is obtained from the coefficients of the ANN output. An example of the ultrawideband (UWB) filter is employed to verify the effectiveness of the Fourier series-based TF.

Automated summary

This letter introduces the Fourier series as a transfer function in the artificial neural network for parametric modeling of microwave filters. The proposed Fourier series-based transfer function is more efficient and involves fewer coefficients compared to the pole-residue-based transfer function. The effectiveness of this method is verified through an example of an ultrawideband filter.