AI-Powered Inverse Design of Ku-Band SIW Resonant Structures by Iterative Residual Correction Network

TL;DR

This work tackles the inverse design of multimode Ku-band SIW filters that exhibit both closely spaced and widely separated resonances. It introduces a three-stage framework—Feedforward Inverse Model (FIM), Hybrid Inverse-Forward Residual Refinement Network (HiFR2-Net), and Iterative Residual Correction Network (IRC-Net)—to map desired S-parameter responses to geometric SIW parameters with minimal electromagnetic simulations. Using a CST-generated dataset of 8,721 samples across six geometry parameters, the IRC-Net iteratively refines predictions, delivering substantial accuracy gains: $\text{MSE}$ reduced from $1.91\times10^{-3}$ to $1.46\times10^{-3}$ and $\text{MAE}$ from $2.62\times10^{-2}$ to $2.09\times10^{-2}$, with prediction times below 1 ms. Experimental fabrication and measurement on RT/Duroid 5880 prototypes validate the approach, showing close agreement between predicted, simulated, and measured $S_{21}$ responses and demonstrating the framework’s potential for rapid, robust prototyping of high-frequency microwave components.

Abstract

Designing high-performance substrate-integrated waveguide (SIW) filters with both closely spaced and widely separated resonances is challenging. Consequently, there is a growing need for robust methods that reduce reliance on time-consuming electromagnetic (EM) simulations. In this study, a deep learning-based framework was developed and validated for the inverse design of multi-mode SIW filters with both closely spaced and widely separated resonances. A series of SIW filters were designed, fabricated, and experimentally evaluated. A three-stage deep learning framework was implemented, consisting of a Feedforward Inverse Model (FIM), a Hybrid Inverse-Forward Residual Refinement Network (HiFR\textsuperscript{2}-Net), and an Iterative Residual Correction Network (IRC-Net). The design methodology and performance of each model were systematically analyzed. Notably, IRC-Net outperformed both FIM and HiFR\textsuperscript{2}-Net, achieving systematic error reduction over five correction iterations. Experimental results showed a reduction in mean squared error (MSE) from 0.00191 to 0.00146 and mean absolute error (MAE) from 0.0262 to 0.0209, indicating improved accuracy and convergence. The proposed framework demonstrates the capability to enable robust, accurate, and generalizable inverse design of complex microwave filters with minimal simulation cost. This approach is expected to facilitate rapid prototyping of advanced filter designs and could extend to other high-frequency components in microwave and millimeter-wave technologies.

Figures (13)

• Figure 1: Schematic diagram of the multi-mode SIW structure.
• Figure 2: Equivalent circuit of the multi-mode SIW structure.
• Figure 3: Simulated S parameters ($S_{11}$ and $S_{21}$) of the proposed multi-mode SIW structure with 4 in-band resonances.
• Figure 4: Effect of varying via diameters $D_1$ and $D_2$ on the S-parameters of the multi-mode SIW structure.
• Figure 5: Schematic overview of the Hybrid Inverse-Forward Residual Refinement Network.