Frequency-domain ANN Non-linear Active Device Model for Harmonic-Balance-based CAD

In this work we demonstrate a feed-forward Artificial Neural Network (ANN) model for microwave active devices, here an FET, where the input and outputs are the device harmonic port waves. The ANN has been trained through an in-house tool implementing back-propagation and gradient descent, with load-pull data sweeping the available input power from back-off to harsh compression. The ANN model can be seamlessly used as a replacement of any active device model in microwave CAD tools implementing the Harmonic Balance (HB) algorithm. As a demonstrator, we developed in MATLAB an HB circuit solver implementing the ANN device model within an external circuit simulating the device with different loading conditions. The ANN modeling approach been successfully applied both in class A and a class B bias.