Multidimensional Load-Pull Extrapolation for Accelerated Computer-Aided Design (CAD) Simulations

Computer-aided design (CAD) is widely used in the design and optimization of power amplifiers. Often CAD tasks require many simulations and the time required can be a limiting factor. Additionally, there are many difficulties with characterizing the behavior of high-power and high-frequency power amplifier devices. The work presented reduces the time required for CAD simulations and can use data collected from a device in a measurable region to approximate behavior under difficult-to-measure operating conditions. This work improves on the Wasserstein Generative Adversarial Network (WGAN) Load-Pull Extrapolation network trained for 2d load-pull extrapolation (A. Egbert, C. Baylis, and R. J. Marks, “Extrapolation of Load-Pull Data: A Novel Use of GAN Artificial Intelligence Image Completion,” IEEE Transactions on Microwave Theory and Techniques, vol. 70, no. 11, pp. 4849–4856, Nov. 2022) and extends it to support a 3^rd dimension, input power. This work can extrapolate as few as 16 simulation queries to 24,176 output powers while predicting the optimal reflection coefficient over changing input power to within a median vector error of 0.1098.