Measurement Validation of AI-Powered Multidimensional Nonlinear Load-Pull Extrapolation

The design and optimization of high-power nonlinear power amplifiers often requires copious time-consuming measurements. To maximize performance, multiple constraints must be satisfied simultaneously in a complex, high-dimensional search space, resulting in a complex and slow design process. This work validates multidimensional load-pull extrapolation using generative adversarial network image completion as a method for reducing the number of measurements required to characterize a power amplifier over multiple design parameters simultaneously. Using 16 output power measurements of a transistor with different input powers and load impedances, an extrapolated three-dimensional load-pull dataset of 24,192 points can be obtained. The extrapolated data accurately predicts the optimal reflection coefficient over input power within a vector distance of about 0.1 on the Smith Chart, resulting in an achieved mean output power error of under 0.2 dB. This can speed up the high-power nonlinear power amplifier design process and can hasten real-time reconfiguration over multiple parameters.