AI-Assisted Deep-Learning-Based Design of High-Efficiency Class F Power Amplifiers

Abstract

This article presents a deep-learning-based approach for designing Class F power amplifiers (PAs). We use convolutional neural networks (CNNs) to predict the scattering parameters of pixelated electromagnetic (EM) layouts. Using a CNN-based surrogate model and an evolutionary algorithm, we synthesize complex Class F output networks. As a proof of concept, we implement a gallium nitride (GaN) HEMT Class F PA, achieving a measured output power of 41.6 dBm and a drain efficiency of 74% at 2.9 GHz. The prototype also linearly reproduces a 20-MHz modulated signal with an 8.5-dB peak-to-average power ratio (PAPR), achieving an adjacent channel leakage ratio (ACLR) of −50.7 dBc with digital predistortion (DPD). To the best of our knowledge, this is the first deep-learning-based Class F PA design using pixelated layout structures.