Deep Learning Driven Design of Highly Efficient Harmonic-Tuned Class F-1 Power Amplifiers

Paper i proceeding, 2025

This paper demonstrates a deep-learning-driven approach for designing inverse class F (Class F−1) power amplifiers (PAs) with pixelated layout structures. We employ convolutional neural networks (CNNs) and evolutionary algorithms to synthesize the complete output network of the Class F−1PA, integrating matching, harmonic-tuning, and biasing functionalities into a unified design. To validate the concept, we constructed a prototype Class F−1PA using GaN HEMT transistors. The experimental results exhibit an output power of 40.9-41.5 dBm and a peak drain efficiency of 75%-83% over a 2.0-2.4 GHz frequency range. To the best of our knowledge, this is the first deep-learning-based Class F−1 PA design with a pixelated output network.