Once-for-All Adaptive Predistortion for Harmonic Cancellation in Wideband HF Power Amplifiers

In HF communication systems, harmonic cancellation for power amplifiers (PA) is challenging due to frequent frequency switching and wide operating ranges. Existing predistortion-based methods require real-time feedback recalibration when signal parameters change, limiting adaptability. This paper presents a novel approach: a neural network (NN)-based harmonic cancellation (HC) model trained once with iterative carrier injection to build a universal predistortion model covering the entire HF band and amplitude range. By learning from sparse, discrete training data, the model eliminates the need for feedback-driven recalibration during dynamic signal changes, enabling broad generalization to arbitrary signal characteristics across the HF spectrum. The proposed method establishes a one-time, fully adaptive HC-NN framework that dynamically cancels harmonics without retraining or real-time parameter updates. Experimental results with a 10W HF-PA show consistent harmonic suppression below -70 dBc for all signals in the HF band, verifying robust, calibration-free adaptability to frequency and amplitude variations.