Low-Rank Adaptation-Based Digital Predistortion of RF Power Amplifiers for Dynamic Scenarios

This paper proposes a parameter-efficient dynamic digital predistortion (DPD) model based on low-rank adaptation (LoRA). By utilizing a fixed Kolmogorov-Arnold network (KAN) backbone to capture shared global nonlinearities, state-specific variations are constrained to low-rank residuals, facilitating efficient adaptation via LoRA matrix. Experimental validation on a broadband Doherty power amplifier (PA) across 27 configurations demonstrates that the proposed architecture achieves excellent linearization performance. Notably, the model maintains high accuracy with 433 running parameters while requiring only 16 updating parameters for state adaptation, verifying its suitability for resource-constrained scenarios.