Toward Interpretable Digital Predistortion: A Frequency-Domain Behavioral Approach

Abstract

We introduce a compact frequency-domain-based model for digital predistortion (DPD) that emphasizes interpretability and efficiency. The model relies on two key hyperparameters: the order of nonlinear Volterra kernels, which controls the level of nonlinearity, and the complexity of the local polynomial used to capture frequency dependencies around the carrier. This structure simplifies traditional modeling approaches while maintaining high accuracy. The frequency-domain formulation ensures band-limited predistortion, making it well-suited for spectral control.