Adaptive Pruning Method of Digital PreDistortion Models Based on DLA Algorithm for LTE/5G Applications

This paper proposes an adaptive pruning method for digital predistortion using a direct learning architecture. To linearize RF power amplifiers, the Generalized Memory Polynomial model is widely used as an inverse dynamical function. However, due to its large number of required terms, many studies focus on reducing implementation complexity. This work mainly uses the doubly orthogonal matching pursuit approach to select active terms while incorporating the Levenberg-Marquardt algorithm to enhance linearization performance. The proposed method achieves a linearity performance nearly equivalent to that of the full model, while significantly reducing the search time for the optimal structure. It’s applied to the linearization of an Open Radio Station for LTE/5G applications, including an asymmetrical two-stage 10W Doherty power amplifier implemented in GaN technology. Comparisons with other state-of-the-art approaches, such as the Gradient method and the Least-Squares algorithm, demonstrate that the proposed method achieves better linearization performance in terms of spectral regrowths reduction and lower computational complexity.