Learnable Edge-Located Activation Neural Network for Digital Predistortion of RF Power Amplifiers

DOI number:10.1109/TMTT.2025.3571780

Affiliation of Author(s):中国科学技术大学信息学院，宁波东方理工大学，GF科技大学

Teaching and Research Group:电子工程与信息科学系

Journal:IEEE Transactions on Microwave Theory and Techniques

Funded by:NNSF 62371436

Key Words:Digital predistortion (DPD), learnable activation function, neural networks (NNs), power amplifiers (PAs).

Abstract:In some application scenarios, radio frequency (RF)  devices face strict power consumption limits, necessitating digital
 predistortion (DPD) models with lower complexity. To address
 the needs of low-complexity scenarios, a novel DPD model called
 learnable edge-located activation neural network (LEANN) is
 developed in this article. Unlike traditional neural network (NN)
 models that use a uniform activation function at the nodes, the
 core idea of the proposed LEANN model is to enhance the
 flexibility and interpretability of nonlinear modeling by employing learnable univariate functions as activation functions on the
 edges of the network. Furthermore, given the varying nonlinear
 characteristics of di erent signal components, a logarithmic
 regularization pruning method suitable for the LEANN model
 is also proposed. This method promotes a greater sparsity in the
 model by reducing the similarity between activation functions.
 Experimental results demonstrate that the proposed LEANN
 model achieves a lower complexity and higher performance
 compared to several classic linear parameter models and NN
 models in linearizing power amplifiers (PAs). Furthermore, the
 pruned LEANN(PLEANN)modelfurther reduces the complexity
 without significantly decreasing the performance.

First Author:Junsen Wang (王俊森)

Co-author:Renlong Han,Qianqian Zhang,Chengye Jiang,Hao Chang

Indexed by:Journal paper

Correspondence Author:Kang Zhou,Falin Liu

Document Code:10.1109/TMTT.2025.3571780

Discipline:Engineering

Document Type:J

Volume:73

Issue:10

Page Number:7105 - 7118

Date of Publication:2025-06-05