DOI码：10.1109/TMTT.2025.3548777

所属单位：中国科学技术大学信息学院

教研室：电子工程与信息科学系

发表刊物：IEEE Transactions on Microwave Theory and Techniques (Early Access)

项目来源：NNSF 62371436

关键字：Behavior modeling, digital predistortion (DPD), feature mapping, parameter identification, power amplifier (PA), shared module, transfer learning.

摘要：The extraction of linear parameters from high-dimensional features is usually time-consuming, which hinders the timely response of the digital predistortion (DPD) module to changes in the nonlinear characteristics of the radio frequency (RF) power amplifier (PA). To address this issue, we propose a novel sparsely shared module (SSM)-based parameter identification for DPD of RF PAs. Initially, we demonstrate the utilization of the shared module to integrate the commonality of PA nonlinear behaviors across different transmission configurations and to reduce the consumption of storage resources by introducing sparse model masks. This is a move that also facilitates feature mapping among heterogeneous DPD models because it avoids several redundant constructions of basis function matrices. Then, based on the sparsity of the columns in the shared module, a novel sparsity-ordering-based dimensionality reduction technique is proposed to further decrease the number of parameters to be extracted. Finally, considering the distinct performance requirements for different configurations, the sparsity of different DPD models in the shared module is flexible by adjusting allowable performance loss, which is beneficial for the on-demand allocation of the system resource. Experimental results indicate that the proposed method can attain satisfactory modeling accuracy by extracting only a few model coefficients.

第一作者：Guichen Yang

合写作者：Renlong Han,Lei Zhang,Ming Chen,Ping Qi

论文类型：期刊论文

通讯作者：Falin Liu

论文编号：10.1109/TMTT.2025.3548777

学科门类：工学

文献类型：J

卷号：Online

期号：Online

页面范围：1-13

是否译文：否

发表时间：2025-03-20

收录刊物：CPCI-S