Behavioral Model With Multiple States Based on Deep Neural Network for Power Amplifiers
Xin Hu, Shubin Xie, Xin Ji, Xuming Chang, Yi Qiu, Boyan Li, Zhijun Liu, Weidong Wang
Abstract
Digital predistortion is widely used to compensate the nonlinear distortion of power amplifiers (PAs). Among the digital predistortion methods, the polynomial or deep neural networks (DNNs) models are only adopted with one specific state. When the operating conditions of PAs change, it is necessary to retrain and update the coefficients of the PA model. The generalization ability of the DNN models cannot be presented. To address this issue, this letter proposes one new modeling method that can build one generalized PA model with multiple states based on DNN. This method embeds a set of coding vectors representing corresponding states to build the generalized model. Compared with the traditional DNN model, experimental results show that the proposed method can construct the PA model containing multiple states while ensuring good modeling performance.