LSTM-CNN for Behavioral Modeling and Predistortion of 5G Power Amplifiers

Abstract

In the Fifth Generation (5G) communication system, power amplifiers (PAs) have serious nonlinear distortion and memory effect. For this reason, this paper proposes a neural network model to linearize PAs. Common PA models such as the long short-term memory (LSTM) network and the deep neural network (DNN) have high complexity problems. Therefore, this paper proposes a behavioral model consisting of LSTM and one dimensional convolutional neural network (1D-CNN), namely LSTM-CNN, for PAs. The LSTM layer is proposed to extract time series information of the input signal to simulate the memory effect of the PA and 1D-CNN structure is used to model the nonlinear characteristics of the PA and reduce model complexity. In addition, the predistortion structure of the PA inverse model based on iteration is used to implement linearization. Finally, modeling results of the class F-PA with the proposed LSTM-CNN show that normalized mean square error (NMSE) can reach about −45 dB. Digital predistortion (DPD) results show that the adjacent channel power ratio (ACPR) can be improved by 14 dB.