Attention-Based Deep Neural Network Behavioral Model for Wideband Wireless Power Amplifiers
Zhijun Liu, Xin Hu, Ting Liu, Xiuhua Li, Weidong Wang, Fadhel M. Ghannouchi
Abstract
The behavior models based on artificial neural networks (ANNs) have been widely used in the wideband power amplifier (PA). However, the selected terms of the input signal significantly affect the complexity of the ANNs. In this letter, a method using an attention-based deep neural network (DNN) is proposed to reduce the number of selected input terms for PA modeling. This method first selects the input terms with large contributions to PA modeling offline using the DNN with an attention mechanism. Then, the selected input items are injected into the DNN to build the PA model online. Experimental results show that the proposed method requiring only 1/3 of the input items can achieve good modeling performance with low complexity.