Augmented-LSTM and 1D-CNN-LSTM Models for Linearization of Wideband Power Amplifiers

Long short-term memory(LSTM) neural networks and convolutional neural networks (CNNs) have been used by modern researchers to compensate power amplifier (PA) distortions. In this research, two digital predistortion (DPD) models based on LSTM and CNN called augmented-LSTM and 1D-CNN-LSTM are proposed. The augmented-LSTM model effectively reduces the distortions in wideband PAs. The measurement results show that the augmented-LSTM model gives better linearization performance compared to other state-of-the-art models designed based on neural networks (NNs). On the other hand, the 1D-CNN-LSTM model is proposed to simplify the augmented-LSTM model by integrating a CNN layer prior the LSTM layer causing reduction in the number of input features to the LSTM layer. The measurement results show that the 1D-CNN-LSTM model offers low-complexity linearization for wideband PAs and provides comparable results to the augmented-LSTM model.