JultikaUniversity of Oulu repository

Abstract

Artificial Neural Networks (ANNs) have gained popularity in modeling the nonlinear behavior of wideband power amplifiers. Recently, modern researchers have used two types of neural network architectures, Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN), to model power amplifier behavior and compensate for power amplifier distortion. Each architecture has its own advantages and limitations. In light of these, this study proposes two digital pre-distortion (DPD) models based on LSTM and CNN. The first proposed model is an augmented LSTM model, which effectively reduces distortion in wideband power amplifiers. The measurement results demonstrate that the proposed augmented LSTM model provides better linearization performance than existing state-of-the-art DPDs designed using ANNs. The second proposed model is a 1D-CNN-LSTM model that simplifies the augmented LSTM model by integrating a CNN layer before the LSTM layer. This integration reduces the number of input features to the LSTM layer, resulting in a low-complexity linearization for wideband PAs. The measurement results show that the 1D-CNN-LSTM model provides comparable results to the augmented LSTM model. In summary, this study proposes two novel DPD models based on LSTM and CNN, which effectively reduce distortion and provide low-complexity linearization for wideband PAs. The measurement results demonstrate that both models offer comparable performance to existing state-of-the-art DPDs designed using ANNs.