CNN Augmented Vector Decomposed Time-Delay Neural Network for Digital Predistortion

Artificial neural networks (ANNs) have been utilized extensively in the field of radio frequency (RF) power amplifier (PA) distortion compensation. However, ANN-based predistortion models with I/Q mapping suffer from limited accuracy and high computational cost. To address this, we propose a CNN-based predistortion model with vector decomposition, which improves linearization while reducing complexity. Compared to existing I/Q-structured digital predistortion (DPD) models, it achieves superior performance. By introducing a 1D convolutional layer before fully connected layers and rearranging input vectors, the model reduces feature dimensionality from two perspectives, maintaining performance comparable to the augmented vector decomposition time-delay neural network (AVDTDNN) with lower computational cost.