Phase-Enhanced Amplitude–Phase Gated Recurrent Unit Neural Network for Digital Predistortion in RF Power Amplifiers

To address the performance limitations and computational redundancy of neural networks in digital predistortion (DPD) for radio frequency (RF) power amplifiers (PAs), this article proposes a phase-aware enhanced amplitude–phase gated recurrent unit (PEAPGRU) model. By structuring the input features within a joint amplitude–phase representation in polar coordinates, the model provides a more cohesive and efficient framework and ensures theoretical compatibility with polar-based RF transmitter architectures. A lightweight module for dynamic feature weighting across channel and spatial dimensions is proposed to suppress redundant information, thereby improving both convergence speed and overall performance. Through a mathematical analysis of distortion mechanisms, the phase-aware mechanism detects temporal phase jumps and enables directional feature enhancement in regions of abrupt phase change, leading to superior nonlinear distortion capture and noise suppression. Compared to state-of-the-art DPD networks, the proposed algorithm achieves a relative improvement of 2.1–4.1dB in ACLR, 1.9–3.9 dB in NMSE, and 1.1–2.5 dB in EVM. Experimental results demonstrate the superior performance of the proposed method in DPD tasks, marking a new advancement in the application of deep learning techniques to the RF front-end design.