Data-driven Digital Pre-Distortion Design via Joint Intermediate and Radio Frequency Optimization

Pre-distortion is a key technique in wireless communication systems, which can be classified into digital pre-distortion (DPD) and analog pre-distortion (APD). Both methods focus on optimizing and assessing the nonlinearity in their own areas while the joint pre-distortion design is still missing in the literature. In this paper, a new DPD approach is proposed to minimize the performance metric of the analog RF-domain (i.e., inter-modulation distortion (IMD) or adjacent channel power ratio (ACPR)) and that of the digital IF-domain (i.e., mean square error (MSE)) jointly. To make the joint design feasible, we derive a new hybrid performance metric, where the analog preferred metric is defined in the form of digital signals. On top of that, an effective DPD scheme is developed based on a new dual time-delayed neural network (TDNN) learning architecture. Numerical results show that the proposed scheme is able to significantly improve the IMD/ACPR performance without compromising the MSE, compared to conventional DPD schemes.