Neural Network-Based Nonlinear Interpolation for Digital Predistortion With Bandlimited Feedback for Wideband Power Amplifiers

Abstract

Transmitter linearization through digital predistortion (DPD) typically relies on a dedicated wideband, high sampling-rate feedback receiver. These dedicated receivers are expensive and consume significant power. To address this challenge, we propose a neural network (NN)-based nonlinear interpolator that reconstructs the full-band high-rate power amplifier (PA) output from a low-rate bandlimited observation of the PA response. The proposed solution is based on a gated recurrent unit (GRU) network that learns the residual signal between the full-band high-rate and bandlimited low-rate PA output observations. Experimental results show that the trained NN-based nonlinear interpolator can accurately reconstruct the full-band full-rate PA output signal of a 500 MHz from the bandlimited low-rate observation, i.e., 166.6 MHz signal, and enable effective DPD linearization. The estimation accuracy of the proposed approach in the normalized mean square error is −37.9 dB.