Neural Network Based Digital Predistortion for Digital MIMO Arrays under Load Modulation

Abstract

In MIMO transmitters, antenna coupling effects can cause load modulation in the active power amplifier (PA) array, changing the nonlinearity of the PAs and impairing the effectiveness of conventional digital predistortion methods. This paper presents a new technique to linearize nonlinear PA arrays in digital MIMO systems, utilizing a novel concept of dual-input time-delay neural networks (NNs). The effectiveness of the proposed dual-input NN approach in linearizing the PA array under load modulation is demonstrated by RF measurement-based experiments at 2.1 GHz. The proposed method shows clear improvements and enhanced performance-complexity tradeoffs over the existing NN-based state-of-the-art models.