Direct Input-to-Output Neural Network for Efficient Digital Predistortion of MIMO Transmitters

Multi-Input Multi-Output (MIMO) systems promise increased spectral efficiency in new generations of radio communications. The improvement in performance adds additional challenges to be solved. One such challenge is the digital predistortion (DPD) of power amplifiers (PA) for multiple paths at the same time. Neural Network (NN) based DPD techniques have been proposed to solve this problem, allowing to apply DPD to multiple paths at the same time. The biggest disadvantage of NN DPD is higher complexity than the conventional polynomial based methods. A new NN-based technique is proposed which reduces the complexity while improving DPD performance. This NN technique adds additional direct input-to-output connections which, as shown by experimental validation, both improves DPD performance and simultaneously reduces the complexity.