Cost-Efficient Baseband DPD for Hybrid MIMO Systems with Shallow Learning Artificial Neural Networks

Enhancing power efficiency and bandwidth of RF power amplifiers (PAs) is always a challenge for future mobile basestation transceivers. Linearization, most commonly as a digital predistortion system (DPD), is essential to enhance power efficiency of individual transmit paths. Especially massive MIMO systems for 5G applications and beyond with multiple transmit paths and PAs are challenging in respect of implementation effort and resource costs. An individual signal processing block for each transmit path is a possible but costly solution. This work shows different solutions for a cost efficient design of a DPD system for hybrid multi-user mMIMO applications with artificial neural networks (ANNs).