Two-Stage Digital Predistortion With Neural-Network-Assisted Virtual Beamforming for Interchannel Effects in MIMO Systems

Existing method to linearize PAs in multiple-input multiple-output (MIMO) systems either rely on deploying over-the-air (OTA) antenna arrays in the far-field or neglect the effects of interchannel. To address these issues, this letter proposes a two-stage digital predistortion (DPD) approach with neural-network (NN)-assisted virtual beamforming (VB) for interchannel effects in MIMO systems. The main DPD compensates for distortion of PAs, while the sub-DPD compensates for distortion of the interchannel effects. An experimental test was performed using a uniform linear array (ULA) at 3.5 GHz. Based on experimental results, the proposed method achieves up to 5.37-dBc improvement in adjacent channel power ratio (ACPR), closely replicates the linearization performance of OTA, and eliminates the need for remote OTA deployment.