An Angle-Inclusive CNN-Based Digital Predistortion Method for Phased Array Transmitters

The proposed angle-inclusive Digital Predistortion (DPD) technique mitigates angle-dependent distortions and load modulation effects in beamforming transmitters. Using baseband signals and steering angles as inputs, angle-inclusive DPD utilizes a 2D Convolutional Neural Network (2D-CNN) to pre-distort signals, achieving continuous linearization across the beam steering range. Training involves organizing input data into 2D images, each representing a specific steering angle. Measurements with a 60 MHz 64-QAM 5G NR signal on a 4-element linear array at 2.35 GHz show that angle-inclusive DPD achieves average ACPR and NMSE values of −48.6 dBc and −34.3 dB, respectively.