Enhancing Power Amplifier Linearity Via Hybrid Power Back-Off and DPD Optimization

The study addresses nonlinear distortion mitigation in High Power Amplifier(HPA) for 5G NR systems through a cooptimization framework integrating Digital Predistortion(DPD) with power back-off techniques. Employing 5G NR as excitation waveforms, we systematically evaluate different DPD architectures, including models based on memory polynomial and models based on deep learning. The research indicates that the performance of the DPD model based on memory polynomials is highly correlated with the magnitude of power backoff. With a baseline power of -12.19 dBm, a backoff of 1 dB can optimize its relative performance, while further backoff will degrade the relative performance of the DPD. In contrast, the performance of the DPD model based on deep learning is not significantly affected by the magnitude of power backoff. Specifically, polynomial methods offer implementational simplicity for moderate linearity requirements, whereas deep learning architectures enable near-full-power operation with enhanced spectral compliance. These findings advance the design of adaptive HPA linearization systems that dynamically balance efficiency, linearity, and computational complexity in 5 G base stations.