Smart Digital Predistortion: Leveraging Cascaded and Artificial Neural Network Models With Pruning for Next-Generation Challenges

In the rapidly evolving telecommunications landscape, digital predistortion (DPD) techniques have emerged as an important solution for improving the linearity and power efficiency of power amplifiers (PAs), especially in the context of 5G New Radio (5G-NR) requirements. This work explores advanced DPD architectures, focusing on the comparative analysis of cascaded models and artificial neural networks (ANNs), with particular emphasis on their effectiveness in addressing the challenges posed by high bandwidth and high peak-to-average power ratio (PAPR) signals. To reduce complexity and avoid overfitting, we implement model order reduction techniques adapted to both model types. Experimental results confirm the superior performance of N-stage cascaded models and ANNs in meeting stringent linearity criteria, including adjacent channel power ratio (ACPR) and error vector magnitude (EVM). Our findings highlight that while both methods effectively address nonlinearity in dual-input wideband PAs, the choice between them depends on specific performance requirements, including computational efficiency and linearization accuracy.