ZP-BAPS: Z-Score Regularized Real-Time Pruning for Energy-Efficient Power Amplifier Linearization

The rapid deployment of 5G/B5G networks necessitates power-efficient solutions for linearizing compact power amplifiers (PAs) in micro-cell base stations. Conventional Basis-Propagating Selection (BAPS) models face challenges including candidate explosion, collinearity, and diminishing contributions of basis functions. This paper proposes the Z-score Pruned BAPS (ZP-BAPS) model, a dual-strategy enhancement framework that integrates statistical significance validation with dynamic basis pruning to optimize digital pre-distortion (DPD). By embedding Z-score hypothesis testing into the greedy basis selection process, the framework systematically evaluates and eliminates statistically insignificant basis functions while preserving structural integrity through leaf-node prioritization. Experimental validation on a 6W GaN Class AB PA demonstrates that ZP-BAPS achieves significant performance improvements under computational complexity constraints (243 FLOPS): compared to conventional BAPS, it reduces normalized mean square error (NMSE) by 0.44 dB and adjacent channel power ratio (ACPR) by 0.71 dB for 64QAM-OFDM signals. Furthermore, it outperforms the DOMP model by 1.03 dB in NMSE and 1.14 dB in ACPR. Hardware implementation on FPGA confirms real-time feasibility with deterministic latency below 1.2 s. The proposed method establishes a statistically regularized paradigm for energy-constrained DPD, addressing the efficiency-accuracy tradeoff in low-power 5G deployments.