A Segment-Refined Digital Predistortion with Residual-Driven Structural Enhancement for RF Power Amplifiers

This paper proposes a segment-refined Digital Predistortion (DPD) model that intelligently allocates modeling resources to efficiently linearize power amplifiers (PAs) with strong local nonlinearities. The method utilizes Gaussian radial basis functions (GRBFs) for smooth soft-partitioning of the PA's input amplitude range. Central to the approach is a residual-driven algorithm that identifies the most nonlinear amplitude regions. This allows the framework to adaptively augment the model by applying specialized expert models exclusively to the critical nonlinear amplitude regions, while leaving the model structure in other regions unchanged. Experimental results on a Doherty PA with a OFDM 100MHz signal demonstrate that the proposed method achieves a superior ACPR of -49 dBc, representing a 2dB improvement over the GMP, while utilizing significantly fewer coefficients.