Adaptive Kernel Function Sharing for Digital Predistortion of RF Power Amplifiers with Dynamic Resource Block Allocation

A novel method of constructing nonlinear kernel functions for digital predistortion (DPD) of radio frequency (RF) power amplifiers (PA), excited with signals under dynamic resource block (RB) allocation, is proposed. By analyzing the coherent model structures within different operation states, a common kernel function chain is constructed with scalable activations depending on the nonlinearity levels that are unsupervisedly classified. To reduce the number of coefficients used, an approach for adding appending terms is also introduced. The proposed method can achieve significant reduction of kernel function diversity and the number of coefficients. It is experimentally validated with a Doherty PA that is excited with 100 MHz 5G new radio (NR) signals with 63 RB allocation patterns,