Pareto-Optimal Multimetric Model Extraction for Digital Predistortion of RF Power Amplifiers for Error Spectrum Redistribution

Abstract

In complex spectrum coordination scenarios in wireless communications, different frequency regions may have distinct linearity requirements. Digital predistortion (DPD), a widely used linearization technique, must address these varying needs properly by effectively redistributing the linearity errors across different frequency regions. To cope with this issue, this article proposes a model extraction method to identify Pareto-optimal coefficients for DPD to meet multimetric requirements, in which distinct requirements on multiple linearity metrics across different frequency regions are posed. Extensive experimental results on a Doherty power amplifier (PA) demonstrate that the proposed method can effectively find the desired solutions during DPD coefficient extraction for a given model, accommodating various spectral error distributions. This allows for enhanced linearization performance in designated frequency regions without damaging performance in other regions. In addition, the proposed method can satisfy certain multimetric requirements with significantly fewer model kernel functions, potentially reducing the digital processing power of DPD.