Scalar-Metric-Based Digital Predistortion Model Extraction Using Out-of-Band Power-Driven Stochastic Optimization

Abstract

This article proposes a scalar-metric-based digital predistortion (DPD) model extraction architecture for radio frequency (RF) power amplifiers (PAs), where DPD coefficients are extracted using only scalar performance metrics rather than time-domain feedback signals. By formulating DPD coefficient adaptation as a black-box optimization problem, the framework avoids large-scale matrix operations and relaxes the requirements on the feedback path. As a representative implementation, out-of-band (OOB) power is adopted as the optimization objective, and a stochastic optimization scheme based on simultaneous perturbation stochastic approximation (SPSA) is employed, with a quadratic update strategy to improve convergence. Simulations are conducted to analyze the characteristics of the proposed method. Experimental results using wideband 5G NR signals demonstrate that effective linearization can be achieved relying solely on scalar OOB feedback, validating the practicality of the proposed low-complexity architecture.