Dynamic PA power mode DPD on base of 2-dimensional Chebyshev polynomials
Keywords: Power Amplifier, Digital Pre-Distortion, Chebyshev Polynomials, Attention
Abstract: This paper explores modern optimization approaches for functionals describing digital predistortion (DPD) in power amplifier (PA) behavioral modeling under non-stationary operating conditions. The considered class of functionals is based on cascade Wiener–Hammerstein models. A key challenge in this domain arises from the dependence of PA nonlinearities on dynamically varying output power and accamulated of previously modified information, necessitating extensions beyond conventional behavioral models such as Generalized Memory Polynomial (GMP) or standard Chebyshev polynomials.
To adress this, a two dimentional chebyshev polynomial model with attention mechanism was designed as effective extractor of nonlinear distrotions during the broad range of the PA power levels.
Optimization methods drawn from neural network training and non-convex multimodal function minimization are employed to enhance convergence properties and improve model accuracy. The proposed approach achieves a significant reduction in distortion. The study provides insights into the structural properties of the model and the behavior of various optimization strategies, identifying the most effective techniques for improving DPD in dynamically varying PA systems.
Submission Number: 29
Loading