Augmented Time-Delay Twin Support Vector Regression-Based Behavioral Modeling for Digital Predistortion of RF Power Amplifier

In this paper, a dynamic behavioral model for digital predistortion (DPD) of RF power amplifier (PA) based on amplitude and phase augmented time-delay twin support vector regression (AP-TSVR) is proposed. Unlike other SVR-based methods, the TSVR model finds a pair of non-parallel planes by solving two related support vector machine (SVM) type problems, namely, the $\varepsilon $ -insensitive up- and down-bound functions. Furthermore, in order to accelerate the training process, an effective linear regression algorithm was used to solve the paired quadratic programming problems (QPPs) of the TSVR model involved. The simulation results show that the proposed model is able to give improved modeling and distortion mitigation capability than the traditional memory polynomial-based model, and reduce CPU training time than the ordinary SVR model, even when the effects of both nonlinear characteristics and memory effects of PA are considered. To verify the effectiveness of the proposed method, experimental verification was performed using single-device gallium nitride (GaN) PA and GaN Doherty PA, respectively. The experimental results show that the new modeling approach can provide very efficient and extremely accurate linearization performance with improving generalization ability.