A Digital Predistortion Technique Based on Improved Sparse Least Squares Twin Support Vector Regression

Abstract

To compensate for the nonlinearity of the power amplifier in the RF front-end of high-capacity satellite communication, more coefficients and higher orders are required in the conventional Digital PreDistortion (DPD) model, which affects severely the resource consumption of the predistortion feedforward path. In this paper, a low-complexity DPD approach based on Improved Sparse Least Squares Twin Support Vector Regression (ISLSTSVR) modeling theory is presented to address this problem. Firstly, the problem that the solution of the Least Squares Twin Support Vector Regression(LSTSVR)model is not sparse is solved by constructing decision function in the original space; At the same time, the truncated least squares loss function is used to increase robustness of the model; Then the low-rank approximation of the kernel matrix is obtained by using Nystrom approximation method, and further Cholesky decomposition is used to reduce the operational complexity of the kernel matrix; Finally, the sparse solution of the model is obtained from low-rank kernel matrix. To verify the effectiveness of the proposed method, experiments are performed using a single-tube gallium nitride (GaN) broadband AB-class power amplifier with a 40 MHz 32 QAM signal for excitation. The experiment result shows that this method can greatly reduce the DPD model coefficients and computational complexity while ensuring the model accuracy, and provides an effective coefficient dimension reduction idea and method for the predistortion technology of the spaceborne RF front-end.