Digital Predistortion Technique Based on Improved Gauss-Newton Method

Gauss-Newton method, a common algorithm in the direct structure of digital predistortion, has a good linearization ability. However, when the predistortion model is configured with small parameters, the Hessian matrix in the Gauss-Newton method is not full rank, and the ill-conditioning problem occurs, so the search direction is not the descending direction. Therefore, this paper proposes a digital predistortion technique based on the improved Gauss-Newton method around these problems. The diagonal terms are injected into the Hessian matrix through the reference ridge regression method. A nonlinear relationship is established between them and the system error. Simulation experiments show that the improved Gauss-Newton method improves the convergence speed by 60% compared with the traditional Gauss-Newton method and reduces the normalized mean square error by 1 dB in the real RF environment.