Parameter Convergent Ridge Regression Algorithm for Digital Predistortion Model

The least squares estimation algorithm is widely used in the parameter extraction of digital predistortion model in power amplifier linearization technology. However, there is a strong correlation between the input and output signals of the actual power amplifier, which makes the ill-conditioned matrix that leads to large errors appear in the parameter extraction of model, thus reducing the linearization performance of the digital predistortion. In order to address the issue mentioned above, a parameter convergence ridge regression (PCRR) algorithm based on digital predistortion model is proposed in this paper. The selection of ridge parameters is determined by the convergence of the actual model parameters, so as to improve the linearity of the output signal of power amplifier. Furthermore, by introducing the ridge parameters, the ill-conditioned matrix is avoided in the least squares estimation algorithm. In order to verify the algorithm proposed in this paper, we carry out simulation test and experimental verification. The experimental conditions are 16 QAM signals with 20 MHz bandwidth. The digital predistortion model adopts the decomposition vector rotation model, and the parameter convergence value is 1. The simulation results show that the proposed method improves the adjacent channel power suppression ratio of the power amplifier output signal from −31.2 dBc to −53.03 dBc, which is 2 dBc higher than that of the least squares algorithm. The experimental results show that the proposed method improves the adjacent channel power suppression ratio of the power amplifier output signal from −35.39 dBc to −57 dBc, which is 0.6 dBc higher than that of the least squares algorithm. Therefore, the proposed algorithm can avoid the occurrence of ill-conditioned matrix in parameter extraction and improve the linearization effect of power amplifier output signal.