A novel data preprocessing method for DPD's polynomial modeling based on principal component analysis

Digital Pre-Distortion (DPD) is a cost-effective way to the linearization of a power amplifier (PA). The polynomial-based model is commonly used to characterize DPD. However, the number of coefficients in polynomial-based model becomes quite large when PA has high-order nonlinearity and deep memory. Moreover, the model often shows numerical problems, which make it unattractive in practical application. In this paper, a novel data preprocessing method using main idea of principal component analysis (PCA) is proposed. The method takes the polynomial terms as ‘characteristics’, and transforms those ‘characteristics’ to some new less ‘characteristics’ by a matrix made up of eigenvectors of a covariance matrix. PCA-based method has the following advantages: it extracts and uses main components to express high nonlinearity and deep memory effects, which decreases the number of coefficients in polynomial-based model and makes the implementation complexity reduce; it reduces the condition number and improves numerical stability; it does not need to know the distribution of input signal. Finally, simulation results indicate the effectiveness of PCA-based method.