Model Order Reduction Techniques For Power Amplifier Digital Predistortion Linearization

To address the known trade-off between linearity and efficiency of the PA, several linearization techniques have been proposed in the literature. Among them, the most widespread technique is now digital predistortion (DPD) a technique that existing mathematical models that model the behavior of the PA is used, these models are composed of several terms each multiplied by a coefficient. Dealing with broadband signals (tens and hundreds of MHz), it implies an increase in the number of coefficients required in the behavioral model of DPD to compensate both nonlinear effects and memory of PAs. Increasing the number of coefficients has a negative impact on the identification of these, as it not only increases the computational complexity of the extraction (usually carried out by linear regression least squares (LS)), but can lead to problems overfitting and uncertainty. The model order reduction DPD has the dual effect of mitigating the computational complexity and improve the conditioning of the matrix of regressors. The strategy used to reduce the model order PAs, is based partly on regularization techniques such as regression Ridge or Tikhonov, and the LASSO (least absolute shrinkage and selection operator) regression. Other techniques will be based on OMP (Orthogonal Matching Pursuit) combined with BIC (Bayesian Information Criterion). Present a study of the amplifier and some behavioral models PAs, then some of the existing linearization techniques, to stop focusing on DPD. Then we will focus on the reduction techniques PA behavioral model and experimental to finish making a comparison of all results.