Low Computational Complexity Digital Predistortion Based on Independent Parameters Estimation

Abstract

In wide-band digital predistortion linearizers, the number of coefficients of a simplified Volterra polynomial model required to model memory effects can increase dramatically, which causes large computational complexity, ill-conditioning or overfitting problems. We propose a novel digital predistortion (DPD) implementation approach called covariance matrix based independent parameters estimation (CM-IPE) method for a direct learning structure (DLA). In the approach, we use the constant transformation matrix to replace the time-varying transformation matrix because of the stationary and ergodic nature of input signals. And then the principal component analysis (PCA) method is applied for independent parameters estimation. The proposed method can reduce computational complexity. And by utilizing the PCA technique, the coefficients can be estimated independently which, at the same time, can prevent ill-conditioning or overfitting problems. Experimental results demonstrate that the proposed approach realizes the equivalent linearization performance as the traditional DLA method at lower computational complexity.