This paper addresses a model order reduction technique based on Ridge’s regression for power amplifier (PA) behavioral models to be used in digital predistortion (DPD) linearization applications. Commonly, the DPD parameters’ extraction is performed by means of a least squares (LS) regression. With Ridge’s regression, the coefficients of the DPD are extracted defining a weighted cost function aimed at minimizing not only the mean square error, but also including a regularization term based on the square of the Euclidean norm of the coefficients’ vector. Taking advantage of this regularization and following a given criterium explained in this paper, it is possible to select the most significant basis functions of the DPD model and thus, not only improving the ovedetermined matrix problem, but also reducing the model’s order and consequently the computational complexity of the DPD linearizer.

Peer Reviewed