Power Amplifier Behavioral Model Dimension Pruning Using Sparse Principal Component Analysis

Abstract

In this paper, an efficient data dimension reduction method which uses sparse principal component analysis (SPCA) is presented for reducing the dimensions of power amplifier (PA) behavioral models. Unlike other model pruning techniques, the SPCA method reduces the data dimension by projecting the variables to a new low dimensional coordinate system while minimizing the model information loss. Meanwhile, the norm L2 and L1 are used as constraint and penalty factor to acquire sparse loadings, which can overcome the non-zero loadings disadvantage of ordinary PCA method and reduce the computational complexity in extracting principal components. Experiment results show that the coefficients of the sparse model can be decreased dramatically using the SPCA method, but almost have the same model performance with the full model.