Jeffreys Prior Based Sparse Bayesian Learning for Digital Predistortion of Power Amplifiers

The article contains possible approach for digital predistorter model sparse estimation. Proposed method sequentially chooses the most likely regressors based on two-layer hierarchical Bayesian prior representation for Relevance Vector Machine (RVM). Significant sparsity is achieved by introduction improper noninformative Jeffreys prior distribution to this scheme. The conducted comparative analysis demonstrates the highest degree of sparsity for trained digital predistorter model with a slight loss of linearization efficiency.