Subsampled Multichannel Blind Deconvolution by Sparse Power Factorization
Kiryung Lee, Elad Yarkony, Yoram Bresler · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · 2013
In this technical report, we show that sparse power factorization (SPF) is an effective solution to the subsampled multichannel blind deconvolution (SMBD) problem when the input signal follows a sparse model. SMBD is formulated as the recovery of a sparse rank-one matrix. Unlike the recovery of rank-one matrix or of sparse matrix, when there are multiple priors on the solution simultaneously, SPF outperforms convex relaxation approaches both theoretically and empirically. We confirm that SPF exhibits the same advantage in the context of SMBD.