Deconvolution of weakly-sparse signals and dynamical-system identification by Gaussian message passing

Lukas Bruderer, Hampus Malmberg, Hans-Andrea Loeliger
2015 2015 IEEE International Symposium on Information Theory (ISIT)  
We use ideas from sparse Bayesian learning for estimating the (weakly) sparse input signal of a linear state space model. Variational representations of the sparsifying prior lead to algorithms that essentially amount to Gaussian message passing. The approach is extended to the case where the state space model is not known and must be estimated. Experimental results with a real-world application substantiate the applicability of the proposed method.
doi:10.1109/isit.2015.7282470 dblp:conf/isit/BrudererML15 fatcat:7xqw34sfp5exzhdpo7ptj4fki4