Training-based and blind algorithms for sparsity-aware distributed learning

In this paper, two novel algorithms for distributed estimation of sparse signals are presented. The algorithms follow an iterative greedy two-step procedure. The first algorithm operates in a training based manner, i.e., the nodes of the network have access to input-output data, whereas the second operates blindly, i.e., the nodes observe output data only. In both cases, the nodes cooperate with each other, by exchanging information with the neighboring nodes. The goal is twofold, first to identify the support set of the unknown signal, and then the non-zero values, which are restricted in the active support set. Theoretical results are outlined and an experimental validation of the proposed algorithms is carried out.