Unsupervised algorithms for distributed estimation over adaptive networks

This work shows how to develop distributed versions of block blind estimation techniques that have been proposed before for batch processing. Using diffusion adaptation techniques, data are accumulated at the nodes to form estimates of the auto-correlation matrices and to carry out local SVD and/or Cholesky decomposition steps. Local estimates at neighborhoods are then aggregated to provide online streaming estimates of the parameters of interest. Simulation results illustrate the performance of the algorithms.