Network topology identification from corrupt data streams

The interconnectivity structure of many complex systems can be modeled as a network of dynamically interacting processes. Identification of mutual dependencies amongst the agents is of primary importance in many application domains that include internet-of-things, neuroscience and econometrics. Moreover, in many such systems it is not possible to deliberately affect the system and thus passive methods are of particular relevance. However, for an effective framework that identifies influence pathways from dynamically related data streams originating at different sources it is essential to address the uncertainty of data caused by possibly unknown time-origins of different streams and other corrupting influences including packet drops and noise. In this article, a method of reconstructing the network topology from corrupt data streams is provided with emphasis on the characterization of the effects of data corruption on the reconstructed network. The structure of the network is identified by observing the sparsity pattern in the joint power spectrum of the measurements.