A major focus of our research is statistical signal processing in wireless sensor networks. We pursue a fully distributed (decentralized, cooperative) approach that does not require a central processing unit. Our goal is to estimate, detect, or classify certain global or local states using only local processing at the individual sensor nodes and local wireless communications between neighboring sensor nodes. An example is the task of detecting moving objects such as vehicles or robots and estimating their locations and velocities, using spatially distributed sensors that sense acoustic or radio signals emitted by these objects.
For distributed sequential estimation in wireless sensor networks, we develop distributed particle filters that use consensus algorithms to disseminate relevant statistical information across the sensor network. In particular, we proposed the likelihood consensus scheme for a distributed calculation of the global likelihood function. The resulting distributed particle filters outperform state-of-the-art methods while requiring less intersensor communications.
We also devise distributed estimation methods based on factor graphs and message passing techniques such as belief propagation. An important aspect of these methods is the reduction of intersensor communications through parsimonious parametric representations of probability distributions. The applications we consider include cooperative sensor self-localization and cooperative sensor synchronization. We were able to accommodate noncooperative network nodes (e.g., moving objects not communicating with the sensor nodes) in the message passing scheme by combining belief propagation with the likelihood consensus scheme. Using these methodologies, we devised distributed algorithms for cooperative simultaneous localization and tracking and cooperative simultaneous localization and synchronization.
Further recent results include Monte Carlo methods for blind deconvolution of sparse sequences, with applications to the analysis of electrocardiogram signals; algorithms and performance bounds for minimum variance estimation of sparse vectors; “soft-heuristic” detection algorithms for large multiple-input multiple-output (MIMO) systems; compressive spectral estimators for nonstationary random processes; and information-theoretic bounds for noncoherent MIMO communications.
Signal processing methods are an essential part of several other research areas. For complementary signal processing research, see the sections Mobile Communications, Communication Theory, Flexible Wireless Systems, and Multimedia Systems.