Information-based clustering and filtering for field reconstruction

A novel communication efficient scheme for field reconstruction is put forth. The proposed framework entails two steps. During the first step sparsity-inducing canonical correlation is utilized to determine different clusters of correlated sensors. The second step relies on least mean-squares adaptive filters to learn, at a cluster head sensor, the data model of all other cluster sensors. The cluster heads send their data and model parameters to a fusion center, which can use them to recover all sensor measurements without the need to talk to all of them. This way the communication cost can be significantly reduced. Numerical tests demonstrate the capability of the proposed scheme in field recovery.