Clustered wireless sensor networks for robust distributed field reconstruction based on hybrid shift-invariant spaces

We develop a clustered architecture for wireless sensor networks in order to perform distributed reconstruction of physical fields with low communication overhead. To this end, we introduce hybrid shift-invariant spaces which extend conventional shift-invariant spaces and allow for localized field reconstruction with a computational complexity that is linear in the number of sensors. In contrast to our previous work, this approach allows adaptation to portions of the field with different smoothness properties. In addition, we consider quantized measurements and imperfect knowledge of the sensor positions. Numerical simulations show that our method is robust against such non-idealities and offers significant performance advantages over global reconstruction.