Optimized power allocation by semidefinite programming and unscented transformation for nonlinear sensor network

This paper presents a novel technique of allocating optimized power to wireless sensor nodes in a nonlinear measurement model. We consider the problem of distributed estimation of a random vector-valued parameter in an energy-constrained sensor network. Noise-corrupted local nonlinear observations are transmitted by spatially distributed sensor nodes towards fusion center where estimation of the vector parameter is carried out. In order to guarantee reliable communication, we minimize mean square error of this estimate subject to a constraint on total power consumed by the network. This optimization problem is then recast into a semi-definite program (SDP) which guarantees globally optimized values of the required power gains at sensor nodes. Estimation performance of this novel technique is demonstrated through examples of nonlinear models. Furthermore, for linear models the proposed strategy provides better performance when compared with the previous sub-optimial techniques.