Adaptive sensing and target tracking of a simple point target with online measurement selection

In previous work we considered the problem of estimating target location parameters using an adaptive sensing paradigm, wherein one attempts to choose the most informative measurement from a set of possibilities characterized by a linear measurement matrix. Here we extend that work to target tracking. At each step in a discrete-time Bayesian filter, a measurement matrix is chosen to illuminate in some near-optimal manner the space of target response vectors in accordance with the prior distribution on the target position. Two methods are proposed. One is based on approximating the manifold of array response vectors as a Euclidean space, and minimizing the Bayesian Cramer-Rao bound on the unknown parameter. The second chooses the rows of the measurement matrix equal to target response vectors at different values of parameter, distributed according to the square root of the prior probability density function for the target location. Performance improvements relative to nonadaptive target tracking methods are quantified for a small number of simulations.