Geo-tracking of a single constant-velocity target using broadband array sensor data from a maneuvering observer: a comparison of tracking techniques for nonlinear observations

We present a new approach to tracking the geographic position of a target. The dynamic behavior of the target is modeled, in a fixed geographic coordinate system, as constant-velocity motion. The observation process is modeled, in an array referenced coordinate system, as the complex broadband output from elements of a sensor array mounted on a maneuvering ship. This observation model includes the highly nonlinear geographic-to-array coordinate transformation, as well as a complex-to-real transformation. Both the dynamic and observation noises are assumed to be additive and Gaussian which leads to a Kalman filter tracker. The mean and covariance estimate equations necessary for the Kalman filter are in the form of integrals that contain nonlinearities and require numerical integration. The performances of three numerical integration techniques are compared for this problem: Gauss-Hermite quadrature, the unscented filter and Monte Carlo integration.