Likelihood sampling particle filter for passive localization by a single observe

Passive localization by a single observer, is a typical nonlinear and non-Gaussian filtering problem, and often suffers large initial estimation error, low observability and limited achievable measurements. Particle filter provides a means to achieve the state estimation in a nonlinear and non-Gaussian system, however it may be very inefficient when applied to single observe passive localization and tracking (SOPLAT) application. Considering that the measurements´ likelihood distribution is more concentrated, a new algorithm of sampling from measurements´ likelihood distribution is presented, in which the proposal likelihood distribution is approximated in modified polar coordinate by linear Kalman filtering with the measurements and is used as the proposal for particle filter. Simulation results of comparing the new algorithm with extended Kalman filter (EKF), unscented Kalman filter (UKF) and the EKF and UKF based hybrid particle filter, demonstrate that the new algorithm is superior in convergence speed, tracking precision and filtering stability to others, and the estimation error can approximate the Cramer-Rao lower bound.