Spectral feature-aided multi-target multi-sensor passive sonar tracking

Passive sonar systems typically provide target bearing estimates that are a nonlinear function of the target state. Multiple state Gaussian Sum extended Kalman filter (EKF) and particle filter (PF) approaches have been combined in a previous multiple hypothesis tracking (MHT) architecture to improve state estimation on bearings-only data. Bearing-only measurements also introduce difficulties in data association as a result of uncertainties, ambiguities, and multiplicity of contacts on a common bearing. In this paper, we shall present an approach to improving data association and filtering by exploiting passive narrowband (PNB) spectral features. The paper identifies the PNB spectral feature target state extension, introduces a Gaussian sum frequency state model, and defines extensions to the likelihood calculations needed for improved data fusion across both kinematic and frequency domains. The track likelihood is based on both kinematic and spectral feature likelihoods. Frequency domain fusion extensions are shown to fit seamlessly into a current MHT architecture.