Performance Improvement of Rao-Blackwellized Particle Filter and Its Application to Terrain Referenced Navigation

Abstract

An integrated navigation system is used to improve the performance and robustness of navigation system using different types of navigation methods. Current navigation systems are based on the integration of different navigation methods to achieve high navigation performance combined with high reliability. The combination of inertial navigation and satellite navigation systems are the most common choice, as can be seen in INS/GNSS (Inertial Navigation System/Global Navigation Satellite System) integrated navigation systems. GNSS can effectively correct the time drift error of INS using the satellite navigation signal. However, there are situations, especially in hostile environments like jamming and spoofing, where the satellite navigation is impossible. In such situations, different methods for navigation are essential parts of integrated navigation systems using other types of position fixing techniques. TRN (Terrain Referenced Navigation) is self-sufficient module and independent of external situations. Jamming is almost impossible, unlike with satellite navigation systems. This makes TRN a reliable way to substitute for the satellite navigation systems as a position fixing navigation method with inertial sensors in hostile environments. This paper introduced the INS/TRN integrated navigation system that is autonomous and accurate. The TRN algorithm is proposed to improve the navigation performance and robustness. To perform the TRN method, Bayesian estimation algorithm based RBPF (Rao-Blackwellized Particle Filter) was adjusted in the INS/TRN integrated system, and the proposed algorithm deals with an adaptive RBPF algorithm using the two-layer structural characteristic of conventional RBPF. The TRN is the most commonly used method as secondary navigation technique with INS on the GNSS denied environment. Especially, the signal jamming and spoofing of GNSS is impossible in the TRN, and these conditions make the navigation system which is autonomous and robust in the external signal environment status. In this paper, the new approach of resampling process for RBPF is proposed, and applied to TRN system. The RBPF has recently been researched to integrate the inertial navigation and TRN system. The RBPF can be divided into the particle filter part for the nonlinear states and the Kalman filter for the linear states. The particle filter uses a resampling process to prevent the sample impoverishment problem. The resampling process is important not to decrease the variance of the current estimates but to allow for better estimates in the future. However, the independency of the particles will be lost. It causes destroying the basic concept for the convergence. An implementation issue arises in the application of particle filter part. Conventional RBPF which was proposed by Noudlund adds the artificial noise to the system to overcome the depletion problem using estimated error covariance. When the particles are not enough, the estimated error covariance is sensitive and does not have the strong correlation to the navigation performance. In this dissertation, therefore, more effective adaptation algorithm for small particles is proposed to design the artificial noise of the particle filter using the estimation of the Kalman filter measurement noise because the particle filter result is used to Kalman filter as measurements in the RBPF. The two-layer structure of RBPF for the computational efficiency also can be an effective solution to overcome the particle filter implementation problem. To estimate the measurement noise, an innovation-based adaptive technique is applied to Kalman filter. As the artificial noise is calculated according to navigation performance every step, small number of particles are enough to allow the sufficient navigation result. To verify the navigation performance of proposed algorithm, the simulation result of RBPF is compared with conventional approaches based on nonlinear filtering methods like EKF (Extended Kalman Filter), PMF (Point Mass Filter), conventional RBPF, and regularized particle filtering methods. The navigation performance of proposed RBPF is also analysed using Monte-Carlo simulation on the several cases of environment. In conclusion, the proposed resampling process of RBPF shows the performance enhancement of navigation and the reliability on the GNSS denied environment. As the proposed RBPF is utilized to TRN system, the navigation system gives more robust and stable solution than conventional Bayesian estimation technique based TRN system.