Previous Abstract Return to Session A2a Next Abstract

Session A2a: Quantum Inertial Sensor Technologies and Applications

Observability and Estimability Analysis of a Hybrid Error State CAI-IMU Filter for Different Dynamic Environments
N.B. Weddig, B. Tennstedt, and S. Schön, Leibniz University Hannover, Institut für Erdmessung Schneiderberg
Location: Deer Valley 1-3
Date/Time: Tuesday, Apr. 29, 2:12 p.m.

An observability analysis of a Quantum Inertial Navigation System (QINS) is presented for multiple realistic dynamic scenarios. It is performed on an Error State Extended Kalman Filter (ESEKF), which contains loosely coupled position and velocity measurements and 3-axis differential Cold Atom Interferometer (CAI) sensor measurements. The CAI-based measurements are hybridized with conventional IMU measurements, which results, in combination with position and velocity estimates, in a filter structure that contains position, velocity, acceleration and angular-rate based observations of the system at the same time. This in turn results in increased estimability and observability of the system, as well as lower position, velocity and attitude drift. As CAI-based measurements are only available for low measurement frequencies (i.e. 1-10 Hz), and are also only valid for low dynamics, the improvement in estimability has to be evaluated in realistic scenarios. To this end, realistic trajectories (low frequency deterministic movement) and realistic vibrations (high frequency correlated deterministic movements) are generated and combined for this analysis. With this data, a numerical observability analysis is performed for different combinations of GNSS-based and CAI-based measurements. Furthermore, differences in estimability and observability between vehicle types (cars, aircrafts, trains or ships) are shown. The results demonstrate that, as in a conventional GNSS-IMU sensor fusion, dynamics improve the observability of e.g. scale factors, lever arm components, or misalignment terms. The inclusion of misalignments in the ESEKF, orientation difference between the CAI and IMU, and the introduction of larger lever arms between the CAI and IMU leads to increased dependencies between different bias terms of the IMU, but also between components of the lever arm and misalignments at the CAI-IMU level. They are accentuated when larger vehicle-dependent oscillations are introduced in the system, which is demonstrated by an analysis of singular vectors of the Fisher Information Matrix (FIM). The article provides relevant information about tradeoffs between CAI-IMU model complexity and occurring dynamics, and it gives insights which components of the system need to be pre-calibrated, as their on the fly estimation may lead to an insufficiently resolved state, due to increased dependencies.

Previous Abstract Return to Session A2a Next Abstract