How well the analysis of an INS/GPS integration algorithm can be divided into two processes

This work deals with approximated performance evaluation for airborne navigation systems composed of GPS and low cost inertial sensors. Usually performance evaluation requires one to design a multivariable Kalman filter and to perform intensive numerical studies. This work follows Berman (1997) and proposes considering an INS/GPS integration algorithm as a fusion of two processes. The first one is called filtering and deals with the sensor error model. It aims to calculate the high-order position derivatives (velocity, acceleration, jerk). The second process is called separation and its goal is to calculate, from filtering results, all navigation and sensor errors. The most important feature of the above partition is the ability to analyze independently sensor error models and flight trajectories. This paper analyzes the accuracy achieved by the calculations based on the filtering-separation partition. First the underlying assumptions are precisely defined. The errors introduced by those assumptions, for different trajectories and sensor models, are analyzed. Algorithms for filtering and separation are discussed. Finally, leveling errors calculated by the combined filtering-separation algorithms are shown to be very close to the ones calculated by the standard, complete Kalman filter