Abstract
The Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) are two main navigation systems with complementary characteristics. In some GNSS challenged environments, due to signal jamming or blockage, INS becomes stand-alone navigation system. INS is a dead reckoning based system, and its performance will gradually degrade with time. This performance degradation is mainly caused by the deterministic errors and stochastic errors according to their sources. Therefore, precise modelling inertial sensors are the first and the most important step for the inertial navigation system. Deterministic errors are due to manufacturing and mounting defects and can be calibrated out from the raw measurements. On the other hand, stochastic errors are the random errors which cannot be removed from the measurements and should be modelled as processes. Therefore, the better identification of the stochastic errors is critical for navigation. Previous work on analysing and modelling inertial sensors is based on the computing the Allan variance of inertial sensors output and using least squares fitting or reading results from the Allan variance log-log plot directly. This paper presents a methodology for estimating the stochastic errors using segmented weighted algorithm. Firstly, individual noise is simulated and the noise spectral density can be obtained through the least square to validate each noise type. Secondly, the simulation studies with the combination of different noise types are carried out to compare with the traditional method. The proposed algorithm demonstrates remarkable improvement, especially in estimate the long-term errors compared with the traditional method. Field test results also confirm the effectiveness of the new method.