Sliding Average Allan Variance for Inertial Sensor Stochastic Error Analysis

Inertial sensor errors include deterministic errors and stochastic errors. Deterministic errors can be calibrated in laboratory by simple computation technique. Stochastic errors can be determined during calibration by adopting special methods because of their random character. The simplest method to determine the stochastic errors for inertial sensors is the Allan variance. This kind of method needs large data to fully characterize the stochastic errors. The normal nonoverlapped Allan variance has quite poor estimation accuracy in long cluster time. The fully overlapping Allan variance and traditional total variance have better estimation accuracy in long cluster time but are quite time consuming for large data set. The not fully overlapping Allan variance and nonoverlapped total variance are suitable for large data set to improve the estimation accuracy in long cluster time with much less time, but their accuracy is still relatively poor in comparison with not fully overlapping total variance. Whereas the not fully overlapping total variance is relatively time consuming and, compared with Allan variance, there is a bias which is not easy to be corrected. This paper proposes a sliding average Allan variance that has comparable estimation accuracy with total variance. The data are not required to extend as the total variance; thus the calculation burden could be reduced greatly. Therefore, it is more suitable for large data set. In addition this method has no bias in comparison with Allan variance, which means no bias correction is required. This method is applied to 12-h static data of three gyroscopes from a position and orientation system with good performance.