Conservative Estimation of Inertial Sensor Errors Using Allan Variance Data

To understand the error sources present in inertial sensors, both the white (time-invariant) and correlated noise sources must be properly characterized. To understand both sources, the standard approach (IEEE standards 647-2006, 952-2020) is to compute the Allan variance of the noise and then use human-based interpretation of linear trends to estimate the separate noise sources present in a sensor. Recent work has sought to overcome the graphical nature and visual-inspection basis of this approach leading to more accurate noise estimates. However, when using noise characterization in a filter, it is important that the noise estimates be not only accurate but also conservative, i.e., that the estimated noise parameters overbound truth. In this paper, we propose a novel method for automatically estimating conservative noise parameters using the Allan variance. Results of using this method to characterize a low-cost MEMS IMU (Analog Devices ADIS16470) are presented, demonstrating the efficacy of the proposed approach.