Towards a novel approach for generating realistic IMU sensor measurements, based on real recorded body frame sensor data

This article shows a novel approach to generate body frame IMU data with realistic noise characteristics, based on real measured sensor data. Realistic noise characteristics as defined here are based not only on the sensor noise itself, but also on the high frequency deterministic components of the vehicle system, which are often ignored in ground-based navigation simulations. Realistic noise processes of the system are derived, based on a spectral analysis of segmented maneuvers of real IMU data, gathered from previous experiments. The measurements were recorded over the last decade at the lnstitut für Erdmessung (1FE) for a variety of experiments and projects. Here, the results from a single experiment are highlighted including spectral analysis for both amplitude and phase data. A potential noise generator is outlined based on these results. As the noise processes are non-normal and nonstationary, a histogram-based distribution of the spectrum amplitudes is used as a basis for the simulation. Phase dependencies between frequency bins of different axes can also be simulated via Markov chain Monte Carlo (MCMC) methods, by fitting a multivariate von-Mises distribution per frequency bin between sensor axes. In this contribution, the analysis process as a whole is described on the basis of one exemplary car trajectory, to give some guidelines as to how simulated measurements for such an IMU generator can be obtained with a given dataset. The results of the analysis are presented and discussed. It is shown that phase dependencies between accelerations and angular rates exist for multiple frequencies, even after gravity-related oscillations of the accelerations are corrected by a computed attitude solution. Furthermore, an explanation for the origin of these sinusoidal correlations between accelerations and angular rates is given. Finally, an outlook is presented, to separate still remaining systematic effects of the car system from the obtained simulation parameters.