A Memory-Based Filter for Long-Term Error De-Noising of MEMS-Gyros

The navigation algorithms which use inertial measurement units (IMUs), such as inertial navigation systems (INSs), always suffer from intrinsic accumulated errors. Bias in gyros induces a significant drift in navigation output especially when micro-electro-mechanical sensor (MEMS) type is used. This error has high- and low-frequency components. De-noising of the long-term error (LTE) (the low-frequency component) is more challenging due to undeterministic behavior and overlapping with carrier motion in the low-frequency band. In this article, a method for de-noising of long-term MEMS-based gyro is presented. In this approach, an auto-regressive (AR) model for the LTE is developed which is being used as the process part of a Kalman filter. To separate the low-band motion dynamic from LTE in the measurement part of the Kalman Filter, the last time epoch of gyro data is subtracted from the current time epoch (memory-based filter). Some static and dynamic experiments have been done for the algorithm evaluation. The static test shows the reduction of LTE by 50%. Also, the method is invoked in INS/Doppler velocity log (DVL) integrated navigation system as a gyro prefilter. The results show that drift and point-to-point final error in position are reduced between 4% and 70% by invoking the de-noising method for gyros.