De-noising MEMS inertial sensors for lowcost vehicular attitude estimation based on singular spectrum analysis and independent component analysis

A novel method to de-noise microelectromechanical system (MEMS) inertial sensors by cascaded integration of singular spectrum analysis (SSA) and independent component analysis (ICA) is proposed to improve the attitude accuracy for low-cost attitude estimation. The low-frequency vibration noise and acceleration disturbance induce large errors to attitude estimation since MEMS accelerometers provide attitude measurement for sensor fusion by measuring the gravity vector. It is proposed to remove the low-frequency vibration noise by SSA and to mitigate the acceleration disturbance by ICA. SSA can effectively separate the trend and periodic vibration noise and ICA can effectively extract the acceleration disturbance with the help of turning rate measured by the yaw gyro. The proposed technique was tested on real road experiments showing significant improvement of attitude accuracy.