De-noising method for MEMS accelerometers based on singular spectral analysis

Micro-electro mechanical systems (MEMS)-based inertial sensors are low-cost but their performances are also degraded because of large uncertainties in their output and effects caused by vibrations. To estimate the attitude of using the MEMS inertial sensors, a pretreatment method to mitigate the noise of the inertial sensors was proposed based on the singular spectral analysis (SSA). With the so-called lagged covariance matrix in this approach, the trend and periodic components were separated with SSA. As a result, the true attitude signal was contained in the trend component, while the vibration signals were contained in the periodic components. Then, the trend component was extracted by use of the number of zero-crossing. Finally, the true attitude measurements pretreated with SSA were utilized as the measurement input of a fusion filter for accurate attitude estimation. The car tests verified the feasibility and the capacity of the method to improve the accuracy of the attitude estimation.