A fast-converge, real-time auto-calibration algorithm for triaxial accelerometer

Online parameter estimation for nonlinear systems are challenging, especially when only limited computational powers are available. The auto-calibration of triaxial accelerometers is essentially a nonlinear parameter estimation problem. In this paper, a simple but efficient auto-calibration method for micro triaxial accelerometers is proposed. In particular, this method is an online calibration approach based on the six-parameter auto-calibration model, which can be implemented in a wearable health-monitoring device to compensate the estimation errors due to parameter drift. To achieve online calibration, the nonlinear model together with its cost function for auto-calibration is linearized, and an online recursive method is exploited to identify the unknown parameters and remove the bias caused by linearization. This online recursive method is developed by modifying the damped recursive least square estimation (MDRLS), where the estimated unknown parameter can converge stably in a short period. The MDRLS-based method can significantly reduce the calculation complexity comparing to the commonly used nonlinear optimization method. Additionally, the proposed method is validated by both simulation and experiment. The simulation results show the estimated parameters are very close to their true values. By comparing the output results before and after calibration in different experimental conditions, the effectiveness of the proposed method is further verified by real-time experiments.