Unknown-input-driven Variable-weight Distributed EM-based Filter to Improve Accuracy of MEMS Sensors Array with Power Reduction of 6.5 times

This paper proposes a variable-weight distributed expectation maximization (VWDEM) filtering method to achieve accuracy improvement and power reduction of MEMS sensors array under unknown disturbance. First, the local measurement data of each sensor node is used to obtain local state estimates via expectation step(E-step). Next, the weight coefficient is adaptively adjusted based on the above state estimate differences between this local sensor node and its neighbor nodes. Through iterative consensus, the local estimate of sensor node can be then updated quickly due to adaptive weight coefficients. The updated local estimate is finally utilized to identify the sensor bias by maximum step(M-step). The results of numerical calculation relying on experimental tests are given. Comparing with the average consensus distributed expectation maximization (ADEM), the bias variance and power consumption of the MEMS sensors array is reduced to 12% and 15.5% by the proposed method, respectively.