Multiscale fuzzy-adaptive Kalman filtering methods for MEMS gyros random drift

A new time series method was proposed to construct the random drift model for the micro electro mechanical sensor (MEMS) gyro. Based on the wavelet multi scale analysis method, the gyro random drift data was decomposed to a series of scale gyro drift data with depth of 4 using bior1.5 wavelet, each scale signal was rebuilt and then constructed the corresponding multi scale time series models to reduce the overall predict error. Moreover, an adaptive Kalman filter algorithm was proposed to improve the compensation performance for the random drift noise. The noise variance was modified by using the fuzzy adaptive system which is based on the mean and variance margin of residual sequence. The effectiveness of the proposed method was proved by a series of experiments compared with multi scale analysis with simple Kalman filter (SKF). Each random item was reduced using Allan variance analysis.