Wavelet Transform With Virtual Data and Cluster Annotation for MEMS Inertial Sensors Signal Denoising

Signal processing plays a critical role in MEMS inertial sensors, where wavelet transform is commonly used for noise reduction. The effect of wavelet transform, is often compromised by various factors. The primary challenges are twofold: the classification and selection of wavelet coefficients and the detection and removal of the pseudo-Gibbs effect. In this paper, a novel wavelet-based denoising method is proposed, incorporating a new cluster-based rule for wavelet coefficients selection and as virtual data based approach for eliminating the pseudo-Gibbs effect. To preserve the effective signal as much as possible, the method creatively applies unsupervised learning for wavelet coefficient classification through cluster analysis, allowing for adaptive, ultra-fine classification based on multiple clustering centers. Furthermore, the proposed boundary processing technique, which entirely eliminates the pseudo-Gibbs effect, utilizes the hypothesis testing theory to detect boundaries and place virtual data on both sides effectively addressing the boundary issues of wavelet transform. The verification methods adopted are simulation and navigation tests based on the IMU (inertial measurement unit) hardware platform. The performance of the proposed method is tested by applying it to signal noise reduction in a MEMS system an inertial turntable. Preliminary results shows that compared to other existing techniques the proposed method offers significant advantages in three areas: 1) SNR;2) mean square error;3) navigation parameters error.