一种新的MEMS陀螺仪信号去噪方法

为提高MEMS陀螺仪输出信号的去噪效果,将稀疏分解(sparse decomposition)与提升小波变换(lifting wavelet transform)相结合,提出了一种新的信号去噪方法.首先,建立MEMS陀螺带噪信号的误差模型,并利用小波提升正变换计算带噪信号的非稀疏的小波系数;然后,利用稀疏分解理论恢复小波系数的稀疏性;最后,再通过小波提升反变换重构信号,从而达到去噪的目的.考虑到梯度投影(gradient projection)算法具有全局最优解,运算效率更高,将梯度投影思想引入恢复信号稀疏性的过程中,提出了基于梯度投影的稀疏分解算法,给出了利用梯度投影算法进行信号系数分解的具体步骤,大大简化了计算复杂度,同时提升了算法的稳定性.为验证所提方法的性能,进行了MEMS陀螺信号去噪的静态实验和跑车实验.实验结果表明,此种方法在动静态条件下都可以有效地去除MEMS陀螺仪输出信号中的噪声,尤其是在静态条件下的去噪效果要优于小波阈值滤波方法.同时采用的梯度投影算法相比于正交匹配追踪算法和基追踪算法具有更高的运算效率.

A novel noise reduction method for MEMS gyroscope

To get a better de-noising effect, a novel noise reduction method combining the sparse decomposition with lifting wavelet transform is proposed. Firstly, the error model is established for the MEMS gyroscope output signal with noise, and wavelet coefficient of signals with noise can be obtained by lifting wavelet transform. Then the sparsity of the coefficient is recovered according to sparse decomposition theory. Finally, signals are reconstructed by lifting wavelet inverse transform, i.e. the de-noised signal is thus obtained. In addition, since the gradient projection algorithm is global optimal algorithm with high computational efficiency, the theory of gradient projection is used in the restoration of sparse signal. Specifically, a sparse decomposition based on gradient projection is designed to simplify the algorithm complexity and improve the stability of the algorithm. To verify the performance of the proposed algorithm, the static experiment and dynamic car test on MEMS gyroscope are implemented. The results show that the denoising performance of the new method is better than that of wavelet filter either under the static or dynamic condition, especially under the latter condition. Meanwhile, the CPU time of gradient projection is less than orthogonal matching pursuit (OMP) and basis pursuit (BP).