为了减小低成本微机电系统(MEMS)陀螺仪输出中的噪声, 提出了一种经验模态分解(EMD)的模糊间隔阈值消噪方法。首先通过EMD将信号分解为多个本征模函数(IMF), 并且IMF特性将这些IMF分为三类, 即噪声主导IMF, 混合噪声与信息的IMF, 信息主导的IMF; 对于混合噪声与信息的IMF, 根据不同阈值的特性确定模糊阈值区域, 并设置隶属度函数, 根据IMF系数对应的隶属度值对IMF进行消噪处理; 最后再将经过消噪处理的IMF与分解得到的信息主导的IMF进行重构, 得到消噪信号。实验首先对一段模拟的“bump”信号进行消噪分析, 然后在MEMS陀螺仪上进行验证, 最后对此方法的消噪性能进行了Allan方差分析。实验结果表明, 该方法能有效去除MEMS陀螺仪输出的噪声分量。静止状态下信号的信噪比提高了5.47 dB, 单轴匀速率旋转状态下信号的信噪比提高了2.64 dB; 陀螺信号的各项误差系数均有所降低。实现了陀螺仪输出中噪声与信号的分离, 改善了信号质量, 可以有效提取和识别出有用信息。

To reduce noise in the output of a low-cost Micro-Electro-Mechanical System (MEMS) gyroscope, an Empirical Mode Decomposition (EMD) fuzzy interval threshold denoising method was proposed. The signal was first decomposed into multiple eigenmode functions/intrinsic mode functions (IMFs) using EMD. Then the IMFs were classified into three categories in terms of their IMF characteristics: noise dominated IMF, mixed noise and information IMF, information-dominated IMF. For the mixed noise and information IMF, the fuzzy threshold region was determined based on the characteristics of different thresholds, and the membership function was set accordingly. The IMF was denoised according to the membership value corresponding to the IMF coefficient. Finally, the denoised mixed IMFs and the information-dominated IMFs were reconstructed to obtain the denoised signal. In the experiments conducted, first a denoising analysis was performed on a simulated bump signal, and then verification of the MEMS gyroscope was performed. Finally, Allan variance analysis was performed on the denoising effect of the method. Experimental results show that this method can effectively remove the noise component of the MEMS gyroscope output. Compared with the original output signal, the signal-to-noise ratio of the signal at rest is amplified by 5.47 dB, and by 2.64 dB for the signal in the single-axis uniform rate rotation state. The analysis of Allan variance shows that all the error coefficients of the gyroscope signal are reduced, the separation of noise and signal in the gyroscope output is realized, and the signal quality improved. Useful information can be effectively extracted and recognized.