基于经验模态分解/高阶统计法实现微机械陀螺降噪

MEMS gyro denoising by EMD-HOS method

针对微机电系统(MEMS)陀螺存在的非线性、非平稳噪声,提出了应用经验模态分解/高阶统计(EMD-HOS)的降噪方法对MEMS陀螺进行降噪。首先,采集MEMS陀螺输出信号,根据EMD算法将信号分解成本征模态函数(IMF)。采用Bootstrap技术分别估计各IMF的峰度值,进行高斯特性检验,滤除高斯IMF。接着,使用方差聚合法分别计算IMF的Hurst指数,根据Hurst指数计算阈值,对各IMF进行软阈值处理。将阈值处理后的剩余IMF进行重构,达到降噪的目的。最后,通过交叠式Allan方差分析对滤波前后数据进行处理,绘制Allan方差与相关时间关系曲线,利用非线性最小二乘拟合方法,计算陀螺噪声各项指标。实验表明,EMD-HOS和软阈值处理能够有效地对MEMS陀螺降噪,其信噪比提高了5.6 d B,各项陀螺随机噪声关键指标提高近一个量级。

For the nonlinear and non-stationary signals existing in a MEMS （Micro-electronic-mechanic system）gyro, a denosing method based on the Empirical Mode Decomposition/High Order Statistic （EMD/HOS） was proposed. Firstly, the MEMS gyro signals were captured, and they were decomposed into a cluster of intrinsic mode function （IMF） based on the proposed EMD/HOS sift process. The IMF peak values were estimated by using Bootstrap technology, respectively, to verify its Gaussianity and the Gaussian components were filtered directly. Then the variance algorithm was used to calculate the Hurst exponent of the IMF. According to the Hurst exponent, the threshold was calculated and the each IMF was processed by soft threshold technology. Finally, the remained IMFs after threshold processing were reconstructed to implement the signal denoise. Moreover, the Allan variance algorithm was introduced to analyze the gyro noise, and the characteristic of gyro noise could be observed via the curve of related time and root Allan variance. The conclusion is that EMD-HOS and soft threshold technology decrease the noise of MEMS obviously, the SNR is increased by 5.6 dB, and each indicator ofMEMS; gyro noise is improved almost by one order.