An Adaptive MEMS Gyroscope Signal Denoising Method Based on ICEEMDAN-ACF

Abstract

An adaptive MEMS gyroscope signal denoising method based on ICEEMDAN-ACF includes the following steps: Step 1: Pre-denoising the MEMS gyroscope signal; Step 2: Decomposing the MEMS gyroscope signal to obtain multiple intrinsic mode functions (IMFs) and residual terms (Res); Step 3: Calculating the variance of the autocorrelation function (ACF) of each IMF, and using the ACF to adaptively threshold and filter the noise-dominant IMFs; Step 4: Reconstructing the MEMS gyroscope signal together with the residual terms (Res) after removing the noise-dominant IMFs; The above steps complete the denoising of the MEMS gyroscope signal. The purpose of this invention is to solve the technical problem that existing empirical mode decomposition (EMD) algorithms for signal denoising suffer from mode aliasing, which leads to the inability to effectively separate noise from useful signals, affecting the subsequent selection of noise-dominant IMFs.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to an adaptive MEMS gyroscope signal denoising method based on ICEEMDAN-ACF. Background Technology

[0002] Micro-Electro-Mechanical Systems (MEMS) gyroscopes are small in size, light in weight, low in power consumption, and moderate in cost, making them widely used in strapdown inertial navigation, industrial control, and consumer electronics. However, the output signal of MEMS gyroscopes contains a large number of different types of noise, and how to suppress noise has always been one of the research hotspots in the field of inertial navigation.

[0003] MEMS gyroscope errors are categorized into deterministic and random errors. Deterministic errors can be calibrated through turntable experiments, while random errors often exhibit weak nonlinearity, non-stationarity, and slow time-varying characteristics, making them susceptible to structural, circuit, and environmental factors. These random errors severely impact the accuracy of gyroscope angular velocity measurements and are a key factor limiting the improvement of MEMS gyroscope accuracy. Current MEMS gyroscope signal denoising methods mainly include low-pass filtering, wavelet thresholding, Kalman filtering, and Empirical Mode Decomposition (EMD). Low-pass filtering is only suitable for processing stationary signals, while MEMS gyroscope signals have non-stationary characteristics. Wavelet thresholding requires pre-setting the wavelet basis and wavelet decomposition level, making adaptive filtering difficult. Kalman filtering relies on establishing an accurate MEMS gyroscope error model, but the parameters of the MEMS gyroscope error system model are time-varying, making it difficult to establish an accurate error model. EMD algorithms require no prior information, can process non-stationary and nonlinear signals, and are widely used in signal denoising.

[0004] Defects and shortcomings of existing technology:

[0005] The EMD algorithm decomposes any signal into several Intrinsic Mode Functions (IMFs) and a remainder (Residue, Res). MEMS gyroscope signals contain a large amount of Gaussian white noise. After EMD decomposition, the white noise is mainly distributed in the lower-order IMFs. The higher the white noise content of the signal, the higher the order of the low-order IMFs dominated by white noise. Removing the white noise-dominated IMFs can effectively suppress white noise in the signal. The key to the EMD denoising algorithm lies in the effective decomposition of the signal and the accurate selection of noise-dominated IMFs.

[0006] Current MEMS gyroscope signal denoising algorithms based on EMD mainly suffer from the following problems:

[0007] (1) Current IMF screening methods are difficult to accurately screen noise-dominant IMFs: The paper "Improved completeensemble EMD: A suitable tool for biomedical signal processing" uses the Hurst exponent method to screen noise-dominant IMFs. However, the threshold of the Hurst exponent screening method is not accurate enough, which will cause some high-frequency signals to be lost, resulting in poor denoising effect under medium and high signal-to-noise ratio conditions. The paper "Medium-MEMS Gyroscope Random Drift Analysis Method Based on EMD" uses the correlation coefficient method to screen noise-dominant IMFs. However, the threshold of the correlation coefficient screening method is only applicable to low signal-to-noise ratio static MEMS gyroscope signals. For high signal-to-noise ratio dynamic gyroscope signals, the correlation coefficient cannot correctly screen noise-dominant IMFs.

[0008] (2) The thresholds for the Hurts index method and the correlation coefficient method are fixed thresholds and cannot be set adaptively.

[0009] (3) Both of the above papers use the EMD algorithm to decompose MEMS gyroscope signals. The EMD algorithm decomposition has mode mixing phenomenon, which cannot effectively separate noise from useful signals, affecting the subsequent screening of noise-dominated IMFs. Summary of the Invention

[0010] The purpose of this invention is to solve the technical problem that existing EMD algorithms for signal denoising suffer from mode aliasing, which makes it impossible to effectively separate noise from useful signals and affects the subsequent screening of noise-dominated IMFs, as well as the technical problem that current IMF screening methods are unable to accurately screen noise-dominated IMFs.

[0011] An adaptive MEMS gyroscope signal denoising method based on ICEEMDAN-ACF, characterized by comprising the following steps:

[0012] Step 1: Perform pre-noise reduction on the MEMS gyroscope signal;

[0013] Step 2: Decompose the MEMS gyroscope signal to obtain multiple intrinsic mode functions (IMFs) and remainder terms;

[0014] Step 3: Calculate the variance of the autocorrelation function of each order of intrinsic mode function (IMF), and use the autocorrelation function (ACF) adaptive threshold to filter noise-dominant IMFs;

[0015] Step 4: After removing the noise-dominant intrinsic mode function (IMF), reconstruct the MEMS gyroscope signal together with the remaining terms;

[0016] The above steps complete the noise reduction of MEMS gyroscope signals.

[0017] In step 1,

[0018] For MEMS gyroscope signals containing low-energy high-frequency components, the autocorrelation function (ACF) adaptive threshold is used to determine the signal-to-noise ratio of the MEMS gyroscope signal.

[0019] For MEMS gyroscope signals with low signal-to-noise ratio and low-energy high-frequency components, wavelet soft thresholding is used for pre-denoising.

[0020] For MEMS gyroscope signals in other cases, no pre-noise reduction processing is performed.

[0021] Specifically, it includes the following sub-steps:

[0022] 1.1) Calculate the variance of the autocorrelation function of the MEMS gyroscope signal; the variance of the autocorrelation function is even-symmetric, and the autocorrelation function takes the right half of the data.

[0023] 1.2) Use the following formula to determine the signal-to-noise ratio level of the MEMS gyroscope signal:

[0024]

[0025] var() calculates the variance. ACF signal The autocorrelation function (ACF) of the MEMS gyroscope signal. ideal_awg Let [the value] be the autocorrelation function of ideal Gaussian white noise, with a value in [1, 0]. 1*(n-1) ], 0 1*(n-1) This is a 1-row, n-1-column zero matrix, where n is the length of the signal to be denoised. t2 is the signal-to-noise ratio (SNR) level judgment threshold. When the ratio in the above formula is less than t2, the MEMS gyroscope signal is judged as a low SNR signal; when the ratio in the above formula is greater than or equal to t2, the MEMS gyroscope signal is judged as a medium or high SNR signal.

[0026] 1.3) Wavelet decomposition of the low signal-to-noise ratio MEMS gyroscope signal is used, with L wavelet decomposition levels, to obtain wavelet coefficients; after decomposition, the wavelet coefficients are processed using a wavelet soft thresholding function, as follows:

[0027]

[0028] c represents the wavelet coefficients, the threshold λ' uses the Rigrsure threshold, and sgn() is the sign function, which has the following form:

[0029]

[0030] In the formula, a is a variable.

[0031] In step 2, the ICEEMDAN algorithm is used to decompose the MEMS gyroscope signal x, resulting in N IMFs and a remainder term Re s;

[0032] Specifically, it involves the following steps:

[0033] 2.1) Calculate the first-order residual r1 and the first-order intrinsic mode function IMF1:

[0034]

[0035] IMF1 = x - r1

[0036] <·> represents the average of multiple signals; M(·) represents the local mean of a signal. Taking the MEMS gyroscope signal x as an example, its local mean is calculated as follows:

[0037]

[0038] env up (x) and env down (x) represents the upper and lower envelopes of x, which can be obtained by interpolating the maxima and minima of x using cubic polynomials, respectively. n1 is the number of Gaussian white noise particles added when constructing the signal, ω i Let β0 represent the i-th Gaussian white noise with a mean of 0 and a variance of 1, and β0 = ε0std(x) / std(E1(ω)). 1 std() calculates the standard deviation, ε0 is the amplitude of the Gaussian white noise; E k (·) represents the k-th IMF obtained from EMD decomposition;

[0039] 2.2) Calculate the second-order residual r² and the second-order intrinsic mode function IMF²:

[0040]

[0041] IMF2 = r1 - r2

[0042] in β1=ε0std(r1).

[0043] 2.3) If k>2, calculate the k-th order residual r. k and k-th order eigenmode function (IMF) k :

[0044]

[0045] IMF k =r k-1 -r k

[0046] in βk =ε0std(r k ), determine r k Is it monotonous, if r k If it is monotonic, then stop the decomposition; if r k If it is not monotonic, increment k by 1 and repeat steps 2 and 3 until r... k monotonous.

[0047] x can be represented as:

[0048]

[0049] In the formula, N is k at the termination of the ICEEMDAN decomposition, and Res is r at the termination of the decomposition. k .

[0050] In step 3, the noise-dominant IMF is filtered using an ACF adaptive threshold. The ACF adaptive threshold t1 for filtering the noise-dominant IMF takes the following form:

[0051]

[0052] Where λ is the adaptive threshold adjustment coefficient. Calculate the variance of the autocorrelation function of each IMF. If it is less than t1, then the IMF is determined to be a noise-dominated IMF.

[0053] In step 4, after removing the noise-dominant IMF, the MEMS gyroscope signal is reconstructed together with the remaining terms; the following reconstruction formula is used for signal reconstruction:

[0054]

[0055] in Let x' be the set of orders of the MEMS gyroscope that do not contain noise-dominated IMFs, and let x' be the reconstructed MEMS gyroscope signal after denoising.

[0056] Compared with the prior art, the present invention has the following technical effects:

[0057] 1) This invention uses the signal ACF adaptive threshold to determine the signal-to-noise ratio. For low signal-to-noise ratio signals containing low-energy high-frequency components, wavelet soft thresholding is used for pre-denoising, which can effectively suppress the aliasing of high-frequency components and noise during the subsequent ICEEMDAN algorithm decomposition. The Hurst exponent method and correlation coefficient method do not use any pre-denoising method and cannot suppress the aliasing of high-frequency components and noise during the subsequent ICEEMDAN algorithm decomposition.

[0058] 2) This invention uses the ICEEMDAN algorithm to decompose gyroscope signals. Compared with the EMD algorithm used in the Hurst exponent method and correlation coefficient method, it can effectively suppress the aliasing of Gaussian white noise and useful signals.

[0059] 3) This invention uses the variance of the autocorrelation function of the IMF to screen noise-dominant IMFs. The IMF autocorrelation function variance threshold is an adaptive threshold, which can accurately screen noise-dominant IMFs based on signals with different signal-to-noise ratios. In contrast, the thresholds in the Hurst exponent method and the correlation coefficient method are fixed thresholds, which cannot be set adaptively and cannot adapt to MEMS gyroscope signals with different signal-to-noise ratios. Attached Figure Description

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0061] Figure 1 This is a flowchart of the present invention.

[0062] Figure 2 This is a diagram showing the denoising result of the 15dB Doppler (dropper) simulation signal in an embodiment of the present invention.

[0063] Figure 3 The figure shows the measured denoising results of the MEMS gyroscope signal according to the present invention. Detailed Implementation

[0064] like Figure 1 As shown, an adaptive MEMS gyroscope signal denoising method based on ICEEMDAN-ACF is characterized by comprising the following steps:

[0065] Step 1: Perform pre-noise reduction on the MEMS gyroscope signal;

[0066] Step 2: Decompose the MEMS gyroscope signal to obtain multiple intrinsic mode functions (IMFs) and remainder terms;

[0067] Step 3: Calculate the variance of the autocorrelation function of each order of intrinsic mode function (IMF), and use the autocorrelation function (ACF) adaptive threshold to filter noise-dominant IMFs;

[0068] Step 4: After removing the noise-dominant intrinsic mode function (IMF), reconstruct the MEMS gyroscope signal together with the remaining terms;

[0069] The above steps complete the noise reduction of MEMS gyroscope signals.

[0070] In step 1,

[0071] For MEMS gyroscope signals containing low-energy high-frequency components, the autocorrelation function (ACF) adaptive threshold is used to determine the signal-to-noise ratio of the MEMS gyroscope signal.

[0072] For MEMS gyroscope signals with low signal-to-noise ratio and low-energy high-frequency components, wavelet soft thresholding is used for pre-denoising.

[0073] For MEMS gyroscope signals in other cases, no pre-noise reduction processing is performed.

[0074] Specifically, it includes the following sub-steps:

[0075] 1.1) Calculate the variance of the autocorrelation function of the MEMS gyroscope signal; the variance of the autocorrelation function is even-symmetric, and the autocorrelation function takes the right half of the data.

[0076] 1.2) Use the following formula to determine the signal-to-noise ratio level of the MEMS gyroscope signal:

[0077]

[0078] var() calculates the variance. ACF signal The autocorrelation function (ACF) of the MEMS gyroscope signal. ideal_awg Let [the value] be the autocorrelation function of ideal Gaussian white noise, with a value in [1, 0]. 1*(n-1) ], 0 1*(n-1) This is a zero matrix with 1 row and n-1 columns, where n is the length of the signal to be denoised. When the ratio of the above formula is less than t2, the MEMS gyroscope signal is determined to be a low signal-to-noise ratio signal. When the ratio of the above formula is greater than or equal to t2, the MEMS gyroscope signal is determined to be a medium or high signal-to-noise ratio signal. t2 is the signal-to-noise ratio level judgment threshold, which is 1920.

[0079] 1.3) Wavelet decomposition was used on the low signal-to-noise ratio MEMS gyroscope signal. The wavelet decomposition layer L was 3 layers, and wavelet coefficients were obtained. After decomposition, the wavelet coefficients were processed using a wavelet soft thresholding function, as follows:

[0080]

[0081] c represents the wavelet coefficients, the threshold λ' uses the Rigrsure threshold, and sgn() is the sign function, which has the following form:

[0082]

[0083] In the formula, a is a variable.

[0084] In step 2, the ICEEMDAN algorithm is used to decompose the MEMS gyroscope signal x, resulting in N IMFs and a remainder term Re s;

[0085] Specifically, it is divided into the following sub-steps:

[0086] 2.1) Calculate the first-order residual r1 and the first-order intrinsic mode function IMF1:

[0087]

[0088] IMF1 = x - r1

[0089] <·> represents the average of multiple signals; M(·) represents the local mean of a signal. Taking the MEMS gyroscope signal x as an example, its local mean is calculated as follows:

[0090]

[0091] env up (x) and env down (x) represents the upper and lower envelopes of x, which can be obtained by interpolating the maxima and minima of x using cubic polynomials, respectively. n1 is the number of Gaussian white noise particles added when constructing the signal, ω i Let β0 represent the i-th Gaussian white noise with a mean of 0 and a variance of 1, and β0 = ε0std(x) / std(E1(ω)). 1 std() calculates the standard deviation, ε0 is the amplitude of the Gaussian white noise; E k (·) represents the k-th IMF obtained from EMD decomposition;

[0092] 2.2) Calculate the second-order residual r² and the second-order intrinsic mode function IMF²:

[0093]

[0094] IMF2 = r1 - r2

[0095] in β1=ε0std(r1).

[0096] 2.3) If k>2, calculate the k-th order residual r. k and k-th order eigenmode function (IMF) k :

[0097]

[0098] IMF k =r k-1 -r k

[0099] in β k =ε0std(r k ), determine r k Is it monotonous, if r k If it is monotonic, then stop the decomposition; if r k If it is not monotonic, increment k by 1 and repeat steps 2 and 3 until r... k monotonous.

[0100] x can be represented as:

[0101]

[0102] In the formula, N is k at the termination of the ICEEMDAN decomposition, and Res is r at the termination of the decomposition. k .

[0103] In step 3, the variance of the autocorrelation function of each order IMF is calculated, and the noise-dominant IMF is screened using the ACF adaptive threshold. The form of the ACF adaptive threshold t1 for screening the noise-dominant IMF is as follows:

[0104]

[0105] Where λ is the adaptive threshold adjustment coefficient, with a value of 1.5. Calculate the variance of the autocorrelation function of each order IMF. If it is less than t1, the IMF is determined to be a noise-dominated IMF.

[0106] In step 4, after removing the noise-dominant IMF, the MEMS gyroscope signal is reconstructed together with the remaining terms; the following reconstruction formula is used for signal reconstruction:

[0107]

[0108] in Let x' be the set of orders of the MEMS gyroscope that do not contain noise-dominated IMFs, and let x' be the reconstructed MEMS gyroscope signal after denoising.

[0109] Example:

[0110] The present invention performs denoising on 15dB Doppler signals in the following manner:

[0111] Step 1: The Doppler signal contains low-energy high-frequency components. Use the following formula to determine the signal-to-noise ratio:

[0112]

[0113] If the ratio in the above formula is greater than or equal to t2, it indicates a medium to high signal-to-noise ratio signal, and wavelet soft thresholding pre-denoising is unnecessary.

[0114] Step 2: Use the ICEEMDAN algorithm to decompose the 15dB Doppler signal to obtain N IMFs and one remainder Res. In the ICEEMDAN algorithm, the amplitude ε0 of the Gaussian white noise is 0.3, and the number of white noises n1 is 150.

[0115] Step 3: Calculate the variance of the autocorrelation function and the adaptive threshold t1 for each order of IMF, where t1 is 0.6844. If the variance of the autocorrelation function of an IMF is less than t1, then the IMF is classified as a noise-dominated IMF.

[0116] Step 4: Remove the noise-dominant IMF, and reconstruct the MEMS gyroscope signal using the remaining IMF and residual terms to obtain the denoised MEMS gyroscope signal.

[0117] For Doppler signals with signal-to-noise ratios of 3dB, 15dB, and 25dB, denoising was performed using wavelet soft thresholding, EMD decomposition followed by Hurst exponent method and correlation coefficient method to screen for noise-dominant IMFs, and the ICEEMDAN-ACF method proposed in this invention. The signal-to-noise ratios before and after denoising are shown in Table 1.

[0118] Table 1

[0119]

[0120] As can be seen from Table 1, the ICEEMDAN-ACF method proposed in this invention has a higher signal-to-noise ratio after denoising than the wavelet soft thresholding method, Hurst exponent method, and correlation coefficient method under low, medium, and high signal-to-noise ratios.

[0121] The denoising results of a Doppler signal with a signal-to-noise ratio of 15dB are as follows: Figure 2 As shown in the figure, the measured denoising results of the MEMS gyroscope signal are as follows: Figure 3 As shown. Figure 2 As can be seen, the ICEEMDAN-ACF method effectively suppresses noise in the Doppler signal without losing the high-frequency components of the useful signal. From... Figure 3 As can be seen, the ICEEMDAN-ACF method effectively suppresses Gaussian white noise in the measured MEMS gyroscope signal.

[0122] This invention uses an adaptive thresholding mechanism (ACF) to determine the signal-to-noise ratio (SNR) and employs wavelet soft thresholding for pre-denoising, effectively reducing the aliasing of high-frequency signal components with white noise during subsequent ICEEMDAN decomposition. Compared to EMD, ICEEMD effectively suppresses mode aliasing when the signal is subsequently decomposed using ICEEMDAN. Then, the ACF adaptive thresholding is used to screen for noise-dominant IMFs, accurately selecting them based on different SNRs and effectively avoiding incorrect selection of noise-dominant IMFs, thus improving the denoising effect.

Claims

1. An adaptive MEMS gyroscope signal denoising method based on ICEEMDAN-ACF, characterized in that, It comprises the following steps: Step 1: pre-noise reduction of the micro-electro-mechanical system MEMS gyroscope signal; Step 2: decompose the micro-electro-mechanical system MEMS gyroscope signal to obtain a plurality of intrinsic mode functions IMFs and a residual term Res; Step 3: calculate the autocorrelation function variance of each order intrinsic mode function IMF, and use the autocorrelation function ACF adaptive threshold to screen the noise dominant intrinsic mode function IMF; Step 4: after removing the noise dominant intrinsic mode function IMF, reconstruct the micro-electro-mechanical system MEMS gyroscope signal together with the residual term Res; Through the above steps, the denoising of the micro-electro-mechanical system MEMS gyroscope signal is completed; In step 2, the MEMS gyroscope signal is decomposed using an improved complete adaptive noise set empirical mode decomposition (ICEEMDAN) algorithm x , to obtain N intrinsic mode functions (IMFs) and a residual term (Res) Specifically, the following sub-steps are included: 2.1) Calculating the first order residual r 1 and the first order intrinsic mode function IMF1: ； ； where <·> denotes the mean of multiple signals; The local mean of the signal is denoted by x For example, the local mean calculation of the MEMS gyro signal is as follows: ； wherein, env up ( x ) and env down ( x ) are x upper and lower envelopes, respectively, which can be obtained by cubic polynomial interpolation of the maximum and minimum values of x ; , i = 1, 2, …, n 1, n 1 is the number of Gaussian white noises added when constructing the signal, ω i represents the i th Gaussian white noise with a mean of 0 and a variance of 1, β 0= ε 0std( x ) / std( E 1( ω 1 )), std() is a standard deviation, ε 0 is the amplitude of the Gaussian white noise; E k (·) represents the k th intrinsic mode function (IMF) obtained by empirical mode decomposition (EMD). 2.2) Calculate second order residual r 2 and the second order intrinsic mode function IMF2: ； ； wherein = 1, 2,... r 1 β 1 E 2 ω i , i = 1, 2,... n 1 β 1 ε 0std r 1 ; 2.3) k >2, compute k order residual r k and the k order intrinsic mode function IMF k : ； ； wherein, , i = 1, 2, …, n 1, β k = ε 0std( r k ), determine r k whether monotonic, if r k monotonic, then stop decomposition; if r k non-monotonic, then k add 1, repeat step 2.3 until r k monotonic; x may be represented as: ； In the formula, N To improve the termination of the fully adaptive noise set empirical mode decomposition (ICEEMDAN) decomposition k , Res is the residual error at the termination of the decomposition r k ; In step 3, the autocorrelation function variance of each order intrinsic mode function IMF is calculated, the noise-dominant intrinsic mode function IMF is screened using the ACF adaptive threshold, and the ACF adaptive threshold of the screened noise-dominant intrinsic mode function IMF t 1 is as follows: ； wherein, λ is an adaptive threshold adjustment coefficient, and the variance of the autocorrelation function of each order intrinsic mode function IMF is calculated, and if it is less than t 1, the intrinsic mode function IMF is determined as a noise-dominant intrinsic mode function IMF.

2. The method of claim 1, wherein, In step 1, For the micro-electro-mechanical system MEMS gyroscope signal containing low-energy high-frequency components, use the autocorrelation function ACF adaptive threshold to judge the signal-to-noise ratio of the micro-electro-mechanical system MEMS gyroscope signal; For the micro-electro-mechanical system MEMS gyroscope signal with low signal-to-noise ratio and containing low-energy high-frequency components, use the wavelet soft threshold for pre-noise reduction; For other micro-electro-mechanical system MEMS gyroscope signals, no pre-noise reduction is performed.

3. The method of claim 2, wherein, In step 1, for the micro-electro-mechanical system MEMS gyroscope signal containing low-energy high-frequency components, use the autocorrelation function ACF adaptive threshold to judge the signal-to-noise ratio of the micro-electro-mechanical system MEMS gyroscope signal; for the micro-electro-mechanical system MEMS gyroscope signal with low signal-to-noise ratio and containing low-energy high-frequency components, use the wavelet soft threshold for pre-noise reduction; Specifically, the following sub-steps are included: 1.1) Calculate the variance of the autocorrelation function of the micro-electro-mechanical system MEMS gyroscope signal; the autocorrelation function variance is even symmetric, and the autocorrelation function is taken from the right half data; 1.2) Use the following formula to judge the signal-to-noise ratio level of the micro-electro-mechanical system MEMS gyroscope signal: ； wherein var() is a calculation of variance, ACF signal is an autocorrelation function of a micro-electro-mechanical system MEMS gyroscope signal, ACF ideal_awg is an autocorrelation function of an ideal Gaussian white noise, and the value is , is a zero matrix of 1 row n -1 column, n is a length of a signal to be denoised, t 2 is a signal-to-noise ratio level judgment threshold, when the ratio of the above formula is less than t 2, the micro-electro-mechanical system MEMS gyroscope signal is determined as a low signal-to-noise ratio signal, and when the ratio of the above formula is greater than or equal to t 2, the micro-electro-mechanical system MEMS gyroscope signal is determined as a medium or high signal-to-noise ratio signal. 1.3) Use wavelet decomposition to decompose the low signal-to-noise ratio MEMS gyroscope signal, the wavelet decomposition layer number is L, to obtain wavelet coefficients; after decomposition, use the wavelet soft threshold function to process the wavelet coefficients, and the wavelet soft threshold function is as follows: ； c for wavelet coefficients, threshold λ ’ using the Rigrsure threshold, sgn() is the sign function, which is of the form ； wherein a are variables.

4. The method of claim 1, wherein, In step 4, after removing the noise dominant intrinsic mode function IMF, reconstruct the micro-electro-mechanical system MEMS gyroscope signal together with the residual term Res; the following reconstruction formula is used for signal reconstruction: ； wherein, φ is a set of orders of intrinsic modal functions, IMFs, which do not contain noise-dominant IMFs, x' is a denoised reconstructed micro-electro-mechanical system, MEMS, gyroscope signal.

Images