Adaptive filtering method, noise reduction system and adaptive filtering method

Abstract

The invention belongs to the technical field of signal processing, and particularly relates to a signal cascade noise reduction method, a noise reduction system and adaptive filtering, and the signal cascade noise reduction method comprises the steps: S1, pre-noise reduction; s2, selecting corresponding adaptive filtering according to a pre-noise reduction result so as to filter noise in the signal in a targeted manner; wherein the optimization parameters of the adaptive filtering are obtained according to the pre-denoised signal, so that the anti-power-frequency interference performance of related products and systems can be improved, and denoising can also be carried out for non-power-frequency interference scenes.

Description

[0001] This application is a divisional application. The application number of the original application is 2022105478298, and the application date is October 4, 2022. The name of the invention is: divisional application for cascade noise reduction method of signal, noise reduction system, and adaptive filtering. Technical Field

[0002] The present invention belongs to the technical field of signal processing, and in particular relates to a signal cascade noise reduction method, a noise reduction system and an adaptive filtering. Background Art

[0003] EEG is a typical weak physiological signal. When the device collects EEG data, it is easily affected by the power frequency signals of other electronic devices and environmental noise, resulting in a decrease in the data signal-to-noise ratio. In severe cases, the data may be completely invalid. Therefore, under the premise of fixed hardware conditions and environment, the quality of the noise reduction method becomes a key factor in determining the data quality of EEG equipment.

[0004] Therefore, based on the above technical problems, it is necessary to design a new signal cascade noise reduction method, noise reduction system, and adaptive filtering. Summary of the invention

[0005] The invention provides a signal cascade noise reduction method, a noise reduction system and an adaptive filtering.

[0006] In a first aspect, the present invention provides a cascade noise reduction method for a signal, comprising the following steps:

[0007] Step S1, pre-noise reduction;

[0008] Step S2, selecting a corresponding adaptive filter according to the pre-noise reduction result to specifically filter out the noise in the signal; wherein

[0009] The optimization parameters of the adaptive filtering are obtained according to the pre-noise-reduced signal.

[0010] Furthermore, the pre-noise reduction includes at least one of removing power frequency interference and removing broadband noise; wherein

[0011] The removing of power frequency interference comprises: pre-filtering fixed strong noise by narrow stopband frequency domain filtering; and

[0012] The broadband noise removal includes: using low-pass filtering to filter out high-frequency noise;

[0013] The method of pre-filtering fixed strong noise by narrow stopband frequency domain filtering includes:

[0014] Set a digital notch filter, that is, make the response of the system function h(z) at the frequency ±ω0 to be zero;

[0015] The response after leaving the frequency ±ω0 quickly returns to a constant, i.e., a pole;

[0016] The system function h(z) is

[0017] Wherein, z represents the independent variable of the system function after Z-transformation, j represents the imaginary unit; the digital notch filter is a second-order IIR filter, and its coefficients are b = [1, -2cosω0, 1], a = [1, -2αcosω0, α 2 ], ω0 is the notch frequency, and 1-α is the normalized stopband width of -3dB.

[0018] Further, the adaptive filtering includes at least one of Wiener filtering, RLS adaptive filtering, LMS adaptive filtering, and Kalman filtering;

[0019] The pre-noise reduction result is whether the pre-noise reduced signal contains stationary noise and non-stationary noise;

[0020] The selecting of the corresponding adaptive filter according to the pre-noise reduction result comprises:

[0021] When the pre-denoising signal contains stationary noise, the adaptive filtering selects at least one of Wiener filtering and RLS adaptive filtering;

[0022] When the pre-denoising signal contains non-stationary noise, the adaptive filtering selects at least one of LMS adaptive filtering and Kalman filtering;

[0023] When the pre-denoising signal contains both stationary noise and non-stationary noise, the adaptive filtering selects at least one of LMS adaptive filtering and Kalman filtering.

[0024] Furthermore, the method for determining whether the pre-denoising signal contains stationary noise and non-stationary noise includes: calculating the statistical characteristics of the noise; if the statistical characteristics change over time, it is non-stationary noise, otherwise it is stationary noise.

[0025] Further, the optimization parameters of the Wiener filter include the order;

[0026] The optimization parameters of the RLS adaptive filtering include stepping and forgetting factors;

[0027] The optimization parameters of the LMS adaptive filtering include stepping;

[0028] The optimization parameters of the Kalman filter include the initial value of covariance;

[0029] When the adaptive filtering selects RLS adaptive filtering or LMS adaptive filtering, the step is optimized according to the proportion of the sine wave in the pre-noise reduction signal, that is,

[0030] The size of the step is positively correlated with the specific gravity value of the sine wave;

[0031] The stepping is optimized according to the proportion of the sine wave in the pre-noise-reduced signal, including:

[0032] Divide the value range of the number of zero crossings / second of the offline signal into several sub-intervals;

[0033] Optimize each subinterval to obtain a corresponding set of steps to form a step table;

[0034] The number of zero crossings / second obtained from the real-time signal is compared with the step table, and the step of the corresponding sub-interval is selected.

[0035] Further, the LMS adaptive filtering includes:

[0036] Delete the sinusoidal component in the observed signal, that is, according to Determine w1 and w2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are weighting coefficients to be estimated;

[0037] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0038] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0039] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum stochastic gradient of the cost function with a fixed step μ: w(k+1)=w(k)+2μ·e(k)y(k), where μ represents the step, e(k)=x(k)―w T (k)y(k),k=0,1,…,K-1;

[0040] The LMS adaptive filter is used to filter out the noise in the signal in a targeted manner, that is, the point-by-point approximation error [e(0), e(1),…, e(K-1)] is used as the filter output.

[0041] Further, the RLS adaptive filtering includes:

[0042] Delete the sinusoidal component in the observed signal, that is, according to Determine w1 and w2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are weighting coefficients to be estimated;

[0043] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0044] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0045] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum overall gradient of the cost function with a fixed step μ, and obtain the sinusoidal interference signal: w(k+1)=w(k)+2μ(r yx ―R yy w(k)), where μ represents the step, R yy and r yx Respectively represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal;

[0046] The RLS adaptive filtering is completed to filter out the noise in the signal in a targeted manner, that is, the sinusoidal interference signal is removed from the known measured signal as the filtering output.

[0047] In a second aspect, the present invention further provides a signal noise reduction system for running the above-mentioned signal cascade noise reduction method, comprising:

[0048] Processor module;

[0049] A pre-noise reduction module, performing pre-noise reduction on the signal; and

[0050] Adaptive filtering module, which specifically filters out noise from the pre-noise-reduced signal;

[0051] The processor module controls the adaptive filtering module to perform corresponding adaptive filtering according to the signal pre-noise reduction result and obtains the optimization parameters of the adaptive filtering.

[0052] In a third aspect, the present invention further provides an LMS adaptive filter for signal noise reduction, comprising:

[0053] Delete the sinusoidal component in the observed signal, that is, according to Determine 1 and 2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated;

[0054] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0055] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0056] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum stochastic gradient of the cost function with a fixed step: w(k+1)=w(k)+2μ·e(k)y(k), where μ represents the step, e(k)=x(k)―w T (k)y(k),k=0,1,…,K-1;

[0057] The LMS adaptive filter is used to filter out the noise in the signal in a targeted manner, that is, the point-by-point approximation error [e(0), e(1),…, e(K-1)] is used as the filter output.

[0058] In a fourth aspect, the present invention further provides an RLS adaptive filtering for signal noise reduction, comprising:

[0059] Delete the sinusoidal component in the observed signal, that is, according to Determine 1 and 2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated;

[0060] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0061] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0062] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum overall gradient of the cost function with a fixed step, and obtain the sinusoidal interference signal: w(k+1)=w(k)+2μ(r yx ―R yy w(k)), where μ represents the step, R yy and r yx Respectively represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal;

[0063] The RLS adaptive filtering is completed to filter out the noise in the signal in a targeted manner, that is, the sinusoidal interference signal is removed from the known measured signal as the filtering output.

[0064] The beneficial effect of the present invention is that the present invention performs pre-noise reduction through step S1; and selects corresponding adaptive filtering according to the pre-noise reduction result in step S2 to filter out noise in the signal in a targeted manner; wherein the optimization parameters of the adaptive filtering are obtained according to the signal after pre-noise reduction, which can improve the anti-power frequency interference performance of related products and systems, and can also perform denoising for non-power frequency interference scenarios.

[0065] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0068] Figure 1 It is a schematic diagram of notch bandwidth selection in the existing method;

[0069] Figure 2 It is a schematic diagram of the universality of existing methods;

[0070] Figure 3 is a flow chart of the cascade noise reduction method of the signal in the present invention;

[0071] Figure 4 It is a schematic diagram of step-by-step filtering of power frequency interference in the present invention;

[0072] Figure 5 It is a schematic diagram of data polluted by power frequency in the present invention;

[0073] Figure 6 It is a schematic diagram of the processing result of comparative scheme 1 in the present invention;

[0074] Figure 7 It is a schematic diagram of the processing results of comparative scheme 2 in the present invention;

[0075] Figure 8 It is a schematic diagram of the processing result of the scheme in the present invention;

[0076] Fig. 9 It is a schematic diagram of the fixed filter coefficient switching channel in the present invention. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] like Figure 1 and Figure 2 As shown in the figure, among the existing methods, frequency domain filtering / notching is the most commonly used noise reduction method. However, it has the problem of frequency band overlap: when the frequency bands of noise and EEG overlap, useful signals will inevitably be lost while noise reduction. Figure 1 Taking power frequency interference as an example, there are two problems when designing a notch solution:

[0079] 1) Notch bandwidth selection: too wide will easily lead to loss of useful signals; too narrow will result in poor notch performance; 2) Poor universality: the notch coefficient is preset and fixed in advance, and is not adaptively adjusted with the subject, channel and environment. The 50Hz notch width is 6Hz, and the 100Hz notch width is 0.4Hz: with the same notch coefficient, channel 7 has a good effect, while channel 1 has a worse effect. In order to solve the above-mentioned problems of frequency band overlap and poor universality, time domain adaptive methods represented by Wiener filtering are used for EEG noise reduction. However, this type of method requires a high input signal-to-noise ratio (in the original data before noise reduction, the noise energy is significantly smaller than the target signal), which is difficult to meet in practical applications.

[0080] Example 1

[0081] like Figure 3As shown, this embodiment 1 provides a cascade noise reduction method for a signal, comprising the following steps: step S1, pre-noise reduction; step S2, selecting a corresponding adaptive filter according to the pre-noise reduction result to filter out the noise in the signal in a targeted manner; wherein the optimization parameters of the adaptive filter are obtained according to the signal after pre-noise reduction; firstly, a fixed, narrow stopband frequency domain filter is used to pre-filter out the fixed strong noise in the environment (such as 50Hz / 60Hz and its multiples of industrial frequency interference), and then a time domain adaptive filter is used to filter out the random noise in the environment in a more targeted manner. Among them, the narrow stopband filtering in the first step does not lose useful signals, and improves the signal-to-noise ratio of the data to a certain extent, creating a favorable premise for the subsequent time domain adaptive filtering performance. It can improve the anti-industrial frequency interference performance of related products and systems, and can also perform denoising for non-industrial frequency interference scenarios.

[0082] In this embodiment, after pre-noise reduction, the corresponding adaptive filtering can be directly selected according to the pre-noise reduction result to filter out the noise in the signal in a targeted manner, and after pre-noise reduction, the step of the LMS filter can be optimized according to the number of zero crossings / second after pre-noise reduction to perform time domain adaptive noise reduction, which can improve the effect and accuracy of noise reduction.

[0083] In this embodiment, the pre-noise reduction includes at least one of removing power frequency interference and removing broadband noise; wherein the removing power frequency interference includes: pre-filtering fixed strong noise by narrow stopband frequency domain filtering; and the removing broadband noise includes: filtering high frequency noise by low-pass filtering; the pre-filtering fixed strong noise by narrow stopband frequency domain filtering includes: setting a digital notch filter, that is, making the response of the system function h(z) at the frequency ±ω0 to be zero; and the response after leaving the frequency ±ω0 quickly returns to a constant, that is, a pole;

[0084] The system function h(z) is

[0085] Wherein, z represents the independent variable of the system function after Z-transformation, and j represents the imaginary unit; according to its system function, it can be known that the digital notch filter is a second-order IIR filter, and its coefficients are b=[1,―2cosω0,1],

[0086] a=[1,―2αcosω0,α 2 ], ω0 is the notch frequency, and 1-α is the normalized stopband width of -3dB.

[0087] What kind of adaptive filtering should be selected for the signal after pre-noise reduction to reduce the difficulty of algorithm implementation, and there is no corresponding theoretical basis at present. In this embodiment, based on specific practice, the following adaptive filtering is listed. The adaptive filtering includes at least one of Wiener filtering, RLS adaptive filtering, LMS adaptive filtering, and Kalman filtering; the pre-noise reduction result is whether the signal after pre-noise reduction contains stationary noise and non-stationary noise; the selection of the corresponding adaptive filtering according to the pre-noise reduction result includes: when the signal after pre-noise reduction contains stationary noise, the adaptive filtering selects at least one of Wiener filtering and RLS adaptive filtering; when the signal after pre-noise reduction contains non-stationary noise, the adaptive filtering selects at least one of LMS adaptive filtering and Kalman filtering; when the signal after pre-noise reduction contains both stationary noise and non-stationary noise, the adaptive filtering selects at least one of LMS adaptive filtering and Kalman filtering.

[0088] In this embodiment, the adaptive filtering is selected by using the indicator of whether the signal after pre-noise reduction contains non-stationary noise signal, and the parameters of the adaptive filtering are optimized to reduce the difficulty of calculation and make it easy to implement. The method for determining whether the signal after pre-noise reduction contains stationary noise and non-stationary noise includes: calculating the statistical characteristics of the noise; if the statistical characteristics change over time, it is non-stationary noise, otherwise it is stationary noise. The reason for selecting the adaptive filtering method based on the stability and non-stationarity of the data is that the past data based on which the current point processing result is obtained by different adaptive filtering methods are different. For example, the LMS method only needs to use the data of one point in the past to obtain the current point processing result, while the RLS method needs to use the data of several points in the past. When it is not stable, the statistical characteristics of the multi-point data in the past are unstable, resulting in a serious decline in the performance of RLS and it is unusable. Therefore, the former can take into account both stable and non-stationary signals, while the latter can only process stable signals. It can be seen that selecting the next adaptive filtering method based on whether the data after pre-noise reduction is stable or not is of great significance to the filtering effect. At the same time, compared with Wiener filtering, RLS adaptive filtering, which also processes stationary noise, does not require prior knowledge of the statistics of the target signal or noise, such as the cross-correlation vector between the target signal and the actual observed signal, the autocorrelation vector of the noise, etc. These statistics are difficult to obtain directly, and the difficulty of indirect estimation is also relatively large. Therefore, feasibility is the advantage of RLS adaptive filtering over Wiener filtering. Compared with Kalman filtering, LMS adaptive filtering, which also processes non-stationary noise, only needs to calculate the error between the estimated value and the actual observed value point by point, and combine the optimizable step parameters to complete the iterative update of the filter coefficient (weighting vector w). The lower computational cost makes LMS adaptive filtering easier to implement in engineering than Kalman filtering.

[0089] In this embodiment, the optimization parameter of LMS filtering is the step, and the reason is that the step is the key parameter that determines the convergence speed and accuracy of the LMS method. The convergence speed of a large step is fast but the accuracy is low, and the precision of a small step is high but the convergence speed is slow. Therefore, it is necessary to select a step that matches different output data. Step optimization strategy for power frequency interference: The number of zero crossings / second can measure the proportion of strong sine waves in the data. After pre-noise reduction, the value is still large, and a larger step can be selected to suppress the power frequency interference more quickly; if the value is small, a smaller step can be selected to filter out the power frequency interference more accurately. A specific embodiment of the preferred step: the value range of the number of zero crossings / second is divided into several sub-intervals, and each sub-interval is optimized based on offline data to obtain a set of steps. When running the noise reduction algorithm, a table is looked up according to the number of zero crossings / second calculated in real time, and the step of the corresponding sub-interval is taken. Of course, the corresponding step can also be obtained through the empirical value of the number of zero crossings / second.

[0090] In this embodiment, the optimization parameters of the Wiener filter include the order; the optimization parameters of the RLS adaptive filter include the step and the forgetting factor; the optimization parameters of the LMS adaptive filter include the step; the optimization parameters of the Kalman filter include the initial value of the covariance; when the adaptive filter selects RLS adaptive filtering or LMS adaptive filtering, the step is optimized according to the proportion of the sine wave in the signal after pre-noise reduction, that is, the size of the step is positively correlated with the proportion of the sine wave; the step is optimized according to the proportion of the sine wave in the signal after pre-noise reduction, including: dividing the value range of the number of zero crossings / second of the offline signal into several sub-intervals; optimizing each sub-interval to obtain a corresponding set of steps to form a step table; comparing the number of zero crossings / second obtained from the real-time signal with the step table, selecting the step of the corresponding sub-interval.

[0091] In this embodiment, the Wiener filter and its LMS implementation are used, and x(k) and n(k) are respectively represented by the target signal and the additive background noise, and then the actual observed signal is:

[0092] y(k)=x(k)+n(k); The basic idea of ​​the Wiener filter is to find an N-dimensional vector w so that its convolution approximates x(k) to the observed signal y(k) N moments ago. The approximation criterion is the minimum mean square error (MMSE), that is:

[0093]

[0094] The above objective function (cost function) is a concave function about w. The w corresponding to its global minimum can be obtained by partial derivative method: R yy and r yx They represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal, respectively. Furthermore, when the noise n(k) and the target signal x(k) are independent of each other, r nn is the noise autocorrelation vector; r yx and r nn The above solutions are only optimal for stationary signals. RLS and LMS methods are usually used in actual filtering.

[0095] In this embodiment, the RLS implementation of the Wiener filter is: the weighted vector w is iteratively approximated point by point in the direction of the maximum overall gradient of the cost function with a fixed step μ:

[0096]

[0097] In theory, when the number of iterations tends to infinity and μ∈(0,1 / λ max ), the RLS method is the optimal approximation of Wiener filtering, where λ max is the observed signal covariance matrix R yy The maximum eigenvalue of

[0098] LMS implementation of Wiener filtering: With a fixed step μ, the weighted vector W is iteratively approximated point by point in the direction of the maximum stochastic gradient of the cost function:

[0099]

[0100] LMS filtering is a stochastic gradient descent method with simple calculation and strong feasibility.

[0101] In this embodiment, the LMS adaptive filtering includes: deleting the sinusoidal component in the observation signal, that is, according to Determine w1 and W2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where W1 and w2 are weighting coefficients to be estimated;

[0102] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0103] Determine the filtered signal, that is, set x(k) as the known measured signal (this method uses reverse thinking and takes the known y(k) as the "virtual" observation signal to be filtered. Use w to perform weighted filtering on y(k) to approximate the sine wave in the real observation signal x(k), and then subtract the approximated sine wave from the observation signal x(k). The approximation error at this time is the filtered signal, let the sine wave w T (K)y(k) approximates x(k), and the approximation error is the filtered signal;

[0104] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum stochastic gradient of the cost function with a fixed step μ: w(k+1)=w(k)+2μ·e(k)y(k), where μ represents the step, e(k)=x(k)―w T (k)y(k),k=0,1,…,K-1;

[0105] LMS adaptive filtering is performed to specifically filter out the noise in the signal, that is, the point-by-point approximation error [e(0), e(1),…, e(K―1)] is used as the filtering output.

[0106] LMS filtering for 50Hz / 100Hz power frequency interference, By determining w1 and w2, the sinusoidal component in the observed signal can be estimated and subtracted from the observed signal. Signal to be filtered: Based on the prior information, generate y(k) = [sin2πfkT s ,cos2πfkT s ] T , f = 50 / 100Hz; X(k) is the known actual observed signal, let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the EEG signal after filtering; Step 1: Initial value of weighted vector: w(0) = [w1, w2] T ; Step μ optimization scheme: based on the number of zero crossings / second after pre-noise reduction; Step 2: Iterative approximation of weighted vector:

[0107] w(k+1)=w(k)+2μ·e(k)y(k),

[0108] e(k)=x(k)―w T (k)y(k),k=0,1,…,K-1;

[0109] Step 3: Complete noise reduction and use the point-by-point approximation error [e(0), e(1),…, e(K-1)] as the filter output.

[0110] In this embodiment, the RLS adaptive filtering includes: deleting the sinusoidal component in the observation signal, that is, according to Determine w1 and w2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are weighting coefficients to be estimated;

[0111] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ]T , where f is the filter frequency corresponding to suppressing interference signals;

[0112] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0113] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum overall gradient of the cost function with a fixed step μ, and obtain the sinusoidal interference signal: w(k+1)=w(k)+2μ(r yx ―R yy w(k)), where μ represents the step, R yy and r yx Respectively represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal;

[0114] The RLS adaptive filtering is completed to filter out the noise in the signal in a targeted manner, that is, the sinusoidal interference signal is removed from the known measured signal as the filtering output.

[0115] In this embodiment, a specific embodiment for power frequency interference is provided. Among them, 1) the power frequency interference in the data is fitted point by point using the LMS method, and then the noise reduction scheme is subtracted, and 2) the LMS step optimization strategy based on the number of zero crossings / second is designed around the power frequency interference. When facing noises such as electromyography and electrooculography, noise reduction processing can also be performed through the "time-frequency domain cascade structure + second-level coefficient optimization": 1) electromyography and electrooculography are broadband noises, and the first-level pre-noise reduction needs to design other filtering methods to replace the notch. Taking electromyography as an example, its typical frequency range is 10-500Hz, and the high-frequency component is more than that of electroencephalography. Therefore, the first-level pre-noise reduction can use a low-pass filter to first filter out the high-frequency electromyography, and the remaining low-frequency electromyography with overlapping frequency bands with electroencephalography is suppressed by the time domain adaptive algorithm in the second level. If the time domain adaptive algorithm has a coefficient to be set, the high-frequency-low-frequency energy ratio can be used to optimize the coefficient, which is similar to the step optimization of LMS using the number of zero crossings / second. 2) Depending on whether the noise is stationary, the second-level time-domain adaptive algorithm can be divided into two categories. One category is more suitable for stationary noise, represented by Wiener filtering and RLS; the other category can deal with both stationary and non-stationary noise, represented by Kalman filtering and LMS. In practical applications, the cascade denoising method of the signal in this embodiment can be combined with the characteristics of different types of noise to optimize the time-domain adaptive algorithm. 3) Parameters to be optimized for different time-domain adaptive filters: Wiener filtering (order), recursive least squares RLS (step, forgetting factor), least mean square algorithm LMS (step), Kalman filtering (covariance initial value).

[0116] In this embodiment, if Figure 4 As shown, the signal cascade noise reduction method is used to filter out the power frequency interference step by step; Figure 5-8 As shown, comparative scheme 1: frequency domain notch only (50Hz stop bandwidth = 6Hz, 100Hz stop bandwidth = 0.4Hz); comparative scheme 2: time domain LMS filtering only (50Hz step = 0.08, 100Hz step = 0.003); the cascade noise reduction method scheme of the signal adopted in this embodiment: pre-noise reduction notch (50Hz stop bandwidth = 1.4Hz, 100Hz stop bandwidth = 0.2Hz); time domain LMS filtering (after optimization according to the number of zero crossings / second: 50Hz step = 0.009, 100Hz step = 0.0015); the cascade noise reduction method scheme of the signal adopted in this embodiment can simultaneously suppress 50Hz and 100Hz power frequency interference without losing high-frequency information. The comparative scheme has poor noise reduction performance and is prone to loss of high-frequency information; as shown Fig. 9 As shown, the filter coefficient is fixed and different channels are switched: the scheme of this embodiment maintains a good noise reduction performance, while the noise reduction performance of the comparative scheme is seriously reduced.

[0117] Example 2

[0118] On the basis of Embodiment 1, this Embodiment 2 further provides a signal noise reduction system for running the above-mentioned signal cascade noise reduction method, including:

[0119] Processor module;

[0120] A pre-noise reduction module, performing pre-noise reduction on the signal; and

[0121] Adaptive filtering module, which specifically filters out noise from the pre-noise-reduced signal;

[0122] The processor module controls the adaptive filtering module to perform corresponding adaptive filtering according to the signal pre-noise reduction result and obtains the optimization parameters of the adaptive filtering.

[0123] Example 3

[0124] On the basis of Example 1, this Example 2 further provides an LMS adaptive filtering for signal noise reduction in Example 1, including:

[0125] Delete the sinusoidal component in the observed signal, that is, according to Determine 1 and 2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated;

[0126] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0127] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0128] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum stochastic gradient of the cost function with a fixed step: w(k+1)=w(k)+2μ·e(k)y(k), where μ represents the step, e(k)=x(k)―w T (k)y(k),k=0,1,…,K-1;

[0129] The LMS adaptive filter is used to filter out the noise in the signal in a targeted manner, that is, the point-by-point approximation error [e(0), e(1),…, e(K-1)] is used as the filter output.

[0130] Example 4

[0131] On the basis of Embodiment 1, this Embodiment 4 further provides an RLS adaptive filtering for signal noise reduction in Embodiment 1, including:

[0132] Delete the sinusoidal component in the observed signal, that is, according to Determine 1 and 2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated;

[0133] Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals;

[0134] Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal;

[0135] Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum overall gradient of the cost function with a fixed step, and obtain the sinusoidal interference signal: w(k+1)=w(k)+2μ(ryx ―R yy w(k)), where μ represents the step, R yy and r yx Respectively represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal;

[0136] The RLS adaptive filtering is completed to filter out the noise in the signal in a targeted manner, that is, the sinusoidal interference signal is removed from the known measured signal as the filtering output.

[0137] To summarize, the present invention performs pre-noise reduction through step S1; and selects corresponding adaptive filtering according to the pre-noise reduction result in step S2 to filter out noise in the signal in a targeted manner; wherein the optimization parameters of the adaptive filtering are obtained according to the signal after pre-noise reduction, which can improve the anti-power frequency interference performance of related products and systems, and can also perform denoising for non-power frequency interference scenarios.

[0138] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0139] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0140] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0141] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. An adaptive filtering method, characterized in that: According to the pre-noise reduction result, the corresponding adaptive filter is selected to filter out the noise in the signal in a targeted manner; The optimization parameters of the adaptive filtering are obtained according to the pre-noise-reduced signal; The adaptive filtering includes at least one of Wiener filtering, RLS adaptive filtering, LMS adaptive filtering, and Kalman filtering; The optimization parameters of the Wiener filter include the order; The optimization parameters of the RLS adaptive filtering include stepping and forgetting factors; The optimization parameters of the LMS adaptive filtering include stepping; The optimization parameters of the Kalman filter include the initial value of covariance; When the adaptive filtering selects RLS adaptive filtering or LMS adaptive filtering, the step is optimized according to the proportion of the sine wave in the pre-noise reduction signal, that is, The size of the step is positively correlated with the specific gravity value of the sine wave; The stepping is optimized according to the proportion of the sine wave in the pre-noise-reduced signal, including: Divide the value range of the number of zero crossings / second of the offline signal into several sub-intervals; Optimize each subinterval to obtain a corresponding set of steps to form a step table; The number of zero crossings / second obtained from the real-time signal is compared with the step table, and the step of the corresponding sub-interval is selected.

2. The adaptive filtering method according to claim 1, characterized in that: The pre-noise reduction result is whether the pre-noise reduced signal contains stationary noise and non-stationary noise; The selecting of the corresponding adaptive filter according to the pre-noise reduction result comprises: When the pre-denoising signal contains stationary noise, the adaptive filtering selects at least one of Wiener filtering and RLS adaptive filtering; When the pre-denoising signal contains non-stationary noise, the adaptive filtering selects at least one of LMS adaptive filtering and Kalman filtering; When the pre-denoising signal contains both stationary noise and non-stationary noise, the adaptive filtering selects at least one of LMS adaptive filtering and Kalman filtering.

3. The adaptive filtering method according to claim 2, characterized in that: The method for determining whether the pre-denoising signal contains stationary noise and non-stationary noise comprises: calculating the statistical characteristics of the noise; if the statistical characteristics change over time, it is non-stationary noise, otherwise it is stationary noise.

4. The adaptive filtering method according to claim 1, characterized in that: The LMS adaptive filtering includes: Delete the sinusoidal component in the observed signal, that is, according to sin[x+tan ―1 (w2 / w1)] to determine w1 and w2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated; Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals; Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal; Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum stochastic gradient of the cost function with a fixed step μ: w(k+1)=e(k)+2μ·e(k)y(k), where μ represents the step, e(k)=x(k)―w T (k)y(K),k=0,1,…,K-1; LMS adaptive filtering is performed to specifically filter out the noise in the signal, that is, the point-by-point approximation error [e(0), e(1),…, e(K-1)] is used as the filtering output.

5. The adaptive filtering method according to claim 1, characterized in that: The RLS adaptive filtering includes: Delete the sinusoidal component in the observed signal, that is, according to sin[x+tan ―1 (w2 / w1)] to determine w1 and w2, complete the estimation of the sinusoidal component in the observed signal, and subtract it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated; Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals; Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal; Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum overall gradient of the cost function with a fixed step μ, and obtain the sinusoidal interference signal: w(k+1)=w(k)+2μ(r yx ―R yy w(k)), where μ represents the step, R yy and r yx Respectively represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal; The RLS adaptive filtering is performed to specifically filter out the noise in the signal, that is, the sinusoidal interference signal is removed from the known measured signal as the filtering output.

6. A signal noise reduction system, characterized in that: include: Adaptive filtering module, which specifically filters out noise from the pre-noise-reduced signal; The processor module is configured to control the adaptive filtering module to perform corresponding adaptive filtering and obtain the optimization parameters of the adaptive filtering according to the signal pre-noise reduction result, that is, to execute the adaptive filtering method according to any one of claims 1 to 5.

7. An LMS adaptive filtering method for signal noise reduction, characterized in that: include: Delete the sinusoidal component in the observed signal, that is, according to sin[x+tan ―1 (w2 / w1)] determines 1 and 2, completes the estimation of the sinusoidal component in the observed signal, and subtracts it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated; Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals; Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal; Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum stochastic gradient of the cost function with a fixed step: w(k+1)=w(k)+2μ·e(k)y(k), where μ represents the step, e(k)=x(k)―w T (k)y(k),k=0,1,…,K-1; LMS adaptive filtering is performed to specifically filter out the noise in the signal, that is, the point-by-point approximation error [e(0), e(1),…, e(K-1)] is used as the filtering output.

8. A RLS adaptive filtering method for signal noise reduction, characterized in that: include: Delete the sinusoidal component in the observed signal, that is, according to sin[x+tan ―1 (w2 / w1)] determines 1 and 2, completes the estimation of the sinusoidal component in the observed signal, and subtracts it from the observed signal, where w1 and w2 are the weighting coefficients to be estimated; Determine the signal to be filtered, that is, according to the prior information y(k) = [sin2πfkT s ,cos2πfkT s ] T , where f is the filter frequency corresponding to suppressing interference signals; Determine the filtered signal, that is, let x(k) be the known measured signal, and let the sine wave w T (k)y(k) approximates x(k), and the approximation error is the filtered signal; Iterative approximation of weighted vector, that is, first set the initial value of weighted vector, that is, w(0) = [w1, w2] T , and then iteratively approximate the weighted vector w point by point in the direction of the maximum overall gradient of the cost function with a fixed step, and obtain the sinusoidal interference signal: w(k+1)=w(k)+2μ(r yx ―R yy w(k)), where μ represents the step, R yy and r yx Respectively represent the observed signal covariance matrix and the cross-correlation vector between the observed signal and the target signal; The RLS adaptive filtering is performed to specifically filter out the noise in the signal, that is, the sinusoidal interference signal is removed from the known measured signal as the filtering output.

Images