CN105933256A - Signal filtering method and system - Google Patents

Abstract

The invention relates to a signal filtering method and system. The method comprises the following steps that the output vector of a first interference signal is constructed according to the first interference signal generated by an interference source, and the output vector is inputted to a self-adaptive neural fuzzy inference system; the first fuzzy rules of each neuron node of the self-adaptive neural fuzzy inference system are set, the degree of membership of the output vector at each neuron node is calculated, and the incentive intensity of the first fuzzy rules is calculated according to the degree of membership; and the number of the neuron nodes is set according to the incentive intensity, the second fuzzy rules of the neuron nodes are set, a second interference signal is determined according to the output vector and the second fuzzy rules, and filtering of useful signals generated by the signal source is performed according to the second interference signal, wherein the second interference signal is the interference signal when the first interference signal is transmitted to the signal source of the useful signals.

Description

Method for filtering signals and system

Technical field

The present invention relates to signal processing technology field, particularly relate to a kind of method for filtering signals and system.

Background technology

Signal is in transmitting procedure, for various reasons, can produce interference signal.Interference signal can produce certain impact to useful signal, time serious, even makes signal to demodulate.Therefore, after receiving signal, generally require the docking collection of letters number and filter, remove interference, obtain useful signal.But, in some scenarios, it tends to be difficult to thoroughly by interference filtering.

As a example by electrocardiosignal, electrocardiosignal (Electrocardiograph, ECG) comprises substantial amounts of physiology even disease information, is the important tool of clinical diagnosis cardiovascular disease.The electrocardiosignal that patient monitor and many signs collecting device actual acquisition arrive is more weak, is easily polluted by multiple interference component, and such as Hz noise, myoelectricity (Electromyography, EMG) disturbs and drift about interference etc..Wherein, Hz noise is relatively fixed, and is easier to remove, but easily brings ST section the problem such as to raise, and affects final judgement；Myoelectricity interference then varies with each individual, and complicated component, it is difficult to remove.The amplitude often interfering with signal is higher, and both frequency aliasings.Therefore, finding robust performance good, the filtering method that separation efficiency is high is significant to extracting pure electrocardiosignal.

The complicated component of myoelectricity interference, the interference produced essentially from the muscular movement of the motion of human muscle group, especially arm, lower limb and thoracic cavity, as waved, chew, move, tremble and the action such as muscular tone.The position of muscular movement is different, and the impact on electrocardiosignal is the most different.At present the research method of myoelectricity interference in electrocardiosignal is included digital filtering, blind source separating method (independent component analysis, canonical correlation analysis), self-adaptive routing, adaptive neural network fuzzy system method.

Owing to the frequency range of ECG is 0～50Hz, the frequency range of EMG is 1～the frequency spectrum of 2000Hz, ECG and EMG intersection aliasing, it is difficult to by conventional digital filtering method by clean for EMG interference filtering.While myoelectricity interference minimizing technology filtering, some effective electrocardiosignal also can be removed；And need manual intervention to identify interference component, thus there is certain subjectivity and the most time-consuming.Blind source separating method is difficult to remove clean by myoelectricity interference, if removing too much myoelectricity component, then useful electrocardiosignal can be caused to lose.Adaptive filter method needs corresponding with reference to leading, thus there is also cross interference problem.The filter effect of adaptive nuero-fuzzy inference system system (Adaptive Neuro-Fuzzy Inference System, ANFIS) need to improve.

In sum, by the above-mentioned analysis done as a example by electrocardiosignal, it appeared that existing signal filtering technique exists cross interference problem, the effect of interference filtering is poor.

Summary of the invention

Based on this, it is necessary to for the problem that the effect of prior art interference filtering is poor, it is provided that a kind of method for filtering signals and system.

A kind of method for filtering signals, comprises the following steps:

According to the output vector of first interference signal structure the first interference signal that interference source produces, described output vector is input to Adaptive Neuro-fuzzy Inference；

First fuzzy rule of described each neuron node of Adaptive Neuro-fuzzy Inference is set, calculates the described output vector degree of membership at each neuron node, calculate the excitation density of described first fuzzy rule according to described degree of membership；

The quantity of neuron node is set according to described excitation density, second fuzzy rule of neuron node is set, determine the second interference signal according to described output vector and described second fuzzy rule, filter according to the useful signal that signal source is produced by described second interference signal；Wherein, described second interference signal is first interference signal interference signal when transmission arrives at the signal source of useful signal.

A kind of signal filtration system, including:

Input module, the output vector of first interference signal structure the first interference signal for producing according to interference source, described output vector is input to Adaptive Neuro-fuzzy Inference；

Computing module, for arranging the first fuzzy rule of described each neuron node of Adaptive Neuro-fuzzy Inference, calculates the described output vector degree of membership at each neuron node, calculates the excitation density of described first fuzzy rule according to described degree of membership；

Filtering module, for arranging the quantity of neuron node according to described excitation density, second fuzzy rule of neuron node is set, determine the second interference signal according to described output vector and described second fuzzy rule, filter according to the useful signal that signal source is produced by described second interference signal；Wherein, described second interference signal is first interference signal interference signal when transmission arrives at the signal source of useful signal.

Above-mentioned method for filtering signals and system, by the output vector of interference signal is input to Adaptive Neuro-fuzzy Inference, first fuzzy rule of described each neuron node of Adaptive Neuro-fuzzy Inference is set, calculate the excitation density of described first fuzzy rule, the quantity of neuron node is set according to described excitation density, second fuzzy rule of neuron node is set, the second interference signal is determined according to described output vector and described second fuzzy rule, and filter according to the described second useful signal disturbing signal that signal source is produced, can effectively obtain interference signal, it is thus possible to extract the purest useful signal from the useful signal being mixed with interference signal.

Accompanying drawing explanation

Fig. 1 is the method for filtering signals flow chart of the present invention；

Fig. 2 is normal electrocardiosignal schematic diagram；

Fig. 3 is the electrocardiosignal schematic diagram being mixed with electromyographic signal；

Fig. 4 is the structural representation of adaptive nuero-fuzzy inference system system；

Fig. 5 is the structural representation of the signal filtration system of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawings the embodiment of the method for filtering signals of the present invention is described.

Fig. 1 is the method for filtering signals flow chart of the present invention.As it is shown in figure 1, described method for filtering signals can comprise the following steps that

S1, according to the output vector of first interference signal structure the first interference signal that interference source produces, is input to Adaptive Neuro-fuzzy Inference by described output vector；

Illustrate as a example by electrocardiosignal and electromyographic signal below.When needing to extract pure electrocardiosignal from the electrocardiosignal being mixed with electromyographic signal, described first interference signal is electromyographic signal；Correspondingly, electrocardiosignal is useful signal.In practical situations both, the first interference signal and useful signal can also be other signals.Concrete signal type will not affect that the enforcement of technical scheme.

Normal its main feature of electrocardiosignal is to have narrow and high QRS wave, and the data beyond QRS wave are substantially at a metastable region, as shown in Figure 2.Its feature waveform of electrocardiosignal being mixed with myoelectricity interference is coarse, rough, causes the useful information contained in QRS to be difficult to extract, as shown in Figure 3.

Adaptive nuero-fuzzy inference system system (ANFIS) theory diagram is as shown in Figure 4.Wherein, s (k) is the preferable ECG signal not containing any interference or interference；Q (k) is main input source, is the ECG signal being mixed into the interference such as EMG, the ECG signal that i.e. electrode directly collects；F is nonlinear dynamic function, reflects human body and produces the path of electrocardioelectrode from myoelectricity interference；N (k) is noise reference input source, the i.e. first interference signal；D (k) is that EMG disturbs signal, the second interference signal when the i.e. first interference signal is at the signal source that transmission arrives useful signal；For the estimated value of d (k), pass throughOperation, obtains the estimated value of preferable ECG signal s (k)

The the first interference signal that can produce interference source postpone several times, the interference signal obtained with each delay as element, the output vector of structure the first interference signal.Such as, r-1 time can be carried out and postpone by described first interference signal through tapped delay line D, obtain output vector X (k)=[x of r dimension₁(k),x₂(k),…,x_r(k)]^T。

S2, arranges the first fuzzy rule of described each neuron node of Adaptive Neuro-fuzzy Inference, calculates the described output vector degree of membership at each neuron node, calculates the excitation density of described first fuzzy rule according to described degree of membership；

Output vector X (k)=[x₁(k),x₂(k),…,x_r(k)]^TEntering the 1st layer of wave filter, each node of this layer is equivalent to an one-dimensional membership function, can choose Gaussian function as membership function, specific as follows:

${\mathrm{\μ}}_{ij}\left(x_l\right)=\exp \[-\frac{{(x_l-c_{ij})}^2}{{\mathrm{\σ}}_j^2}\]---\left(1\right)$

Wherein, described wave filter can be sef-adapting filter, Chebyshev filter etc..The degree of membership that can obtain each neuron node according to above-mentioned membership function is:

$O_j^{\left(1\right)}=\underset{l=1}{\overset{r}{\mathrm{\Σ}}}{\mathrm{\μ}}_{lj}\left(x_l\right)---\left(2\right)$

In formula,For described output vector at the degree of membership of jth neuron node, μ_lj(x_l) it is the l element x in described output vector_lAt the membership function of jth neuron node, r is the dimension of described output vector, j=1,2 ..., n is the numbering of membership function, and wherein, n is the number of membership function, c_ljFor the l element in described output vector at the center of the membership function of jth neuron node, σ_jWidth for the membership function of jth neuron node.In practical situations both, it is possible to choose other functions as membership function, such as trigonometric function, chebyshev function etc..Through reality test, use Gaussian function to be obtained in that filter effect more more preferable than other functions as membership function, clinical data behaves oneself best.

As the input of the 2nd layer, can can calculate the excitation density of every fuzzy rule according to described degree of membership, when membership function is Gaussian function, described excitation density can be:

$O_j^{\left(2\right)}=\underset{l=1}{\overset{r}{\mathrm{\Π}}}{\mathrm{\μ}}_{lj}\left(x_l\right)=\exp \[-\frac{\left|\right|{(x_l-c_{lj})}^2\left|\right|}{{\mathrm{\σ}}_j^2}\]---\left(3\right)$

In formula,For the excitation density of j-th strip fuzzy rule, μ_lj(x_l) it is the l element x in described output vector_lAt the membership function of jth neuron node, r is the dimension of described output vector, and n is the number of membership function, c_ljFor the l element in described output vector at the center of the membership function of jth neuron node, σ_jWidth for the membership function of jth neuron node.In practical situations both, other membership functions, such as, chebyshev function etc. can also be used.

S3, the quantity of neuron node is set according to described excitation density, second fuzzy rule of neuron node is set, determines the second interference signal according to described output vector and described second fuzzy rule, filter according to the useful signal that signal source is produced by described second interference signal；Wherein, described second interference signal is first interference signal interference signal when transmission arrives at the signal source of useful signal.

In the prior art, typically willAs the input of the 3rd layer, calculated the proportionality coefficient of each rule by equation below, the ratio of the excitation density sum of its excitation density being output as j-th strip rule and strictly all rules:

$O_j^{\left(3\right)}=\frac{O_j^{\left(2\right)}}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{O}_i^{\left(2\right)}}---\left(4\right)$

In formula,For the excitation density of j-th strip fuzzy rule,WithBeing respectively i-th fuzzy rule and j-th strip fuzzy rule, n is the quantity of fuzzy rule.

Finally, at the 4th layer, by equation below, the output of strictly all rules is sued for peace, calculates and always export y:

$y=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{O}_j^{\left(3\right)}---\left(5\right)$

In order to improve signal strainability, the design by adaptive-filtering on-line learning algorithm of this step, realize adaptive-filtering the most exactly, neuron number for the adaptive neural network shown in Automatic adjusument step 2, it is achieved the on-line tuning of filter construction and the on-line study of algorithm.

Specifically, the excitation density of each fuzzy rule can be compared with the first threshold preset, if the excitation density of all fuzzy rules is all not less than described first threshold, described Adaptive Neuro-fuzzy Inference build new neuron node.After building new neuron node, available least-squares estimation (Least Squares Estimate, LSE) adjusts fuzzy rule consequent parameter (i.e. c described in formula 3_ljAnd σ_j).Owing to having increased neuron newly, adjusting parameter with greater need for Parameter Estimation Method accurately, LSE algorithm improves the degree of accuracy of parameter estimation.Also can delete neuron node adaptively, such as, can be according to ERR (Error Reduction Ratio, error attenuated rate) criterion deletes neuron node adaptively, specifically, error fall off rate and the weight of each neuron can be calculated, described error fall off rate is compared with the Second Threshold preset, described weight is compared with the 3rd threshold value preset, error fall off rate is exceeded described Second Threshold, or weight exceedes the neuron deletion of described 3rd threshold value.After deleting neuron node, available recursive least-squares (Recursive Least Squares, RLS) adjusts fuzzy rule consequent parameter (i.e. c described in formula 3_ljAnd σ_j).RLS is without storing all data, and algorithm amount of calculation is little, and real-time is high.In other embodiments, also can adjust fuzzy rule consequent parameter according to other modes, such as, steepest descent algorithm can be used.

Wherein, described first threshold can be arranged according to equation below:

F_gen=min [F_minδ^-i,F_min] (6)

In formula, F_genFor described first threshold, F_minFor default excitation intensity threshold, δ ∈ (0,1) is attenuation constant, and i represents the number of Current neural unit.

Specifically, the correlation coefficient of described second interference signal and described useful signal, the degree coefficient filtered according to described correlation coefficient checking signal can be calculated；Wherein, described degree coefficient is for characterizing the degree that signal filters.This correlation coefficient embodies ECG signal and the degree of relevancy of myoelectricity before and after filtering, thus can quantitative analysis myoelectricity interference filter degree.In one embodiment, the useful signal before described second interference signal can being calculated according to equation below and filter or the correlation coefficient of filtered useful signal:

${\mathrm{\ρ}}_{xy}=\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{x}_n\left(t\right)y_n\left(t\right)/{\[\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{x}_n^2\left(t\right)\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{y}_n^2\left(t\right)\]}^{1/2}---\left(7\right)$

In formula, ρ_xyFor described correlation coefficient, n is sample number, x_nT () is described second interference signal, y_nT () is the useful signal before filtering the second interference signal or the useful signal after filtering the second interference.

In order to verify the performance of adaptive filter algorithm designed by the present invention, the dependency of myoelectricity interference and ECG signal before and after filtering is compared, result such as following table:

Table 1 ECG signal and the dependency of EMG interference

Through the pure electrocardiosignal more that above-mentioned filter filtering obtains, on the one hand can show on many sign device of electrocardio module, monitor equipment comprising, as individual or doctor's detection, the basis of diagnosis；On the other hand, data basis accurately and reliably is provided for ensuing rate calculation, arrhythmia analysis.

With existing method is compared, the invention have the advantages that

(1) online adaptive adjusts FL-network structure, it is possible to responds change real-time, and makes corresponding reply for change.

(2) relative to fixing network structure and the wave filter of parameter, it is possible to the information for each moment makes corresponding filtering, it is possible to farthest retain effective information, filtering interfering, it is ensured that the reliability of filtered data and accuracy.

(3) on-line learning algorithm is more suitable for being actually needed, and more disclosure satisfy that the requirement of engineer applied.

(4) evaluation criterion quantitative to signal filter effect is provided.

Above-mentioned electrocardiosignal and electromyographic signal are one embodiment of the present of invention, and the method for filtering signals of the present invention is not limited to above-mentioned embodiment electrocardiosignal filtered out from electromyographic signal, hereby illustrates.

Below in conjunction with the accompanying drawings the embodiment of the signal filtration system of the present invention is described.

Fig. 5 is the structural representation of the signal filtration system of the present invention.As it is shown in figure 5, described signal filtration system comprises the steps that

Input module 10, the output vector of first interference signal structure the first interference signal for producing according to interference source, described output vector is input to Adaptive Neuro-fuzzy Inference；

Illustrate as a example by electrocardiosignal and electromyographic signal below.When needing to extract pure electrocardiosignal from the electrocardiosignal being mixed with electromyographic signal, described first interference signal is electromyographic signal；Correspondingly, electrocardiosignal is useful signal.In practical situations both, the first interference signal and useful signal can also be other signals.Concrete signal type will not affect that the enforcement of technical scheme.

Normal electrocardiosignal is as in figure 2 it is shown, its main feature is to have narrow and high QRS wave, and the data beyond QRS wave are substantially at a metastable region.Be mixed with myoelectricity interference electrocardiosignal as it is shown on figure 3, its feature waveform is coarse, rough, cause the useful information contained in QRS to be difficult to extract.

Adaptive nuero-fuzzy inference system system (ANFIS) theory diagram is as shown in Figure 4.Wherein, s (k) is the preferable ECG signal not containing any interference or interference；Q (k) is main input source, is the ECG signal being mixed into the interference such as EMG, the ECG signal that i.e. electrode directly collects；F is nonlinear dynamic function, reflects human body and produces the path of electrocardioelectrode from myoelectricity interference；N (k) is noise reference input source, the i.e. first interference signal；D (k) is that EMG disturbs signal, the second interference signal when the i.e. first interference signal is at the signal source that transmission arrives useful signal；For the estimated value of d (k), pass throughOperation, obtains the estimated value of preferable ECG signal s (k)

The the first interference signal that can produce interference source postpone several times, the interference signal obtained with each delay as element, the output vector of structure the first interference signal.Such as, r-1 time can be carried out and postpone by described first interference signal through tapped delay line D, obtain output vector X (k)=[x of r dimension₁(k),x₂(k),…,x_r(k)]^T。

Computing module 20, for arranging the first fuzzy rule of described each neuron node of Adaptive Neuro-fuzzy Inference, calculates the described output vector degree of membership at each neuron node, calculates the excitation density of described first fuzzy rule according to described degree of membership；

Output vector X (k)=[x₁(k),x₂(k),…,x_r(k)]^TEntering the 1st layer of wave filter, each node of this layer is equivalent to an one-dimensional membership function, can choose Gaussian function as membership function, specific as follows:

${\mathrm{\μ}}_{ij}\left(x_l\right)=\exp \[-\frac{{(x_l-c_{ij})}^2}{{\mathrm{\σ}}_j^2}\]---\left(1\right)$

Wherein, described wave filter can be sef-adapting filter etc..The degree of membership that can obtain each neuron node according to above-mentioned membership function is:

$O_j^{\left(1\right)}=\underset{l=1}{\overset{r}{\mathrm{\Σ}}}{\mathrm{\μ}}_{lj}\left(x_l\right)---\left(2\right)$

In formula,For described output vector at the degree of membership of jth neuron node, μ_lj(x_l) it is the l element x in described output vector_lAt the membership function of jth neuron node, r is the dimension of described output vector, j=1,2 ..., n is the numbering of membership function, and wherein, n is the number of membership function, c_ljFor the l element in described output vector at the center of the membership function of jth neuron node, σ_jWidth for the membership function of jth neuron node.In practical situations both, it is possible to choose other functions as membership function, such as trigonometric function, chebyshev function etc..Through reality test, use Gaussian function to be obtained in that filter effect more more preferable than other functions as membership function, clinical data behaves oneself best.

As the input of the 2nd layer, can can calculate the excitation density of every fuzzy rule according to described degree of membership, when membership function is Gaussian function, described excitation density can be:

$O_j^{\left(2\right)}=\underset{l=1}{\overset{r}{\mathrm{\Π}}}{\mathrm{\μ}}_{lj}\left(x_l\right)=\exp \[-\frac{\left|\right|{(x_l-c_{lj})}^2\left|\right|}{{\mathrm{\σ}}_j^2}\]---\left(3\right)$

In formula,For the excitation density of j-th strip fuzzy rule, μ_lj(x_l) it is the l element x in described output vector_lAt the membership function of jth neuron node, r is the dimension of described output vector, and n is the number of membership function, c_ljFor the l element in described output vector at the center of the membership function of jth neuron node, σ_jWidth for the membership function of jth neuron node.In practical situations both, other membership functions, such as, chebyshev function etc. can be used, use which kind of membership function can't affect the enforcement of the application subsequent step.

Filtering module 30, for arranging the quantity of neuron node according to described excitation density, second fuzzy rule of neuron node is set, determine the second interference signal according to described output vector and described second fuzzy rule, filter according to the useful signal that signal source is produced by described second interference signal；Wherein, described second interference signal is first interference signal interference signal when transmission arrives at the signal source of useful signal.

In the prior art, typically willAs the input of the 3rd layer, calculated the proportionality coefficient of each rule by equation below, the ratio of the excitation density sum of its excitation density being output as j-th strip rule and strictly all rules:

$O_j^{\left(3\right)}=\frac{O_j^{\left(2\right)}}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{O}_i^{\left(2\right)}}---\left(4\right)$

In formula,For the excitation density of j-th strip fuzzy rule,WithBeing respectively i-th fuzzy rule and j-th strip fuzzy rule, n is the quantity of fuzzy rule.

Finally, at the 4th layer, by equation below, the output of strictly all rules is sued for peace, calculates and always export y:

$y=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{O}_j^{\left(3\right)}---\left(5\right)$

In order to improve signal strainability, the filtering module 30 design by adaptive-filtering on-line learning algorithm, it is achieved adaptive-filtering the most exactly.On-line learning algorithm flow chart is as it is shown in figure 5, be used for the neuron number of the adaptive neural network shown in Automatic adjusument computing module 20, it is achieved the on-line tuning of filter construction and the on-line study of algorithm.

Specifically, the excitation density of each fuzzy rule can be compared with the first threshold preset, if the excitation density of all fuzzy rules is all not less than described first threshold, described Adaptive Neuro-fuzzy Inference build new neuron node.After building new neuron node, available least-squares estimation (Least Squares Estimate, LSE) adjusts fuzzy rule consequent parameter (i.e. c described in formula 3_ljAnd σ_j).Also can delete neuron node adaptively, such as, neuron node can be deleted adaptively according to ERR criterion, specifically, error fall off rate and the weight of each neuron can be calculated, described error fall off rate is compared with the Second Threshold preset, described weight is compared with the 3rd threshold value preset, error fall off rate is exceeded described Second Threshold, or weight exceedes the neuron deletion of described 3rd threshold value.After deleting neuron node, available recursive least-squares (Recursive Least Squares, RLS) adjusts fuzzy rule consequent parameter (i.e. c described in formula 3_ljAnd σ_j).In other embodiments, also can adjust fuzzy rule consequent parameter according to other modes, such as, steepest descent algorithm can be used.

Wherein, described first threshold can be arranged according to equation below:

F_gen=min [F_minδ^-i,F_min] (6)

In formula, F_genFor described first threshold, F_minFor default excitation intensity threshold, δ ∈ (0,1) is attenuation constant, and i represents the number of Current neural unit.

Specifically, the correlation coefficient of described second interference signal and described useful signal, the degree coefficient filtered according to described correlation coefficient checking signal can be calculated；Wherein, described degree coefficient is for characterizing the degree that signal filters.This correlation coefficient embodies ECG signal and the degree of relevancy of myoelectricity before and after filtering, thus can quantitative analysis myoelectricity interference filter degree.In one embodiment, can be according to the correlation coefficient of the described second interference signal of equation below calculating with described useful signal:

${\mathrm{\ρ}}_{xy}=\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{x}_n\left(t\right)y_n\left(t\right)/{\[\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{x}_n^2\left(t\right)\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{y}_n^2\left(t\right)\]}^{1/2}---\left(7\right)$

In formula, ρ_xyFor described correlation coefficient, n is sample number, x_nT () is described second interference signal, y_nT () is the useful signal before filtering the second interference signal or the useful signal after filtering the second interference.

In order to verify the performance of adaptive filter algorithm designed by the present invention, the dependency of myoelectricity interference and ECG signal before and after filtering is compared, result such as following table:

Table 1 ECG signal and the dependency of EMG interference

Through the pure electrocardiosignal more that above-mentioned filter filtering obtains, on the one hand can show on many sign device of electrocardio module, monitor equipment comprising, as individual or doctor's detection, the basis of diagnosis；On the other hand, data basis accurately and reliably is provided for ensuing rate calculation, arrhythmia analysis.

With existing method is compared, the invention have the advantages that

(1) online adaptive adjusts FL-network structure, it is possible to responds change real-time, and makes corresponding reply for change.

(2) relative to fixing network structure and the wave filter of parameter, it is possible to the information for each moment makes corresponding filtering, it is possible to farthest retain effective information, filtering interfering, it is ensured that the reliability of filtered data and accuracy.

(3) on-line learning algorithm is more suitable for being actually needed, and more disclosure satisfy that the requirement of engineer applied.

(4) evaluation criterion quantitative to signal filter effect is provided.

Above-mentioned electrocardiosignal and electromyographic signal are one embodiment of the present of invention, and the signal filtration system of the present invention is not limited to above-mentioned embodiment electrocardiosignal filtered out from electromyographic signal, hereby illustrates.

The signal filtration system of the present invention and the method for filtering signals one_to_one corresponding of the present invention, technical characteristic that the embodiment at above-mentioned method for filtering signals illustrates and beneficial effect thereof, all be applicable to the embodiment of signal filtration system, hereby give notice that.

Each technical characteristic of embodiment described above can combine arbitrarily, for making description succinct, the all possible combination of each technical characteristic in above-described embodiment is not all described, but, as long as the combination of these technical characteristics does not exist contradiction, all it is considered to be the scope that this specification is recorded.

Embodiment described above only have expressed the several embodiments of the present invention, and it describes more concrete and detailed, but can not therefore be construed as limiting the scope of the patent.It should be pointed out that, for the person of ordinary skill of the art, without departing from the inventive concept of the premise, it is also possible to make some deformation and improvement, these broadly fall into protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (10)

1. a method for filtering signals, it is characterised in that comprise the following steps:

According to the output vector of first interference signal structure the first interference signal that interference source produces, by described defeated Outgoing vector is input to Adaptive Neuro-fuzzy Inference；

First fuzzy rule of described each neuron node of Adaptive Neuro-fuzzy Inference is set, calculates Described output vector, in the degree of membership of each neuron node, calculates described first according to described degree of membership and obscures The excitation density of rule；

The quantity of neuron node is set according to described excitation density, the second fuzzy rule of neuron node are set Then, the second interference signal is determined, according to described second according to described output vector and described second fuzzy rule The useful signal that signal source is produced by interference signal filters；Wherein, described second interference signal is first Interference signal when interference signal is at the signal source that transmission arrives useful signal.

Method for filtering signals the most according to claim 1, it is characterised in that produce according to interference source The step of the output vector of first interference signal structure the first interference signal includes:

The the first interference signal producing interference source postpones several times；

The interference signal obtained with each delay as element, the output vector of structure the first interference signal.

Method for filtering signals the most according to claim 1, it is characterised in that calculate described output vector Step in the degree of membership of each neuron node of Adaptive Neuro-fuzzy Inference ground floor includes:

According to the equation below described degree of membership of calculating:

$O_j^{\left(1\right)}=\underset{l=1}{\overset{r}{\Σ}}{\mathrm{\μ}}_{lj}\left(x_l\right);$

Wherein,

In formula,For described output vector at the degree of membership of jth neuron node, μ_lj(x_l) it is described output The l element x in vector_lAt the membership function of jth neuron node, l=1,2 ..., r is described output vector Dimension, j=1,2 ..., n is the number of membership function, c_ljFor the l element in described output vector in jth The center of the membership function of neuron node, σ_jWidth for the membership function of jth neuron node.

Method for filtering signals the most according to claim 3, it is characterised in that described excitation density is:

$O_j^{\left(2\right)}=\underset{l=1}{\overset{r}{\Π}}{\mathrm{\μ}}_{lj}\left(x_l\right)=\exp \[-\frac{\left|\right|{(x_l-c_{lj})}^2\left|\right|}{{\mathrm{\σ}}_j^2}\];$

In formula,For the excitation density of j-th strip fuzzy rule, μ_lj(x_l) it is the l unit in described output vector Element x_lAt the membership function of jth neuron node, r is the dimension of described output vector, and n is degree of membership letter The number of number, c_ljFor the l element in described output vector at the membership function of jth neuron node Center, σ_jWidth for the membership function of jth neuron node.

Method for filtering signals the most according to claim 1, it is characterised in that according to described excitation density The step of the quantity adjusting the neuron node of described Adaptive Neuro-fuzzy Inference includes:

The excitation density of each fuzzy rule is compared with the first threshold preset；

If the excitation density of all fuzzy rules is all not less than described first threshold, at described adaptive neural network mould Stick with paste in inference system and build new neuron node.

Method for filtering signals the most according to claim 5, it is characterised in that by each fuzzy rule Excitation density with preset first threshold compare before, further comprising the steps of:

According to equation below, first threshold is set:

F_gen=min [F_minδ^-i,F_min]；

In formula, F_genFor described first threshold, F_minFor default excitation intensity threshold, δ ∈ (0,1) is for decay often Number, i represents the number of Current neural unit.

Method for filtering signals the most according to claim 1, it is characterised in that according to described excitation density The step of the quantity adjusting the neuron node of described Adaptive Neuro-fuzzy Inference also includes:

Calculate error fall off rate and the weight of each neuron；

Described error fall off rate is compared with the Second Threshold preset, by described weight and the preset Three threshold values compare；

Error fall off rate is exceeded described Second Threshold, or weight exceedes the neuron of described 3rd threshold value Delete.

Method for filtering signals the most according to claim 1, it is characterised in that further comprising the steps of:

Calculate the correlation coefficient of described second interference signal and described useful signal；

The degree coefficient filtered according to described correlation coefficient checking signal；Wherein, described degree coefficient is used for table The degree that reference number filters.

Method for filtering signals the most according to claim 8, it is characterised in that calculate described second interference Signal includes with the step of the correlation coefficient of described useful signal:

Correlation coefficient according to the described second interference signal of equation below calculating with described useful signal:

${\mathrm{\ρ}}_{xy}=\underset{n=0}{\overset{\mathrm{\∞}}{\Σ}}{x}_n\left(t\right)y_n\left(t\right)/{\[\underset{n=0}{\overset{\mathrm{\∞}}{\Σ}}{x}_n^2\left(t\right)\underset{n=0}{\overset{\mathrm{\∞}}{\Σ}}{y}_n^2\left(t\right)\]}^{1/2};$

In formula, ρ_xyFor described correlation coefficient, n is sample number, x_nT () is described second interference signal, y_n(t) be Useful signal after filtering the useful signal before the second interference signal or filtering the second interference.

10. a signal filtration system, it is characterised in that including:

Input module, the output of first interference signal structure the first interference signal for producing according to interference source Vector, is input to Adaptive Neuro-fuzzy Inference by described output vector；

Computing module, for arranging the first of described each neuron node of Adaptive Neuro-fuzzy Inference Fuzzy rule, calculates the described output vector degree of membership at each neuron node, according to described degree of membership meter Calculate the excitation density of described first fuzzy rule；

Filtering module, for arranging the quantity of neuron node according to described excitation density, arranges neuron joint Second fuzzy rule of point, determines the second interference signal according to described output vector and described second fuzzy rule, Filter according to the useful signal that signal source is produced by described second interference signal；Wherein, described second dry Disturbing signal is first interference signal interference signal when transmission arrives at the signal source of useful signal.