JPH0795968A - Noise removal method and apparatus - Google Patents

Abstract

(57) [Abstract] [Purpose] When a signal is detected at multiple measurement points for a common signal source, a noise component whose frequency band overlaps with the original signal component of the signal source is distorted to the original signal. Provided is a noise removal method that can be removed without any problem. [Configuration] In each measurement point, by detecting the detection signal V _IN at one less than the least that number, and V _IN of any measuring point, specific amplitude relationship between V _IN of the remaining measuring points The parameter α _ij that defines is calculated. Detecting a fundamental component signal V in V _IN, by subtracting V from V _IN to generate a residual signal V _r. The comparison signal U corresponding to V _r at any measurement point is calculated from the calculation of V _{r at the} remaining measurement point and the parameter α _ij . The noise index in V _r is calculated by comparing V _r and U, and the original signal component therein is detected by reducing the signal amplitude of V _r at any measurement point according to the noise index. By adding this original signal component and V, the original signal from which the mixed noise is removed is reproduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention eliminates noise that is mixed with a detection signal and is generated at a place other than the signal source when an original signal generated from a common signal source is detected at a plurality of measurement points. The present invention relates to a noise removing method and device for performing the above.

[0002]

2. Description of the Related Art For example, in an electrocardiograph using a heart as a signal source,
Electrocardiogram signals evoked by the heart are detected on multiple channels by attaching electrodes to the chest, left and right hands and feet. When the noise mixed in the detection signal and having the frequency domain common to the electrocardiogram signal is removed, the electrocardiogram signal is distorted and removed by the filter.

[0003]

In summary, it is generally a current challenge to remove noise in the frequency domain of an electrical signal without distortion. Incidentally, an addition and averaging method for canceling noise by averaging a plurality of same signals by aligning the phases and multiply-adding the signals and averaging is well known, but it is not applied to signals that cannot periodically become the same signal. Therefore,
For example, PVC (premature ventri) mixed in the electrocardiogram signal
cular contraction), heart fibrillation or other arrhythmias are difficult to remove.

In view of the above points, the present invention is directed to detecting signals at a plurality of measurement points with respect to a common signal source.
In a system in which at least approximately a linear or non-linear functional relationship is maintained with respect to a common signal source, a noise component whose frequency band overlaps with the original signal component of the signal source can be removed without distorting the original signal. An object of the present invention is to provide a noise removing method and device.

[0005]

When the original signal is detected at a plurality of locations (N channels) with respect to a common signal source, the detection signal V _i of an arbitrary channel is a unique amplitude relation between the channels, that is, It can be expressed as the following Expression 1 by a linear combination of the detection signals V _j of other channels, which are weighted by the parameter α _ij defining the distribution ratio. This method satisfies the condition that the distributed signal source is approximated by a linear or non-linear model with respect to the distributed signal, and that the number of measurement points (the number of channels) is sufficient to determine the parameters of the model.
For example, if the object of measurement is modeled for a linear system with a maximum number L of linear independent variables, the number of channels N is N> L to determine the parameters of that model.
Conditions are required. In fact, the majority of measurement systems
This condition is satisfied.

[0006]

[Equation 1]

Learning data V _i (j) (i = 1, 2, ..., N and j = 1, 2, ..., M, i is a channel, j Is the sampling time point). Using these learning data, Equation 1 is applied to an arbitrary channel i to obtain a simultaneous equation according to Equation 2 below.

[0008]

[Equation 2]

Α _ij is calculated as an M-element simultaneous linear equation under the condition of M + 1 ≧ N on the assumption of the well-known principle of the least squares method in linear algebra. That is, α _ij
Is calculated, the detection signal V _{i of} each channel is calculated.
Is calculated from the detection signals V _j of the remaining channel groups.

Similarly, in a system in which a non-linear functional relationship is maintained at a plurality of measurement points with respect to a common signal source, the intrinsic amplitude relationship between the channels is assumed on the assumption of the principle of the non-linear least squares method. The parameter α _ij that defines is calculated.

The present invention has been made in view of the fact that the difference between a signal calculated by using this algebraic arithmetic method and an actual signal can be used for noise discrimination, and the above-mentioned object is achieved as follows. That is, when detecting an original signal generated from a common signal source at a plurality of measurement points, in a noise reduction method for reducing noise mixed in the detection signal, M number of signals derived from the detection signal are detected. By sampling at the sampling time point, a parameter for weighting the detection signal of an arbitrary measurement point as a linear combination of the detection signals of the remaining measurement points is calculated (where M + 1 is the number of measurement points or more), and the detected signal is detected. Detect the basic component signal of, generate the residual signal by subtracting the basic component signal from the detected signal, add the residual signal of the remaining measurement point weighted by the parameter, and add the residual error of the arbitrary measurement point. A comparison signal corresponding to the signal is calculated, and the noise index that reflects the degree of noise contamination included in the detection signal at each sampling time at any measurement point is compared with the comparison signal at each sampling time and the residual error. Calculated by comparing the items, by reducing in accordance with the signal amplitude of residual signal of an arbitrary measurement point to noise index, to detect a residual signal Nakanoharu signal component,
By adding the original signal component and the basic component signal, the original signal with reduced mixed noise at any measurement point is reproduced.

[0012]

The parameter α _ij at an arbitrary measurement point for the common signal source is obtained in advance in a state where the noise is low or the noise is reduced by the conventional method. When a basic component signal that is close to the original signal and contains almost no noise is separated from the detection signals detected at a plurality of measurement points, a residual signal remains.
Since this residual signal contains the original signal component obtained by removing the basic component signal from the original signal and noise, the signal-to-noise ratio becomes low, and the noise level becomes relatively high, making noise identification easy. Become. The original signal component in the residual signal has a unique amplitude relationship because it is a part of the original signal. That is, the original signal component contained in the residual signal is reproduced according to the equation 1, and the noise generated on the one side other than the signal source does not follow the equation 1, and therefore the comparison signal obtained from the residual signal changes according to the noise. . Therefore, the comparison signal obtained from the residual signal of the remaining measurement points using the parameter α _ij for any measurement point is reproduced as it is with respect to the original signal component, and the noise component is reconstructed with different amplitudes. To be done.

As a result, the residual signal in which the noise at an arbitrary measurement point is relatively large is compared with the comparison signal, and the noise index reflecting the degree of noise contamination is calculated from the magnitude of the amplitude difference. Ask. The larger the noise index is, the larger the signal amplitude of the residual signal is reduced, so that the noise in the residual signal is compressed and the original signal component is detected.

Therefore, by adding this original signal component to the pre-separated basic component signal, the original signal in the detection signal at an arbitrary measurement point is in a state in which noise having a frequency domain in common with this original signal is removed. Will be reproduced faithfully. That is, the noise from the noise source at a location other than the signal source is reduced because it has a different relationship from the calculated inherent amplitude relationship between the plurality of measurement points with respect to the signal source.

[0015]

1 is a functional block diagram showing the basic configuration of an eight-channel electrocardiograph to which the present invention is applied. In the figure,
The signal source 1 is the heart, and detects an 8-channel electrocardiogram signal, which uses the signals of the right hand (RA), the left hand (LA), and the chest C1 to C6 as the original signals. Mathematical models of the heart have been studied for a long time, and it has been found that a linear model of the heart with up to several 7 independent parameters can detect an electrocardiogram with high accuracy.

Reference numeral 2 denotes a parameter calculating means, which is unique to the detection signal V _IN of an arbitrary channel and the detection signal V _IN of the remaining channels by sampling at M (M + 1 ≧ N) sampling points of the detection signal. A parameter calculation that defines the amplitude relationship is performed and stored. This parameter is a detection signal V _IN ⁱ (i = 1, 2 ...
, N) are detected signals V _IN ^j (i = 1,
2 ........., N, j ≠ i) are weighted linear combinations. For example, as shown in FIG. 2, with respect to an R wave having a sufficiently large SN ratio, a sampling signal is detected for each channel at least at time points t _{1 to} t ₇ , so that 8 channels are detected from the detection signal V _IN at time point 7. Assuming that it is a linear system, α _ij
To calculate.

Reference numeral 3 denotes a basic component signal detecting means for detecting a basic component signal in the detection signal, which is constituted by a low pass filter having no phase shift so that the band 0.
For example, the band 12H in the detection signal V _{IN of} 05 to 100 Hz
The basic component signal V equal to or less than z is detected. 4 is the detection signal V _IN
Is a subtraction means for subtracting the basic component signal V from and generating a residual signal V _r .

Reference numeral 5 is a comparison signal calculation means for generating a comparison signal U corresponding to the residual signal V _r at each measurement point from the calculation of the residual signal V _r of the remaining 7 channels and the parameter α _ij , for example, 1ch. Comparison signal U _{1 with} respect to the residual signal V _r1 of
From the residual signals V _{r2 to} V _r8 of the remaining channels,
Calculate according to. The comparison signals U _{2 to} U ₈ are similarly calculated for the remaining channels.

U ₁ = α ₁₂ V _r2 + α ₁₃ V _r3 …… + α ₁₈ V _r8 …… (3)

Reference numeral 6 is a noise index calculating means for comparing the residual signal V _r of an arbitrary channel with the comparison signal U to calculate a noise index in the residual signal V _r . This noise index is calculated as coherence (coherence) C by the following equation 4 from the amplitudes of the actual residual signal V _r of each channel and the associated comparison signal U, for example.

[0021]

[Equation 3]

This coherence factor C is also used in physics and reflects the degree of coincidence of both signals V _r , U. In other words, when the residual signal V _r does not include noise, the comparison signal U has an inherent amplitude relationship with the signal source, and therefore the residual signal V _r
It becomes equal to _r and C = 1. On the contrary, when the amount of noise is larger than that of the original signal component, the calculated U approaches O because it does not follow the inherent amplitude relationship, and thus C changes between 0 and 1.

Reference numeral 7 is a noise removing means for detecting the original signal component in the residual signal V _r by reducing the signal amplitude of the residual signal V _{r of} an arbitrary channel according to the noise index.
For example, the residual signal V _{r1 to} V _{r8 of} each channel is sequentially multiplied by the associated noise index C to detect the original signal component of each channel.

Reference numeral 8 denotes an adding means for adding the basic component signal V and the original signal component in the residual signal V _r to reproduce the original signal from which the mixed noise of any channel is removed.
The reproduced original signal is sent to a processing means for the purpose of the apparatus, for example, processing means such as derivation of standard 12-lead to display a waveform.

At the time of measurement, the R wave of each channel is detected in advance, the baseline levels are made uniform, and α _ij is calculated in advance from the amplitude data at the sampling points of t ₁ to t ₇ (FIG. 2).
The table below shows α _ij (i = 1, 2 ...
.. 8; j = 1, 2, ...

[0026]

[Table 1]

When the measurement is started, the detection signal V _IN (FIG. 3A) of each channel is detected, the basic component signal detection means 3 detects the basic component signal V (FIG. 3B), and the subtraction means 4 is detected. A residual signal V _r (FIG. 3C) is generated. In this case, the residual signal V _r contains a relatively large amount of noise.

The comparison signal calculation means 5, with respect to the residual signal V _r of each channel, computes the comparison signal (FIG. 4D) from the rest of the channel residual signal V _r to the affiliation of alpha _ij. Since the noise level of the comparison signal does not depend on the calculated α _ij , it has a different amplitude over all channels from that of the residual signal V _r .

The noise index calculation means 6 successively calculates C and multiplies it by the residual signal V _r to detect the original signal component (FIG. 4E) in which noise is reduced. The adding means 8 adds the original signal component and the basic signal component V to reproduce the electrocardiogram waveform signal (FIG. 4F) of each channel.

Note that the noise index calculating means 6 has a characteristic that noise has a statistical characteristic, and therefore, in consideration of a temporal statistical effect, at each sampling time point n at an arbitrary measurement point,
Comprehensive evaluation values for C at a plurality of consecutive sampling points, for example, summing up the noise index C (n) by using the V _ri and U _i values at w sampling points before and after the sequential calculation by Equation 5 below. The noise removal accuracy is further improved.

[0031]

[Equation 4]

Furthermore, the noise of the comparison signal is redistributed among the channels by the addition of the weighted V _r of the different channels, based on the averaged common noise over all channels at the sampling instant n. As an index, C (n) can be calculated according to the following equation 6 by the values of V _r and U of all channels. The noise reduction degree is improved as compared with Expression 4.

[0033]

[Equation 5]

Given the time and space effects of noise distribution, C (n) is calculated over time window through all channels according to Equation 7 below. Similarly, the noise reduction degree is improved as compared with Equation 4.

[0035]

[Equation 6]

Further, in order to enhance the effect of the noise index as the noise removing means 7, a function f given by the following equation 8 is added to C.
By using the coefficient C · f (c) multiplied by (c), a large noise removal effect can be obtained.

[0037]

[Equation 7]

That is, in the function f (C) of C shown in FIG. 5A, b
A better effect can be obtained when = 1. This is because the relationship between the noise and the noise index is essentially statistical even if the noise index directly reflects the noise level in a certain range. As can be seen from the figure, if C is a high value close to 1 (in this case, the probability of noise contamination is low), then V _r is reduced to avoid signal distortion as much as possible. On the other hand, if C has a relatively low value (when the probability of noise presence is high), f
The value of (C) rapidly approaches 0 to make the denoising effect more effective.

For an intermittent signal, such as an electrocardiogram signal, the noise index for the baseline (in which case there is no signal) is given by f in equation 9 below to improve the noise removal effect.
(C) can be used. Where C _b is the baseline V _r
And U are used and pre-computed in the learning phase together with α _ij . For electrocardiographic signals, it is measured as the average value of the sampling time points over the 40 ms period of the baseline. FIG. 5B shows the characteristics when C _b is set to 0.5 and b is set to 3.

[0040]

[Equation 8]

FIG. 6 shows a conventional 12-lead type electrocardiograph in which the present invention is implemented by using a microcomputer. As is well known, the microcomputer 14 has eight channels of signals for the right hand (RA), the left hand (LA) and the chests V1 to V6, the average signal of the averaging circuit 18 for the RA, LA and left foot (LL) leads and the right foot. The preamplifier 10 differentially amplifies the potential via the feedback circuit 19, the sampling and holding circuit 11 performs sampling and holding, the multiplexer 12 sequentially selects it, and the A / D converter 13 digitizes and supplies the data. To be done. Then, the built-in program learns the parameters at the time of measurement and calculates the detection signal at the time of measurement based on the parameters. Further, for the standard 12-lead, each lead signal is calculated from the above-mentioned 8-channel electrocardiographic signals detected by the calculation processing.

Reference numeral 15 is a display device for displaying each waveform of 12-lead, 16 is a recorder thereof, and these respective parts are mutually connected by a data and control bus 17.

The operation of the electrocardiograph configured as described above is shown in FIG.
This will be described with reference to the flowchart shown in FIG.

The electrocardiogram signal which is the detection signal of each channel is held by the sampling and holding circuit 11 at a sampling frequency of 1 kHz, and the multiplexer 1
2 is selected, digitized by the A / D converter 13, and taken into the microcomputer 14.

In the measurement, when a learning command signal is input from the control panel as a preliminary measurement, the R wave of each channel is detected, the time axis and the base line at 7 points are aligned, and the sampling signal is detected. α _ij is calculated and stored (see FIG. 7A).

When the measurement start signal is input, the sampling signal of each channel is smoothed by sequentially averaging, for example, the front and rear 16 signals, and the basic component signal V of each channel is detected. Next, for each channel, a residual signal V _r is generated by subtraction, and a comparison signal U for each channel is created according to the α _ij unique to the subject learned in advance. C is calculated at each sampling time of each channel, and the noise index C is sequentially calculated. The electrocardiogram signal component is detected by multiplying the residual signal V _r , which is each sampling signal of each channel, by the noise index C to reduce the noise therein.
This electrocardiogram signal component is added to the basic component signal at the time of sampling belonging to the belonging channel and reproduced. Further, a standard 12-lead ECG signal is created by a well-known method based on the reproduced ECG signal of 8ch. Display device 15
Then, the electrocardiogram signal is D / A converted, displayed, and recorded in the recorder 16 (see FIG. 7B).

8 to 13 show the test results for confirming the noise elimination effect of the standard 12-lead ECG signal waveform output by such an electrocardiograph. First, FIGS. 8 and 9 show a case where the muscle is tense and an EMG signal is generated during the measurement.
Shows the case where the process of the present invention was performed, and FIG. 9 shows the case where the process of the present invention was not performed. FIGS. 10 and 11 show the case where the above-mentioned electromyogram is mixed so much that the original signal is buried in noise, and FIG. 10 shows the case where the processing of the present invention is performed.
FIG. 11 shows the case where the process of the present invention is not performed. Figure 1
2 and FIG. 13 are cases where ham occurred due to poor contact of unspecified electrodes, FIG. 12 is the case where the treatment of the present invention was performed, and FIG. 13 is the case where the treatment of the present invention was not performed. .

FIG. 14 shows the reproduced original signal when the electrocardiogram signal of the test according to FIG. 12 is processed by using the above-mentioned f (c) instead of C, and the noise is higher than that of FIG. It is confirmed that the original signal is accurately removed and the original signal is reproduced more faithfully.

[0049] In such an embodiment, alpha _ij has been the calculation directly from the detection signal V _IN itself as being derived from the detection signal V _IN signal, as shown in Figure 15, by modifying the arrangement of Figure 1 remaining It can also be calculated by the difference signal V _r . Further, it is also possible to use the basic component signal V as shown by the dotted line in the figure.

The present invention is not limited to the above-mentioned biological signals, but has a unique amplitude relationship due to the signals propagating from the signal source at least at approximate points to each other by a linear or non-linear functional relationship. It can be applied to various systems that detect signals at a plurality of locations.

[0051]

According to the present invention, in a system in which a common signal source is measured at a plurality of points, noise in the detection signal band can be removed without distorting the original signal. For example, when applied to an electrocardiograph, it is possible to remove random noise in the electrocardiogram signal region, myocardial signal, hum, etc. without distorting the electrocardiogram signal. Further, even an electrocardiogram mixed with arrhythmia or the like caused by PVC or cardiac fibrillation can be detected without distortion by removing noise.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing a basic configuration of an electrocardiograph to which the present invention is applied.

FIG. 2 is a signal waveform diagram explaining the operation of the electrocardiograph.

FIG. 3 is a signal waveform diagram for explaining the operation of the concentricity gauge.

FIG. 4 is a signal waveform diagram for explaining the operation of the electrocardiograph.

FIG. 5 is a diagram illustrating another embodiment of the present invention.

FIG. 6 is a diagram showing a configuration of an electrocardiograph in which the present invention is implemented by a microcomputer.

FIG. 7 is a flowchart illustrating an operation of the microcomputer.

FIG. 8 is a signal waveform diagram of a test example for confirming the effect of the electrocardiograph.

9 is a signal waveform diagram of the comparison test of FIG.

FIG. 10 is a signal waveform diagram of a test example for confirming the effect of the electrocardiograph.

11 is a signal waveform diagram of the comparative test of FIG.

FIG. 12 is a signal waveform diagram of a test example for confirming the effect of the electrocardiograph.

13 is a signal waveform diagram of the comparison test of FIG.

FIG. 14 is a signal waveform diagram for confirming the effect of the embodiment of FIG.

FIG. 15 is a functional block diagram showing a modified example of the basic configuration of an electrocardiograph to which the present invention is applied.

[Explanation of symbols]

 1 signal source

Claims (9)

[Claims] 1. A noise reduction method for reducing noise mixed in a detection signal when detecting an original signal generated from a common signal source at a plurality of measurement points, the signal being derived from the detection signal. By sampling at M sampling points of, a parameter for weighting the detection signal at an arbitrary measurement point as a linear combination of the detection signals at the remaining measurement points is calculated (where M + 1 is the measurement value). The number of points or more), detecting the basic component signal in the detection signal, to generate a residual signal by subtracting the basic component signal from the detection signal, of the remaining measurement points weighted by the parameter The residual signal is added to calculate a comparison signal corresponding to the residual signal at any measurement point, and the detection signal is obtained at each sampling time point at any measurement point. The noise index that reflects the Murrell noisy degree,
The comparison signal and the residual signal at each sampling time are compared and calculated, and the residual signal is reduced by decreasing the signal amplitude of the residual signal at any measurement point according to the noise index. A noise removal method characterized by detecting an original signal component in a signal and adding the original signal component and the basic component signal to reproduce an original signal with reduced mixed noise at an arbitrary measurement point. . 2. The noise elimination method according to claim 1, wherein the signal derived from the detection signal is the detection signal itself. 3. The noise removing method according to claim 1, wherein the signal derived from the detection signal is a fundamental component signal detected from the detection signal. 4. The noise removing method according to claim 1, wherein the signal derived from the detection signal is a residual signal generated by subtracting the basic component signal from the detection signal. 5. The noise elimination method according to claim 1, wherein the noise index is calculated based on the coherence calculation of the comparison signal and the residual signal at the sampling time point of any of the measurement points. 6. A noise index is a function of coherence of a comparison signal and a residual signal, and this function is 1 when the noise index is 1 and is 1 when the noise index is relatively high.
6. The denoising method according to claim 5, wherein the value becomes relatively close to 0 and rapidly becomes close to 0 when relatively low. 7. The noise index at each sampling time point at an arbitrary measurement point is sequentially calculated as an average value of coherence of the comparison signal and the residual signal before and after each sampling time point. Noise removal method. 8. The noise index is a common noise index for the plurality of measurement points at each sampling, by calculating an average value of coherences of the comparison signal and the residual signal of the plurality of measurement points at each sampling. The noise removal method according to claim 5, wherein the noise removal method is performed as follows. 9. A noise reduction device for reducing noise mixed in a detection signal when detecting an original signal generated from a common signal source at a plurality of measurement points, and a signal derived from the detection signal. By sampling at M sampling points of, a parameter for weighting the detection signal of any of the measurement points as a linear combination of the detection signals of the remaining measurement points (where M + 1 is the measurement point Parameter calculation means for calculating, a basic component signal detection means for detecting a basic component signal in a detection signal, a subtraction means for subtracting the basic component signal from the detection signal to generate a residual signal, A comparison signal for adding the residual signals at the remaining measurement points weighted by parameters and calculating a comparison signal corresponding to the residual signal at any measurement point. No. calculation means, and compares the comparison signal and the residual signal at each sampling time of any of the measurement points, and calculates a noise index that reflects the degree of noise contamination included in the detection signal at each sampling time. Noise index calculation means, and noise reduction means for detecting the original signal component in the residual signal by reducing the signal amplitude of the residual signal at any of the measurement points according to the noise index, And a means for adding the basic component signal and the original signal component in the residual signal to reproduce an original signal with reduced mixed noise at any of the measurement points. Removal device.