WO2016203862A1

WO2016203862A1 - Signal processing method, biosignal processing method, signal processing device, and biosignal processing device

Info

Publication number: WO2016203862A1
Application number: PCT/JP2016/063672
Authority: WO
Inventors: 洋平冨田
Original assignee: フォスター電機株式会社
Priority date: 2015-06-15
Filing date: 2016-05-06
Publication date: 2016-12-22
Also published as: JP2017000625A; JP6501643B2

Abstract

[Problem] To find an estimated signal from which noise has been removed using a filter process on an information signal. [Solution] Provided is a signal processing method that extracts an information signal by removing noise from a measured signal that is measured in circumstances in which the information signal and noise are mixed, wherein the method obtains a plurality of measured signals from different measurement locations, detects correlations among the plurality of measured signals, detects phase synchronicity among the plurality of measured signals, and outputs a plurality of estimated signals that are estimated to be the information signal by removing uncorrelated components with low correlation and non-phase synchronous components with low phase synchronicity from each of the plurality of measured signals.

Description

Signal processing method, biological signal processing method, signal processing apparatus, and biological signal processing apparatus

The present invention relates to a signal processing method, a biological signal processing method, a signal processing device, and a biological signal processing device that extract information signals by removing noise from a measurement signal measured in a state where information signals and noise are mixed.

In the health care field, it is possible to configure a pulse wave acquisition earphone by incorporating a sensor in the earphone.
That is, a sensor for measuring a pulse wave, which is a signal linked to the heartbeat, is attached to the earphone, so that a biological signal can be acquired and analyzed on a daily basis.

The results obtained by measuring pulse waves in this way can be applied to daily health management, lifestyle-related diseases, chronic stress diagnosis, etc., and are useful for healthcare problems that are an issue in developed countries. It is expected.
For example, during exercise for maintaining health, it is becoming common to exercise while measuring the heart rate in order to make the effect of exercise effective. For such heart rate management during exercise, the above-described pulse wave acquisition earphone can be used in a portable biometric device or the like.

In this type of pulse wave acquisition earphone, there is a problem that noise is mixed due to body movement or disturbance light. For this reason, it is important to remove the noise from the measurement signal acquired by the pulse wave acquisition earphone and extract the component of the pulse wave.
For example, various proposals have been made in the following patent documents related to a technique for accurately measuring this type of pulse wave.

Patent No. 5060186 Japanese Patent No. 3735774 Japanese Patent No. 3705814

In general, body motion noise has a frequency close to that of a pulse wave and has a large amplitude, so it cannot be easily removed. For this reason, there are two main approaches to countermeasures.
The first is hardware design for stabilizing the contact state of the sensor, and the second is noise removal by signal processing.

The hardware design includes reduction of the earphone cable swing due to wireless communication and a special shape for fixing the sensor. However, since the optimal hardware form largely depends on the usage scene, it cannot cope with all situations.
On the other hand, noise removal by signal processing has high utility value because it is not so limited by the usage scene.

In the method described in Patent Document 1 described above, a body motion sensor that detects the body motion of the subject is provided to separate the pulse wave signal included in the measurement signal from the body motion noise. As a body motion sensor in this case, a triaxial acceleration sensor or the like is necessary, and it is difficult to incorporate it in a small earphone. In addition, there is a problem that noise other than body movement (noise due to disturbance light) cannot be removed.

In the method described in Patent Document 2 described above, an estimated waveform is generated based on the heartbeat interval and the cardiac contraction time, and the biological signal processing is executed using the estimated waveform. However, this method has a problem that accurate signal processing cannot be performed because the estimated waveform does not correspond correctly when the heart rate changes suddenly or when an arrhythmia appears.

In the method described in Patent Document 3 described above, it is proposed to perform noise processing using two signals measured by light having different wavelengths. However, in order to perform this, there is a problem that it is necessary to select an appropriate frequency and to adjust in advance.
Further, as various methods other than these, even when a plurality of measurement signals are used, there is a problem that noise cannot be accurately separated when pulse waves and noise are correlated.

Further, in order to remove noise from the measurement signal, the following methods and the like have been proposed as existing methods for signal processing. Below, various existing methods and their problems are listed.
a. Frequency selection:
The frequency of the measurement signal is selected using a band pass filter or the like. In this case, the structure is simple and easy to design, but there is a problem that it is difficult to remove noise close to the frequency of the pulse wave.

b. Sudden noise removal:
Remove sudden noise by using a median filter. In this case, even if the amplitude is large, it can be removed, but there is a problem that it is difficult to remove if the sudden noise is continuous.
c. Remove trend:
Noise is removed using polynomial fitting, Savitzky-Golay smoothing filter, etc. In this case, improvement in pulse rate estimation accuracy can be expected, but there is a problem that high-frequency noise is difficult to remove.

The existing methods a to c have a common problem that the effect is small for noise having a frequency close to that of the pulse wave.
The present invention has been made to solve the above-described problems, and a signal processing method, a biological signal processing method, and a signal that can obtain an estimated signal from which noise has been removed by filtering the information signal. An object is to provide a processing device and a biological signal processing device.

(1) The present invention is a signal processing method for extracting an information signal by removing noise from a measurement signal measured in a state where information signals and noise are mixed, and acquiring a plurality of measurement signals at different measurement locations. Detecting a correlation between the plurality of measurement signals, detecting phase synchronization between the plurality of measurement signals, and detecting a non-correlated component having a low correlation from each of the plurality of measurement signals and the phase synchronization. And a plurality of estimated signals estimated as the information signal are output.

(2) In the above (1), a correlation is detected by calculating a mutual spectrum of a plurality of measurement signals.
(3) In the above (1), the cross spectrum of the plurality of measurement signals is calculated to detect the correlation, and the phase synchronization is detected by a non-phase synchronization cancel function of the plurality of measurement signals. It is characterized by that.

(4) The plurality of measurement signals are two pulse wave signals obtained by measuring pulse waves at left and right target positions of the subject, and the measurement signals according to any one of (1) to (3) above This is biological signal processing characterized by using signal processing.

(1) In the present invention, a plurality of measurement signals at different measurement locations are acquired, the correlation between the plurality of measurement signals is detected, the phase synchronism between the plurality of measurement signals is detected, and each of the plurality of measurement signals is detected. Since the uncorrelated component and the non-phase-synchronous component are removed from the output signal and the multiple estimated signals that are estimated as the information signal are output, the correlation and synchronism between the signals are evaluated and it does not depend on the frequency information Even when the frequency of the information signal and the noise is close, it is possible to remove the noise from the measurement signal in which the information signal and the noise are mixed and extract a desired information signal.

(2) In the above (1), since the mutual spectrum of a plurality of measurement signals is calculated and the correlation is detected and the filter process is performed, the noise is removed from the measurement signal in which the information signal and the noise are mixed and the desired signal is obtained. An information signal can be extracted.
(3) In the above (1), in order to detect the correlation by calculating the mutual spectrum of the plurality of measurement signals and to detect the phase synchronization by the non-phase synchronization cancellation function of the mutual spectrum of the plurality of measurement signals, It is possible to extract a desired information signal by removing the noise from the measurement signal in which the information signal and the noise are mixed.

(4) In biological signal processing using two pulse wave signals obtained by measuring a pulse wave at left and right target positions of the subject as a plurality of measurement signals, the pulse wave signal has phase synchronization and correlation. To detect the correlation by calculating the mutual spectrum of multiple measurement signals and detecting the phase synchronization by the non-phase synchronization cancellation function of the mutual spectrum of multiple measurement signals. Even if the frequency of noise is close to that of the noise, the noise and the information signal can be separated by filtering focusing on the correlation and phase synchronization, and the noise from the measurement signal in which the information signal and noise are mixed Thus, it is possible to extract a desired information signal.

It is a block diagram which shows the structure of the signal processing apparatus of embodiment of this invention. It is explanatory drawing which shows the characteristic in the signal processing of embodiment of this invention. It is explanatory drawing which shows the experimental condition in the signal processing of embodiment of this invention. It is explanatory drawing which shows the characteristic in the signal processing of embodiment of this invention. It is explanatory drawing which shows the characteristic in the signal processing of embodiment of this invention.

The best mode for carrying out the present invention (hereinafter referred to as an embodiment) will be described below in detail with reference to the drawings.
[Configuration of Embodiment]
The configuration of the signal processing apparatus 100 that executes the signal processing method according to the embodiment of the present invention will be described with reference to FIG.

Sensor 11 and sensor 12 are a plurality of sensors arranged at different measurement locations that measure pulse waves, which are signals linked to the heartbeat of the subject. Note that the sensor 11 and the sensor 12 can be configured by known optical sensors or the like. Moreover, in this embodiment, it is preferable to mount a plurality of

sensors

11 and 12 at left and right positions of the subject and measure pulse waves.

For example, not only those that are built into the stereo earphones or headphones that the subject wears, but also those that are attached in a clip format so that both ear lobes are sandwiched, those that are installed near the left and right temples of the headband, It may be the one attached to both wrists, the one attached to both ankles, or the like. The plurality of sensors are at least two, but may be more than that.

In the present embodiment, a description will be given using a specific example in which two

sensors

11 and 12 are provided and signal processing is executed based on two measurement results.
These

sensors

11 and 12 measure pulse waves at different measurement locations, and generate information signals from the measurement results of pulse waves that are desired measurement targets. In addition, these

sensors

11 and 12 output noise as measurement signals in a state where measurement signals other than pulse waves due to body movements, disturbance light, and the like of the subject are mixed in the information signal.

The signal processing apparatus 100 removes the noises n1 and n2 from the plurality of measurement signals x1 and x2 measured in a state where the information signals (pulse waves) s1 and s2 and the noises n1 and n2 coexist, thereby removing the information signals (pulses). Wave) components are extracted and a plurality of estimated signals s′1 and s′2 are output.
Each of a plurality (two types) of measurement signals x1 and x2 includes an information signal and a noise signal (x1 = s1 + n1, x2 = s2 + n2), but the information signal and the noise are uncorrelated (s1 and n1). The inner product is 0, the inner product of s1 and n2 is 0, the inner product of s2 and n1 is 0, and the inner product of s2 and n2 is 0).

Therefore, the signal processing apparatus 100 includes a signal acquisition unit 101 and a signal acquisition unit 102 that acquire a plurality of measurement signals x1 and x2 from the sensor 11 and the sensor 12. Further, the signal processing apparatus 100 includes a signal output unit 131 and a signal output unit 132 in order to output a plurality of estimated signals s′1 and s′2 subjected to signal processing.

The signal processing apparatus 100 includes a filter coefficient generation unit 110 and an information signal extraction unit 120 in order to execute signal processing.
The filter coefficient generator 110 includes a cross spectrum calculator 111, a phase synchronization detector 112, a power spectrum calculator 113, a power spectrum calculator 114, an inverse processor 115, an inverse processor 116, and an operator 117. And a calculation unit 118.

Here, the cross spectrum calculation unit 111 calculates a cross spectrum (mutual spectrum) C _x1x2 meaning cross-correlation by multiplying the frequency components of the plurality of input measurement signals x1 and x2. When n1 and n2 are uncorrelated, the cross spectrum of x1 and x2 corresponds to the power spectrum of s1 (= s2).

The phase synchronization detection unit 112 calculates phase synchronization related to the phase difference between the plurality of input measurement signals x1 and x2. A function (non-phase synchronization cancellation function) α meaning phase synchronization is introduced. Here, when 0 ≦ α ≦ 1 and n1 and n2 are non-phase-synchronized, α using the phase of the cross spectrum as an input variable approaches 0.

The power spectrum calculation unit 113 squares the Fourier spectrum of one measurement signal x1 to calculate a power spectrum (self spectrum) that means autocorrelation. Similarly, the power spectrum calculation unit 114 squares the Fourier spectrum of the other measurement signal x2 to calculate a power spectrum that means autocorrelation. The reciprocal processing unit 115 generates a reciprocal C ⁻¹ _X1X1 of the power spectrum of one measurement signal x1. The reciprocal processing unit 116 generates the reciprocal C ⁻¹ _X2X2 of the power spectrum of the other measurement signal x2.

The information signal extraction unit 120 includes a filter processing unit 121 and a filter processing unit 122. Here, the filter processing unit 121 and the filter processing unit 122 perform the filter processing generated by the filter coefficient generation unit 110 on the plurality of measurement signals x1 and x2 from the

sensors

11 and 12, thereby causing noise n1. , N2 and a plurality of estimated signals s′1, s′2 are extracted.

[Operation of Embodiment]
Hereinafter, as an operation of the embodiment of the present invention, a signal processing method which is an operation of the signal processing apparatus 100 will be described with reference to FIG. Here, it is assumed that the sensor 11 and the sensor 12 are installed in each stereo earphone.

[Prerequisites]
The sensor 11 is installed on one of the earphones, and generates a measurement signal x1 measured in a state where the information signal (pulse wave) s1 and the noise n1 are mixed. Similarly, the sensor 12 is installed on the other side of the earphone, and generates a measurement signal x2 measured in a state where an information signal (pulse wave) s2 and noise n2 are mixed.

Here, the information signal s1 and the information signal s2 are caused by a pulse wave that is a signal linked to the heartbeat of the subject, and from both ears that are the left and right target positions of the subject. Since it was measured, it has high correlation and high phase synchronization.
On the other hand, the information signal s1 and the noise n1 are caused by completely different vibration sources and often do not have correlation and phase synchronization (non-correlation and non-phase synchronization). Further, the information signal s2 and the noise n2 are caused by completely different vibration sources, and often have neither correlation nor phase synchronization (non-correlation and non-phase synchronization).

On the other hand, it is unlikely that the timing of the noise and the magnitude of the noise will be the same for a plurality of sensors, such as when the subject shakes his head or touches the sensor. In addition, when there is disturbance light on the sensor, the incident optical path is different, and therefore there is little possibility that the noise timing and the noise magnitude match among the plurality of sensors. Therefore, the noise n1 and the noise n2 often have no correlation and phase synchronism, or have low correlation and phase synchronism (non-correlation and non-phase synchronism).

[Problem setting]
・ Measured signal model:
There are two types of measurement signals x1 and x2, each of which comprises information signals s1 and s2 and noises n1 and n2.
x1 = s1 + n1,
x2 = s2 + n2,
Here, the information signal and noise are uncorrelated (the inner product of s1 and n1 is 0, the inner product of s1 and n2 is 0, the inner product of s2 and n1 is 0, and the inner product of s2 and n2 is 0).
・ Purpose of processing:
An estimated signal from which the noise signal is removed is obtained by filtering the measurement signal including the information signal and noise.
・ Solution policy:
A Wiener filter that minimizes the mean square error between the information signal and the estimated signal can be used.

Here, in order to derive the Wiener filter, the following two methods (conventional methods) can be considered.
Conventional method Estimate the spectrum of the information signal. However, the information signal may not be directly measured.

Conventional method 2.
The average value of the noise power spectrum is estimated by a spectral subtraction method, and the noise is reduced by subtracting it from the power spectrum of the measurement signal including noise. However, there are cases where only noise cannot be obtained.
·solution:
The above conventional methods cannot be solved. Therefore, the following solution is proposed.

Solution 1 A cross spectrum is introduced in the derivation of the Wiener filter. In this case, if the noise is uncorrelated (the inner product of n1 and n2 is 0), the effect can be expected.
Solution 2 In addition to the above solution technique 1, a non-phase synchronization cancellation function (α) is further introduced. In this case, even if the noise is correlated, if it is non-phase-synchronized (the inner product of n1 and n2 is not 0, but the phase difference is not 0), it can be expected to exert an effect.

[Outline of processing]
Here, the cross spectrum calculation unit 111, the power spectrum calculation unit 113, the power spectrum calculation unit 114, the reciprocal number processing unit 115, and the reciprocal number processing unit 116 detect the correlation between the plurality of measurement signals x1 and x2. A correlation detection unit is configured. Here, the Wiener filter can be obtained using only the measurement signal by assuming that there is no correlation between the noise and the pulse wave, two correlations between the noise and the like.

Also, the phase synchronization detection unit 112 constitutes a phase synchronization detection unit that detects the phase synchronization of the plurality of measurement signals x1 and x2. Here, even if the two noises have a correlation, a filter coefficient for removing the noise can be obtained if the phases are out of phase and non-phase-synchronous.

The filter processing unit 121 and the filter processing unit 122 are each composed of a time series filter, and the correlation is obtained from each of the plurality of measurement signals based on the filter coefficient generated by the correlation and the phase synchronization. The non-correlated component having a low phase and the non-phase synchronous component having a low phase synchronism are removed to constitute an information signal extraction unit 120 that extracts an information signal. Here, by adding the phase synchronism to the correlation, it is possible to expect a noise removal effect that cannot be realized by the conventional basic Wiener filter configuration.

That is, filter coefficients are generated so as to extract high components from two measurement signals obtained from the two

sensors

11 and 12 while suppressing low correlation and phase synchronization components from the above conditions. Generated by the unit 110. Then, the information signal extraction unit 120 using the filter coefficient extracts high components from the two measurement signals obtained from the two

sensors

11 and 12 while suppressing components having low correlation and phase synchronization. Noise is removed by filtering.

[Details of processing]
The measurement signal when noise is mixed in the pulse wave signal is expressed as follows.
x1 = s1 + n1,
x2 = s2 + n2, (1)
x1 is a column vector of finite length T, and it is considered that noise n1 is added to the pulse wave signal s1. Similarly, x2 is a column vector of finite length T, and it is considered that noise n2 is added to the pulse wave signal s2. In the present embodiment, noise removal is signal processing for obtaining estimated signals s′1 and s′2 close to s using only the measurement signals x1 and x2.

In the present embodiment, the Wiener filter is a filter that regards signals and noise as stochastic processes and minimizes the mean square error that is an evaluation function.
Taking one measurement signal x1 of the above equation (1) as an example, the estimation signal s'1 is a convolution integral of the Wiener filter h1 and the measurement signal x1,
s'1 = h1 * x1 (2)
It is expressed as

Furthermore, assuming the stationarity of a certain section, in the frequency domain,
S1 (ω) = H1 (ω) X1 (ω) (3)
It becomes.
The error of the estimated value S′1 (ω) with respect to the measurement signal is
E1 (ω) = S (ω) −S1 (ω)
= S (ω) −H1 (ω) X1 (ω) (4)
It becomes.

And the mean square value of the error, i.e.
ε [| E1 (ω) | ² ] = ε [| S (ω) −H1 (ω) X1 (ω) | ² ] (5)
Minimizing the noise suppresses the noise to the maximum.
Here, ε [] means an expected value of the expression in [].

Therefore, if the result of partial differentiation of equation (5) with H1 (ω) is 0,
H1 (ω) = ε [X ^* 1 (ω) X1 (ω)] ⁻¹ ε [S ^* (ω) X1 (ω)] (6)
It becomes.
Thereby, the winner filter H1 (ω) is obtained.
Here, the one with * on the right shoulder of a character means a complex conjugate.

The same applies to the other measurement signal x2.
H2 (ω) = ε [X ^* 2 (ω) X2 (ω)] ⁻¹ ε [S ^* (ω) X2 (ω)] (7)
It becomes.
By the way, in the above formulas (6) and (7), S (ω) is unknown and cannot be used directly. Therefore, assuming that the pulse wave S (ω) and the noise N (ω) are uncorrelated,
S ^* (ω) X1 (ω)
= S ^* (ω) (S (ω) + N1 (ω))
= S ^* (ω) S (ω) (8)
It becomes.

Further, assuming that the noise N1 (ω) and the noise N2 (ω) are uncorrelated and the cross spectrum (cross spectrum) of the measurement signals X1 and X2 is calculated by the cross spectrum calculation unit 111, X ^* 1 (ω) X2 (ω)
= (S (ω) + N1 (ω)) ^* (S (ω) + N2 (ω))
= S ^* (ω) S (ω) (9)
It becomes.

Since the same applies to the other sensor, the equations (6) and (7)
H1 (ω)
= Ε [X ^* 1 (ω) X1 (ω)] ⁻¹ ε [X ^* 1 (ω) X2 (ω)]
= C ^-1 _X1X1 (ω) C _X1X2 (ω) (10)
H2 (ω)
= Ε [X ^* 2 (ω) X2 (ω)] ⁻¹ ε [X ^* 1 (ω) X2 (ω)]
= C ^-1 _X2X2 (ω) C _X1X2 (ω) (11)
It becomes. That is, the filter coefficient can be obtained using only the measurement signal.

If N1 (ω) and N2 (ω) are not uncorrelated, Equations (8) and (9) will not be equal, but noise non-phase synchrony is assumed and the noise phase difference is considered. To further increase the noise effect.
For this reason, in the phase synchronization detection unit 112 using a function (non-phase synchronization cancellation function) α meaning phase synchronization,
The above C _X1X2 (ω) is changed to α (ω) C _X1X2 (ω) (12)
And

here,
α (ω) = {(cos (∠C _X1X2 (ω)) + 1) / 2} ^p (13)
Then, the filter coefficient is calculated.
Here, p is a parameter for adjusting how far non-phase synchronism is allowed.

FIG. 2 shows α (ω) when p is changed. When the angle is 0 [rad], that is, the signal whose phase is synchronized, α (ω) = 1, and the above C _X1X2 (ω) is stored as it is.
On the other hand, when the phase shift increases, α (ω) decreases, the value of the above equation (12) decreases, and its component is suppressed by the Wiener filter. Here, as p is increased, the effect of removing the non-phase-synchronous component is increased. Therefore, it is desirable to increase p.

Finally, it can be considered that the filter characteristics do not satisfy 0 ≦ | H | (ω) ≦ 1 when the noise is extremely large. Therefore, if | H (ω) | ≧ 1,
H (ω) = H (ω) / (| H (ω) |) (14)
To adjust the filter coefficient.

The filter is generated by αC _X1X2 calculated by the cross spectrum calculation unit 111 and the phase synchronization detection unit 112 as described above, and C ⁻¹ _X1X1 calculated by the power spectrum calculation unit 113 and the reciprocal processing unit 115, and the filter The filter coefficient H1 (ω) supplied to the processing unit 121 is
H1 (ω) = C ⁻¹ _X1X1 αC _X1X2 (10 ′)
It becomes.

Similarly, it is _generated by αC _X1X2 calculated by the cross spectrum calculation unit 111 and the phase synchronization detection unit 112 as described above, and C ⁻¹ _X2X2 calculated by the power spectrum calculation unit 114 and the reciprocal processing unit 116. The filter coefficient H2 (ω) supplied to the filter processing unit 122 is
H2 (ω) = C ⁻¹ _X2X2 αC _X1X2 (11 ′)
It becomes.

That is, it was possible to obtain filter coefficients for removing uncorrelated components and non-phase-synchronized components using only the measurement signal.
The filter processing unit 121 configured with a time-series filter removes noise from the measurement signal using uncorrelated components and non-phase-synchronized components by executing processing using the filter coefficient of the above equation (10 ′). Thus, an estimated signal estimated as a pulse wave can be output.

Similarly, the filter processing unit 122 configured with a time-series filter performs processing using the filter coefficient of the above equation (11 ′), so that noise is generated from the measurement signal by uncorrelated components and non-phase-synchronized components. And an estimated signal estimated as a pulse wave can be output.
[Method of implementation]
The noise removal method proposed in this embodiment is expressed by the block diagram shown in FIG.

Here, the estimated signals s′1 and s′2 are obtained from the measurement signals x1 and x2 by the filter coefficients h1 and h2.
In this method, since the self spectrum and the cross spectrum are calculated, the filter coefficient is sequentially updated by the frame processing.

If the filter feature at frame k−1 is H (k−1, ω), the filter feature H (k, ω) at frame k is
H (k, ω) = μH (k, ω) + (1−μ) H (k−1, ω) (15)
As shown in FIG. Here, 0 <μ ≦ 1.

When μ is increased, the convergence is quicker, but it is necessary to consider how often frame processing is performed.
[Results of treatment experiment]
Hereinafter, the operation and effect of this embodiment will be verified.

Here, it is assumed that the pulse wave component is replaced with a sine wave and non-phase-synchronous noise is mixed in the two measurement signals. The experimental conditions in this case are shown in FIG.
A measurement signal in a state where a sine wave noise of 1.5 [Hz] and white noise are superimposed on a sine wave pulse signal of 1.0 [Hz] is used. Here, it is assumed that the sine wave noise of 1.5 [Hz] is shifted in phase by 0.1π on the left and right. Further, the allowable parameter p of non-phase synchronicity is 100, the update coefficient μ is 0.5, the number of FIR taps is 100, the frequency conversion window length is 15 [s], and the frame shift time is 0.5 [s]. As an experiment.

FIG. 4 (1) and (2) show that the measurement signals x1 and x2 are mixed with sinusoidal pulse wave signals and sinusoidal noise. In this state, the pulse wave component cannot be read from the waveforms of x1 and x2. When this embodiment is applied, as shown in FIGS. 4 (3) and 4 (4), the estimated signals s ′ 1 and s ′ 2 with the noise removed and the pulse wave s component appear are extracted. can do.

On the other hand, the experiment was also performed with the pulse wave based on the heartbeat of the subject and the jaw moved as the body motion of the subject. The measurement signal x1 shown in FIG. 5 (1) shows a state in which the pulse wave signal of the subject and the body movement noise of the subject are mixed. In this state, the pulse wave component cannot be read from the waveform of x1.

And when this embodiment is applied, as shown in FIG. 5 (2), it is possible to extract the estimated signal s′1 in a state where the noise is removed and the component of the pulse wave s appears. In FIG. 5, the measurement signal x2 and the estimation signal s′2 are not shown. Similarly, although the pulse wave component cannot be read from the waveform of x2, by applying this embodiment, The estimated signal s'2 in the state where the noise was removed and the component of the pulse wave s appeared could be extracted.

Note that the experimental results of FIGS. 4 and 5 described above show that the noise processing is started from 15 [s] because the Fourier transform window is set to 15 [s].
[Other Embodiments (1)]
In the above description, the

sensors

11 and 12 are supposed to be attached to the left and right positions of the subject, for example, both ears. However, the present invention can be applied even when the sensors are arranged at different positions other than both ears.

For example, for example, noise non-correlation and non-phase synchronization can be assumed for fingers and ears. However, since the phase shift of the pulse wave is expected, it is necessary to calculate and give the phase shift φ in advance when determining the parameter α described above.
α (ω) = {(cos (∠C _X1X2 (ω) ± Φ) +1) / 2} ^p (13 ′)
It is possible to respond by changing.

[Other embodiment (2)]
In the above embodiment, a plurality of measurement signals are required. Specifically, it is necessary to obtain measurement signals from both ears of the subject. For this reason, when the contact state of one ear in an earphone or the like is extremely bad, only one measurement signal can be obtained. It is desirable to add a function for determining such an abnormal state.

In such an abnormal state in which only one of the measurement signals can be obtained, it is also preferable to provide a notification unit that notifies the abnormal state. Alternatively, it is possible to switch to a conventional method in which signal processing is performed using only one measurement signal in an abnormal state where only one measurement signal is obtained.

[Other embodiment (3)]
In this embodiment, a part of the pulse rate calculation function can be used. When obtaining the pulse rate, a method of detecting the peak value in the frequency domain after Fourier transforming the pulse wave is usually used. In the present embodiment, since the value obtained by equation (12) corresponds to the Fourier transform of the pulse wave, the pulse rate can be calculated with reference to this value.

In this case, the noise changes from moment to moment, but the peak frequency and intensity of the pulse wave change only within a certain range. Therefore, by specifying the pulse wave change range in advance, the pulse rate detection accuracy can be increased.
[Other embodiment (4)]
In the above description of the embodiment, the pulse wave has been described using a specific example of removing noise included in the measurement signal. However, the present invention can be applied to various biological signals other than the pulse wave.

Further, when non-correlated and non-phase-synchronized noise is mixed with various signals other than biological signals, it is possible to perform signal processing for noise removal.
[Effect obtained by the embodiment]
In this embodiment, a plurality of measurement signals at different measurement locations are acquired, the correlation between the plurality of measurement signals is detected, the phase synchronism between the plurality of measurement signals is detected, and there is no difference from each of the plurality of measurement signals. When outputting a plurality of estimated signals estimated as information signals by removing the correlation component and non-phase synchronization component, the correlation and synchronization between the signals are evaluated and do not depend on the frequency information. Even if the frequency of the signal and the noise is close, it is possible to extract the desired information signal by removing the noise from the measurement signal in which the information signal and the noise are mixed.

In addition, since the mutual spectrum of a plurality of measurement signals is calculated to detect the correlation and filter processing is performed, it is possible to remove the noise from the measurement signal in which the information signal and the noise are mixed to extract a desired information signal. It becomes possible.
In addition, information signals and noise are mixed because the cross spectrum of multiple measurement signals is calculated to detect correlation and the phase synchronization is detected by the non-phase synchronization cancellation function of the mutual spectrum of multiple measurement signals. It is possible to extract a desired information signal by removing noise from the measured signal.

Further, in the biological signal processing using two pulse wave signals obtained by measuring the pulse wave at the left and right target positions of the subject as a plurality of measurement signals, the pulse wave signal includes phase synchronization and correlation. In order to detect the correlation by calculating the mutual spectrum of the plurality of measurement signals and to detect the phase synchronization by the non-phase synchronization cancellation function of the mutual spectrum of the plurality of measurement signals, Even when the frequency of the noise is close, the noise and the information signal can be separated by filtering focusing on the correlation and phase synchronization, and the noise can be removed from the measurement signal mixed with the information signal and noise. The desired information signal can be extracted by removing the signal.

11, 12 Sensor 100 Signal processor 110 Filter coefficient generator 120 Information signal extractor

Claims

A signal processing method for extracting information signals by removing noise from a measurement signal measured in a state where information signals and noise are mixed,
Acquire multiple measurement signals at different measurement locations,
Detecting a correlation between the plurality of measurement signals;
Detecting phase synchronism between the plurality of measurement signals;
Removing a non-correlated component with low correlation and a non-phase synchronous component with low phase synchronization from each of a plurality of measurement signals, and outputting a plurality of estimated signals estimated as the information signal,
And a signal processing method.
Calculating a mutual spectrum of a plurality of the measurement signals to detect the correlation;
The signal processing method according to claim 1.
Calculating the cross spectrum of the plurality of measurement signals to detect the correlation;
Detecting the phase synchronism by a non-phase synchronization cancellation function of a mutual spectrum of a plurality of the measurement signals;
The signal processing method according to claim 1.
The plurality of measurement signals are two pulse wave signals obtained by measuring pulse waves at left and right positions of the subject,
A biological signal processing method using the signal processing method according to claim 1.
A signal processing apparatus that extracts information signals by removing noise from a measurement signal measured in a state where information signals and noise are mixed,
A signal acquisition unit for acquiring a plurality of measurement signals at different measurement locations;
A correlation detection unit for detecting correlation between the plurality of measurement signals;
A phase synchronism detecting unit for detecting phase synchronism between the plurality of measurement signals;
An information signal extraction unit that generates a plurality of estimated signals estimated as the information signal by removing the uncorrelated component having low correlation and the non-phase synchronizing component having low phase synchronization from each of the plurality of measurement signals When,
A signal processing apparatus comprising:
A signal processing device according to claim 5,
Using two pulse wave signals obtained by measuring pulse waves at left and right positions of the subject as a plurality of measurement signals, and generating a plurality of estimation signals that are estimated as information signals,
A biological signal processing apparatus.