KR101641024B1 - Movement noise detection algorithm using wavelet transform - Google Patents

Abstract

The present invention relates to a motion noise determination method using wavelet transform that can determine whether motion noise is included in a PPG signal that is one of biological signals using wavelet transform, comprising: measuring a PPG signal; Generating a signal from which the high frequency noise is removed from the measured PPG signal; Setting a reference signal in the signal from which the high-frequency noise is removed; Selecting a base signal based on the set reference signal; Performing wavelet transform using the base signal; And determining a motion noise using the wavelet-transformed value.

Description

{MOVEMENT NOISE DETECTION ALGORITHM USING WAVELET TRANSFORM}

The present invention relates to a motion noise determination method using wavelet transform, and more particularly, to a motion noise determination method using a wavelet transform that can determine whether motion noise is included in a PPG signal, which is one of biological signals, .

As the incidence of various diseases has increased recently, there is an increasing interest in methods for preventing the health and various diseases of modern people. Many medical devices that determine the presence or absence of a disease in the body and treat the disease are also developed and can be identified by a simple method. However, these methods need to be directly inspected by the hospital, and society members seek a way to check their body condition in real time rather than go to the hospital, so that they can receive medical services without restriction of time and space. Research on U-healthcare technology, which is a ubiquitous and remote healthcare service utilizing technology, is actively being conducted.

Electroencephalography (EEG), electrocardiography (EMG), electromyography (PPG), photodynamic pulse wave, and electroogeography (EOG)

The signal that is easy to measure and monitor in various biological signal processing is PPG. In the case of EEG and ECG, the electrodes should be attached to the scalp and chest, and movement is very limited. However, since PPG uses one sensor attached to the body end region such as the finger, earlobe, and toe, It is advantageous in that the user's motion can be secured relative to the EEG. However, it is necessary to determine whether the noise is included in the signal processing due to the problem that the signal is distorted due to the noise of the measurement environment itself and the movement of the user.

Korean Patent Publication No. 10-2010-0008239 (Jan. 25, 2010)

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method for determining whether motion noise is included in a PPG signal, which is one of biological signals, using a period and wavelet transform.

According to an aspect of the present invention, there is provided a method of determining motion noise using wavelet transform, the method comprising: measuring a PPG signal using wavelet transform to determine whether motion noise is included in a PPG signal; Generating a signal from which the high frequency noise is removed from the measured PPG signal; Setting a reference signal in the signal from which the high-frequency noise is removed; Selecting a base signal based on the set reference signal; Performing wavelet transform using the base signal; And determining a motion noise using the wavelet-transformed value.

In the motion noise determination method using the wavelet transform, the step of generating the signal from which the high frequency noise is removed may include using an FIR filter, selecting the first period after the predetermined time elapses, The wavelet transform step repeats the wavelet transform so that the frequency of the low frequency band has a frequency band of 0 to 4 Hz.

The step of determining the dynamic noise includes the steps of: determining that the dynamic noise is included when the detected signal period is less than a predetermined value; determining whether the value obtained by adding the power per period of the predetermined frequency band is equal to or greater than a certain multiple , Wherein the predetermined multiple is 1.5 and the predetermined frequency range is 32 to 64 Hz.

According to the motion noise determination method using the wavelet transform according to the present invention configured as described above, it is possible to effectively determine whether motion noise is included in the PPG signal, which is one of biological signals.

1 is a signal graph showing a PPG signal detected through a PPG sensor.
2 is a signal graph showing a PPG signal including high frequency noise and dynamic noise.
3 is a diagram illustrating a process of wavelet-transforming and decomposing an original signal according to the present invention.
4 is a flowchart illustrating a method of determining whether a PPG signal includes motion artifacts according to the present invention.
5 is a flow chart showing the step of determining the number of wavelet transforms according to the present invention.
6 is a flow chart showing steps of a method for determining motion noise according to the present invention.
FIG. 7 is a graph showing the determination result of the motion noise according to the present invention. FIG.
8 is a graph showing a result of the determination of the dynamic noise according to whether or not the period is used according to the present invention.
9 is a graph showing a result of motion noise determination according to the wavelet transform according to the present invention.

The present invention may have various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the detailed description. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a signal graph showing a PPG signal detected through a PPG sensor. Generally, the PPG signal period is used to detect the heart rate and the PPI (Peak to Peak Interval) is used to detect the period, which means the time interval between the maximum value of one periodic signal and the maximum value of the next periodic signal . Thus, the PPI of the PPG signal can be used to determine the heart rate. Generally, a person has about 70 heartbeats per minute, but heart rate can be changed by physical activity or external stimuli. When the body is tensed due to external stimuli, the heart rate increases and the PPI becomes smaller. In the relaxed state, the heart rate decreases and the PPI increases. Such a change in the PPI can be used to determine the psychological state of a measurer or to determine the presence or absence of a disease.

2 is a signal graph showing a PPG signal including high frequency noise and dynamic noise. When high-frequency noise is included in the PPG signal, distortion of the signal occurs as shown in Fig. 2 (a). High-frequency noise does not make a big difference in the signal itself, but it should be removed as it is an obstacle to detection of PPI. A typical PPG signal is activated in the frequency range of 0 to 4 Hz, but high frequency noise is activated in a larger frequency band and can be removed using a low pass filter.

When the PPG signal includes motion noise, the distortion of the signal can not be detected as shown in Fig. 2 (b). Such motion artifacts occur in various frequency bands and can not be removed by simple filters because they affect the frequency range of 0-4 Hz where the PPG signal is activated. Since the motion noise is a serious obstacle when detecting a PPI, that is, a period of a signal, a signal processing technique for solving motion noise is required.

3 is a diagram illustrating a process of wavelet-transforming and decomposing an original signal according to the present invention. In the discrete wavelet transform, the original signal S can be represented by a sum of a signal that has passed through the low-pass filter and a signal that has passed through the high-pass filter. The original signal passes through the low-pass filter and the high-pass filter and then the approximation coefficients A [n] are output from the high-pass filter and the detail coefficients D [n] from the low-pass filter. Each output has a frequency band that is half the original signal. The original signal is the sum of the divided detailed coefficients and the approximate coefficients and can be expressed by the following equation.

S [n] = A [n] + D [n]

Each time the wavelet transform is performed, the frequency band is divided into two bands, the low frequency band component A [n] and the high frequency band component D [n].

When the signal A [n] that has passed through the low-pass filter is subjected to wavelet transformation in such a manner, signals are separated in the form as shown in FIG. 3, and the signal after the sixth wavelet transform can be expressed by the following equation.

&Quot; (2) " S = D1 + D2 + D3 + D4 + D5 +

4 is a flowchart illustrating a method of determining whether a PPG signal includes motion artifacts according to the present invention. As shown in FIG. 4, the method for determining whether or not a PPG signal according to the present invention includes motion noise includes the steps of measuring a PPG signal (S110), removing high frequency noise (S120), setting a reference signal A step S140 of selecting a similar base signal, a step S150 of performing wavelet transformation using a base signal, and a step S160 of determining a motion noise using the converted value.

When measuring a signal to apply an algorithm for the first time, the user measures the PPG signal without moving for a predetermined time in a stable state (S110). At this time, since the base signal to be used in the subsequent wavelet transform is determined by using the measured signal, it is measured in an environment where maximum noise is not included. During signal measurement, the user determines the end of the body to attach the PPG sensor. When determining the site of measurement, it is desirable to measure the signal by adopting the site with the smallest effect of the disease as the signal is distorted by the disease in the terminal part of the body where the blood flow due to disease frequently changes.

In the measurement of the signal on the body terminal, the measured pulse wave signal includes noise due to physical and mechanical characteristics as well as motion noise due to user's motion. The noise due to these characteristics is generally included in the high frequency range and should be removed because it is an obstacle in the detection of the PPI used for detecting the period of the PPG signal.

In step S120 of removing the high-frequency noise, a low-pass filter (LPF) can be used to remove the signal noise in the high-frequency band.

In the PPG signal measurement step (S110), the signal noise in the high frequency band due to the mechanical and physical characteristics exists in the PPG signal. The signal noise in the high frequency band is an obstacle to the detection of the period of the PPG signal, and therefore a high-frequency noise is eliminated by using a low-pass filter. Since wavelet transform is performed using FWT (Fast Wavelet Transform) based on a finite impulse response (FIR) filter, wavelet transform is used to remove high frequency noise using an appropriate FIR filter. Various filters such as a moving average filter, a Bartlett filter, a Hanning filter, and a Hamming filter can be used as the FIR filter. In the present invention, the Hamming filter having the largest minimum band-stop attenuation is used, but the present invention is not limited thereto. In the embodiment of the present invention, the cutoff frequency is set to 6 Hz, but the present invention is not limited thereto.

Next, a reference signal is set (S130) in order to set a base signal to be used in wavelet transformation in a signal in which high frequency noises are removed. The reference signal shall be measured by the user in a stable state and shall have no motion during the measurement. Even if a signal is measured under these conditions, a signal smaller than the size of a user's general pulse wave is initially detected at the characteristic of the sensor, and amplitude and size are not uniform. Therefore, since the signal at the initial stage of measurement can not be used as a reference signal, a reference signal is set using a normal signal after a predetermined time elapses. In this case, since the period of the general PPG signal has a period of 0.6 to 1.2 seconds, the first period after 3 seconds corresponding to about 3 periods can be set as the reference signal. PPG signals have different characteristics and shapes for different people and have different shapes depending on the region to be measured. It is easy to judge the motion noise by selecting the reference signal reflecting the pulse wave characteristics of each user and determining the wavelet transform base signal based on the reference signal.

Next, a base signal to be used in the wavelet transform is set using the reference signal of the user set in step S130 (S140). At this time, the base signal adopts the most similar function among the base functions by using the reference signal of the user, the basis function and the cross-correlation technique. There are several base functions used for wavelet transform. Of these, there are scaling functions and FIR filter can be applied to fit FWT. The basic functions 'Harr (harr)', 'Daubechies (dbN)', , 'Symbols (symN)', 'Coiflets (coifN)', 'Biothogonal wavelets (biorN)', 'Reverse biothogonal wavelets (rbioN)' and 'Discrete approximation of Meyer wavelet The reference signals are compared using a cross-correlation technique. The basis function closest to the reference signal among the seven basis functions compared is set as the number of wavelet transform sub-frames to be used later. In choosing the baseline function that is most similar to the reference signal, the maximum value of each cross correlation technique is compared by applying the cross correlation technique to the 7 basis functions and the reference signal, and the largest value is selected as the base signal. In order to quantify the similarity, the cross-correlation function of matlab is used to find the result of the cross-correlation method, and the maximum value is found by using the max function. Then, seven values corresponding to the maximum value of each of the seven base signals The base signal is determined in such a manner that the maximum value is again measured.

Next, the wavelet transform (S150) is performed using the basis function set in S140. A [n] is a low frequency band, and D [n] is a high frequency band. In this case, the signal is divided into an approximation (A [n]) part and detail (D [n]) part. At this time, the wavelet transform is repeated so that A [n] has a frequency band of 0 to 4 Hz. The reason for this repetition is that the normal heartbeat is less than 4 Hz, and the conversion step is set to include an interval of less than 4 Hz in order to easily detect one period. Generally, the sampling frequency is used in the conversion step determination, and the wavelet conversion frequency determination method will be described with reference to FIG.

5 is a flow chart showing the step of determining the number of wavelet transforms according to the present invention. The initial value of a is 1 (S151), and it is determined in step S152 whether f _s ? 2 ^a-1 is satisfied. If it is not satisfied, 1 is added to the value of a (S153), and the step of S152 is repeated.

fs is the number of times the smallest a-3 satisfying fs < 2 > ^{a-1 is} to be repeated. (n) is wavelet-transformed by the number of times of the smallest a-3 that satisfies fs? 2 ^a-1 (S154). Frequency domain and low-frequency domain by the wavelet transform, it is divided into A [n] and D [n] by a relative standard.

In the embodiment of the present invention, the wavelet transform is performed on A [n], but the present invention is not limited thereto. The values of A [n] and D [n] may vary depending on which part is transformed, It is also possible to divide A [n] by setting D [n] as a conversion target as well as continuously dividing it.

Specifically, in order to obtain a desired band, it is possible to determine which signal A [n] and D [n] is selected and divided at the time of one-time division. For example, if you want to obtain a signal in the (32-48) frequency band, you can use (0-128), (128-256), (0-64), (64-128) (128- 256) (A [n] conversion), three conversions (0-32), (32-64), (64-128) Since the frequency is divided into (0-32), (32-48), (48-64), (64-128), (128-256) . In this manner, it is possible to select which signal is selected and divided each time the wavelet transform is performed once, and the signal is divided into A [n] and D [n] each time the wavelet transform is performed.

Since the sampling frequency of the signal used in the embodiment of the present invention is 256 Hz, it includes the section where A [n] converted from the sixth order is 0 to 4 Hz. Table 1 shows the frequency distribution of a signal obtained by wavelet-transforming the original signal by the method shown in Fig.

n One 2 3 4 5 6 7 A [n] 0-128 0-64 0-32 0-16 0-8 0-4 0-2 D [n] 128-256 64-128 32-64 16-32 8-16 4-8 2-4

The sum of D [1] + D [2] + D [3] + D [4] + D [5] + D [6] + A [6] as in Equation Frequency signal.

Next, one period of the user's pulse wave is detected using a period in which the heartbeat is activated, and it is determined whether motion noise is included in the detected signal using the period (S160). A method for determining whether or not the motion noise is included will be described with reference to FIG.

6 is a flow chart showing steps of a method for determining motion noise according to the present invention. The signal period is detected using the PPI of the signal when detecting a period using the heartbeat activated interval (S161). If there is no dynamic noise, PPI appears at regular intervals, but if dynamic noise is included, the PPI changes. The normal heartbeat is about 0.8 seconds, and the period in which the motion noise occurs is rapidly changed from 0.3 to 1.6 seconds. Therefore, it is determined whether the period is less than 0.3 seconds (S162), and if it is smaller, it is determined that motion noise is included (S166). In step S162, the cycle of the determination reference is set to 0.3 seconds, but it may be set differently depending on the situation.

In determining the motion noise, a frequency band that is affected by heartbeat but whose degree is smaller than other frequency bands and whose motion noise is more affected than heartbeat is used. The portion corresponding to this frequency band corresponds to about 32 to 64 Hz. The heartbeat period is about 0.8 seconds on average, and if the component value of this frequency band is large, it is judged that the signal contains motion noise. In the determination of the dynamic noise, the power of the signal of each period and the signal of the frequency band of 32 to 64 Hz of the reference signal are summed (S163), and it is determined whether the signal of each period is a certain multiple of the reference signal or not If it is more than a predetermined multiple, it is determined that the dynamic noise is included in the period (S166). If the period is less than the certain period, it is determined that the dynamic signal is a normal signal (S165). In the process of determining whether the reference signal is more than a certain multiple of the reference signal, the multiple may be set to 1.4 to 1.6, preferably 1.5.

FIG. 7 is a graph showing the determination result of the motion noise according to the present invention. FIG. The PPG signal generating the motion noise can be confirmed in three short PPIs in FIG. 7 (a). After the reference signal is set and the basis function is set, the wavelet transform is performed so that the A [n] section in the range of 0 to 4 Hz and the D [n] section in the frequency range of 32 to 64 Hz appear, ] Period is equal to or more than 1.5 times the power of the corresponding section of the reference signal is determined as the period in which the dynamic noise occurs. The amplitude of the reference signal may not be sufficiently large because the reference signal is set at the initial signal of the signal measurement. Even if the user does not move, there may be a change of heartbeat due to psychological reasons such as change of emotions or tension. It is preferable to set it to 1.5 times.

As a result of the actual experiment, when the dynamic noise is judged based on 1.2 times, the normal signal is judged to be the motion noise as well as the motion noise, and when it is judged based on 1.7 times, the motion noise can not be normally discriminated Therefore, the experiment was conducted based on the most appropriate 1.5 times.

7 (b) is a result of adding 0 to 4 Hz, 7 (c) is a period of 32 to 64 Hz, and 7 (d) is a power of 32 to 64 Hz of each cycle in order from the beginning of one cycle to the next cycle. As shown in FIG. 7 (c), in the case of a normal pulse wave, the power of the D [n] portion appears to be very small, and a large value is shown when motion noise is included.

As shown in FIG. 7 (e), if it is determined that the dynamic noise is present, it is determined that the dynamic noise is not included in the output signal during the corresponding period In the case of judging, 1 is outputted during the corresponding period, and it can be confirmed that it is judged correctly when the motion noise is included.

8 is a graph showing a result of the determination of the dynamic noise according to whether or not the period is used according to the present invention. As a result of analyzing the signal without using the period, an error occurs that the motion noise is included but the motion noise is not included as shown in FIG. 8 (a). This is an error that occurs because the noise is included in the signal but the measured period is shorter than the period of the reference signal, so that the magnitude of the accumulated noise during that period is not larger than the reference signal. FIG. 8 (b) is a result of the dynamic noise judgment experiment using the period. As a result of adding the algorithm that the dynamic noise is included when the signal period is 0.3 or less, it can be confirmed that the signal includes the dynamic noise correctly. Accordingly, it is preferable to include a process of determining that a signal having a period equal to or shorter than a predetermined range includes motion noise (S166) as in S162 of FIG.

9 is a graph showing a result of motion noise determination according to the wavelet transform according to the present invention.

When the dynamic noise is determined using only the period, the dynamic noise is included as shown in FIG. 9 (a). However, when the peak appears at a constant interval, it is not accurately determined whether or not the dynamic noise is included. However, In the case of using all the wavelet transforms, the motion noise is more accurately judged than the result using the periodicity only. When the motion noise is judged in the signal containing the motion noise for 20% of the time, the experiment using the periodicity only includes the motion noise for 16.31% And that the experiment using both the period and wavelet transform included motion artifacts for 20.88% of the time. That is, it can be confirmed that the motion noise is relatively more accurately judged when both the period and the wavelet transform are used.

Table 2 below shows the result of determining whether motion noise is included in the measured pulse wave signal using a signal of a specific frequency band by separating the signal into various frequency bands using a base signal similar to the user's reference signal.

Experient A Experient B Experient C Experient D 20% 20.18 19.82 22.56 21.95 30% 29.95 30.46 31.67 31.24

The experimenter randomly included noise for 20% and 30% of the total signal measurement time, respectively. An error of not exactly 20% exists in the experimental result. However, this error occurs because it is determined that noise is included in one cycle including noise, and when there is motion noise at the beginning or end of one cycle, an error occurs do. The longer the measurement time, the less the effect is. Experiments A and B show the results for 200 seconds. Experiments C and D show the results for 100 seconds. As the measurement time increases, the error decreases.

Through the above process, it can be confirmed whether or not the motion noise occurring in various frequency bands is included in the signal.

The embodiments of the present invention described in the present specification and the configurations shown in the drawings relate to the most preferred embodiments of the present invention and are not intended to encompass all of the technical ideas of the present invention so that various equivalents It should be understood that water and variations may be present. Therefore, it is to be understood that the present invention is not limited to the above-described embodiments, and that various modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims. , Such changes shall be within the scope of the claims set forth in the claims.

Claims (11)

Measuring a PPG signal;
Generating a signal from which the high frequency noise is removed from the measured PPG signal;
Setting a reference signal in the signal from which the high-frequency noise is removed;
Selecting a base signal based on the set reference signal;
Performing wavelet transform using the base signal; And
Determining a motion noise using the wavelet-transformed value;
And determining a dynamic noise using the wavelet transform.
The method according to claim 1,
Wherein the step of generating the signal from which the high frequency noise is removed uses an FIR filter.
The method according to claim 1,
Wherein the reference signal selects a first period after a predetermined time elapses.
The method according to claim 1,
Wherein the base signal is selected by applying a cross-correlation technique to a plurality of base functions and a reference signal capable of performing discrete wavelet transform and comparing the maximum value of each cross-correlation technique, and selecting the largest value as a base signal Determination of motion noise using wavelet transform.
The method according to claim 1,
Wherein the step of performing the wavelet transform using the base signal repeats the wavelet transform so that the frequency of the low frequency band has a frequency band of 0 to 4 Hz.
6. The method of claim 5,
Wherein the wavelet transform is repeated a number of times smaller than the smallest a-3 in which the frequency of the low-frequency band satisfies 2 ^a-1 or less.
The method according to claim 1,
Wherein the determining of the motion noise comprises determining that motion noise is included when the detected signal period is less than or equal to a predetermined value.
The method according to claim 1,
Wherein the step of determining the motion noise comprises the step of determining whether a value obtained by summing the power of each predetermined period of the frequency band is equal to or greater than a predetermined multiple of the reference signal.
9. The method of claim 8,
Wherein the constant multiple is 1.4 to 1.6.
9. The method of claim 8,
Wherein the predetermined frequency band is 32 to 64 Hz.
The method according to claim 1,
Wherein the step of determining the dynamic noise includes the steps of: determining that the dynamic noise is included when the detected signal period is less than a predetermined value; determining whether the value obtained by adding the power per period of the predetermined frequency band is equal to or greater than a predetermined multiple And determining the motion noise using the wavelet transform.

Images