KR20190021631A - Wearable multichannel photo plethysmography measuring device using singular value decomposition and method for removing noise from a signal using the same - Google Patents

Abstract

Disclosed are a wearable multichannel photoelectric pulse wave measuring device using a singular value decomposition (SVD) and a method for removing noise from a signal using the same. According to an aspect of the present invention, the method for removing noise from a signal in a wearable multichannel photoelectric pulse wave measuring device comprises: a step of simultaneously collecting a plurality of PPG signals measured at different measurement sites by a sensing part; a step of removing noise by applying a cut SVD to the collected plurality of PPG signals; and a step of measuring a heart rate in the noise-removed signal.

Description

TECHNICAL FIELD The present invention relates to a wearable multi-channel optoelectronic pulse wave measuring device using singular value decomposition, and a method of removing noise from a signal using the same. BACKGROUND ART [0002]

The present invention relates to a wearable multichannel photoelectrochemical pulse wave measuring apparatus using singular value decomposition and a method for removing noise from a signal using the apparatus. More particularly, the present invention relates to a wearable multichannel photoelectrical pulse wave measuring apparatus using singular value decomposition (SVD) Channel multichannel photoelectrical pulse wave measuring device using singular value decomposition that minimizes the influence of motion artifacts and eliminating noise from a signal using the same.

The PPG signal means a pulse wave, which is a biological signal, and can be measured using a photoplethysmography. The PPG signal consists of direct current (DC) and alternating current (AC) components. The AC component represents arterial blood changes that occur between the systolic and diastolic phases of the cardiac cycle. The DC component corresponds to light intensity detected from non-pulsatile components of tissue, venous and arterial blood. At this time, PPG sensors for obtaining PPG signals are classified into transmission and reflectance types. A transmission type faces a pair of light emitting diodes (LEDs) and a photodetector. The light of the LED travels through absorbing material such as skin pigmentation, biological tissue bones, arterial blood and venous blood, and the photodetector receives the transmitted light and is quantized through a filter and an analog-to-digital converter (ADC). A representative example of the transfer type is the fingertip pulse oximeter, which is widely used clinically. On the other hand, in the reflection type, the pair of the LED and the photodetector are arranged on the same side. The light from the LED is reflected by the reflective material, the photodetector receives the reflected light, and the reflected light is similarly quantized through the filter and the ADC filter. The reflection type is mainly used when the material is thick (for example, the wrist and the forehead). Therefore, the reflection type can be various forms such as a band, a clock, a patch, and the like. As the need for a reflective type as a wearable sensor increases, a considerable number of wearable PPG sensors are available that can provide real-time heart rate. However, most reflective PPG sensors are limited in use due to their relatively high signal-to-noise ratio (SNR) caused by relatively large amounts of skin, muscle and fat and relatively small pulsatile components of arterial blood. Also, the reflective wearable PPG sensor had problems with motion artifacts. Therefore, many studies have been conducted to reduce the influence of these motion artifacts, one of which was to additionally measure the reference signals associated with motion artifacts. A typical example of this is to use an accelerometer. However, the accelerometer does not always adequately represent some type of real motion artifacts, such as finger tapping, wrist twisting, and fist spreading / unfolding. In fact, finger, wrist and fist movements have a significant impact on the signal quality of the wrist, that is, the PPG sensor. Other studies have attempted to reconstruct a damaged PPG signal using signal processing techniques such as principle component analysis (PCA), independent component analysis (ICA), and empirical mode decomposition (EMD) did. These algorithms work well in motion artifacts in many cases, but they can not be used for severely damaged PPG signals. Therefore, research is needed to overcome severely damaged motion artifacts.

(Non-Patent Document 1) J. Lee, " Motion artifacts reduction from PPG using cyclic moving average filter, " Technology and Health Care, vol. 22, pp. 409-417, 2014.

Disclosure of Invention Technical Problem [8] The present invention has been proposed in order to solve the above problems, and it is an object of the present invention to eliminate motion artifacts from a signal using singular value decomposition (SVD) The present invention provides a wearable multi-channel optoelectronic pulse wave measuring apparatus using singular value decomposition, which provides a wearable multi-channel photoelectrical pulse wave measuring apparatus, and a method of removing noise from a signal using the same.

Other objects and advantages of the present invention can be understood by the following description, and will be more clearly understood by one embodiment of the present invention. It will also be readily apparent that the objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims.

According to an aspect of the present invention, there is provided a method of removing noise from a signal in a wearable multi-channel photoelectrochemical pulse wave measurement device, the method comprising: sensing a plurality of PPG signals measured at different measurement sites; At the same time; Removing noise by applying cut singular value decomposition (SVD) to the collected plurality of PPG signals; And measuring a heart rate in the noise canceled signal.

The step of applying the singular value decomposition (SVD) to the plurality of collected PPG signals to remove noise includes: expressing all the collected channel signals by a two-dimensional matrix; Decomposing the two-dimensional matrix using SVD; And extracting the noise-removed signal by approximating the decomposed matrix.

In the step of representing all of the collected channel signals by a two-dimensional matrix, all of the collected channel signals are represented by a two-dimensional matrix as shown in the following Equation 1, and each of the rows corresponds to a pulse signal from each channel k .

[Equation 1]

Where k is the number of channels and N is the number of samples.

In the step of decomposing the two-dimensional matrix using the SVD, the two-dimensional matrix is decomposed using the SVD according to the following equation (2).

&Quot; (2) "

)는 행렬 P의 특이값임.Where U and V are the left and right singular vectors, respectively, and sigma

) Is the singular value of the matrix P.

In the step of extracting the noise-eliminated signal by approximating the decomposed matrix, the decomposed matrix is expressed as a singular value and a vector to be approximated as shown in Equation (3) below, And extracts a noise-removed signal from the matrix.

&Quot; (3) "

는 특이값,

,

는 각각 특이값의 좌우 특이 벡터임.here,

Lt; / RTI >

,

Are specific left and right specific vectors, respectively.

&Quot; (4) "

Where k is the number of channels and N is the number of samples.

에 대응하는 각각의 행 벡터는 각각의 k 번째 채널에서 잡음이 제거된 신호인 것을 특징으로 한다.remind

Is a signal in which noise is removed in each k-th channel.

The sensing unit includes a reflectance type PPG sensor in which pairs of LEDs and photodetectors are arranged on the same side.

According to another aspect of the present invention, there is provided a wearable multi-channel photoelectrochemical pulse wave measuring apparatus using singular value decomposition, comprising: a sensing unit for simultaneously collecting a plurality of PPG signals measured at different measurement sites; ; A noise eliminator for removing noise by applying cut singular value decomposition (SVD) to the collected plurality of PPG signals; And a heart rate measuring unit for measuring a heart rate in the noise canceled signal.

The sensing unit may include a reflectance type PPG sensor in which pairs of LEDs and photodetectors are disposed on the same side; An LED drive circuit including a metal oxide silicon field effect transistor and a digital-to-analog converter for controlling a current supplied to the LED; A transimpedance amplifier for converting the PPG signal obtained from the photodetector into a voltage signal; And an active filter for filtering the converted voltage signal, wherein the filtered signal is converted into digital data by an analog-to-digital converter embedded in the microcontroller unit.

The sensing unit may be a two-sided printed circuit board (PCB), a bottom layer may include an analog circuit on the top face, and a sensing device may be included on the bottom face. Wherein the top layer includes a user interaction component on the top face and includes a digital signal processing signal component on the bottom face.

In the bottom layer, nine multichannel PPG sensors made of a pair of a light sensor and an LED for directly contacting the wrist and measuring the PPG signal are disposed on the bottom face.

According to an aspect of the present invention, there is an effect that performance of a PPG signal can be improved by effectively removing motion artifacts from a signal using singular value decomposition (SVD). Accordingly, it is possible to accurately measure the heart rate even during high-intensity exercise.

The effects obtained in the present invention are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the following description .

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments of the invention and, together with the specific details for carrying out the invention, And shall not be construed as limited to the matters described.
1 is a schematic diagram of a wearable multi-channel optoelectronic pulse wave measuring apparatus according to an embodiment of the present invention;
2 is a diagram illustrating an example of the configuration of a sensing unit according to an embodiment of the present invention;
3 is a view showing an example of a double-sided printed circuit board according to an embodiment of the present invention,
FIG. 4 is a schematic flowchart of a method for removing noise from a signal using a wearable multi-channel optoelectronic pulse wave measuring apparatus according to an embodiment of the present invention.
5 is a view for explaining noise canceling effect of a cut SVD according to an embodiment of the present invention,
6 is a view for explaining noise canceling effect of a cut SVD according to another embodiment of the present invention,
FIG. 7 is a graph showing distribution of RMSE for each exercise intensity and each method according to an embodiment of the present invention, FIG.
Figure 8 shows an example segment of the results for various methods,
FIG. 9 is a diagram showing an RMSE distribution for a medium intensity exercise and a hard intensity exercise according to the number of used channels.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, in which: There will be. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

Throughout the specification, when an element is referred to as " comprising ", it means that it can include other elements as well, without excluding other elements unless specifically stated otherwise. In addition, the term "Quot; and " part " refer to a unit that processes at least one function or operation, which may be implemented in hardware, software, or a combination of hardware and software.

FIG. 1 is a schematic block diagram of a wearable multi-channel photoelectric pulse wave measuring apparatus according to an embodiment of the present invention; FIG. 2 is a diagram illustrating an example of a configuration of a sensing unit according to an embodiment of the present invention; 2 is a diagram showing an example of a double-sided printed circuit board according to an embodiment of the present invention.

Referring to FIG. 1, the wearable multi-channel photoelectrochemical pulse measurement apparatus 100 includes a sensing unit 110, a noise removing unit 120, and a heart rate measuring unit 130. The wearable multichannel photoelectrochemical measurement device 100 according to the present embodiment may be of a timepiece type.

The sensing unit 110 simultaneously collects a plurality of PPG signals measured at different measurement sites. The sensing unit 110 may include a multi-channel PPG sensor, and the multi-channel PPG sensor may be a multiple photo sensor. The multi-channel PPG sensor may be a reflectance type PPG sensor in which a pair of the LED 113 (see FIG. 2) and the photodetector 111 (see FIG. 2) are arranged on the same side. In the case of the photo sensor, as shown in FIG. 2, the photo detector 111 and the 570-nm LED 113 may emit green light. On the other hand, the photosensor may be NJL5303R. More specifically, referring to FIG. 2, nine photosensors may be disposed to perform sensing, and the distance between the photosensors may be 7 mm. At this time, each photo sensor may be a 1 mm protruding type coated with acryl, and the clock base may be painted black for optical and sweat shielding. The clock face can be printed through a 3D printer.

The sensing portion 110 includes an LED driving circuit including a metal oxide silicon field effect transistor and a digital-analog converter for controlling a current supplied to the LED 113. [ The LED driver circuit includes a metal oxide silicon field effect transistor (MOSFET) and a digital-to-analog converter (DAC) to control the current. The brightness of the LED 113 changes according to the value of the DAC, and the amplitude of the reflected PPG signal can be adjusted according to the brightness. Each PPG signal obtained from each photodetector 111 is converted into a voltage signal through trans-impedance amplifiers. Subsequently, the converted small voltage signal with noise is amplified and filtered through an analog filter. Each voltage signal is amplified and filtered using an active filter (MCP6004, Microchip) with cutoff frequencies of 0.5 and 10 Hz. The filtered signal is converted to digital data using a 12-bit analog-to-digital converter (ADC) embedded in a microcontroller unit (MCU, TM4C123GH6PMI, Texas Instruments). Digital data is converted to a 100 Hz sampling rate and stored on a micro SD card (SDSQUNB-016G, SanDisk). The data can also be communicated via Bluetooth (HM-11, JNHuaMao Technology). The Bluetooth communication function enables you to implement smartphone applications that can be used on Android and iPhone. These applications can communicate wirelessly with the developed sensor. The power required for the sensor is designed as a low dropout regulator with a 3.3V output.

More specifically, the sensing unit 110 may be configured as a two-layered double sided printed circuit boards, as shown in FIG. 3 (a) shows the top face and the bottom face of the bottom layer, and FIG. 3 (b) shows the top face of the top layer. ) And the bottom face. The bottom layer includes analog circuitry on the top face and includes a sensing element on the bottom face. The top layer includes a user interaction component on the top face and a digital signal processing signal component on the bottom face, as shown in Figure 3 (b). In the bottom layer, nine photosensors 311 associated with each pair of photodetector and 570nm (green) LEDs 113 are placed on the bottom face to measure the PPG signal in direct contact with the wrist. do. LED drive circuitry 312, nine active filters 313, and a board to board connector 314 are disposed on the top face. The filtered signal is transmitted to the top layer PCB via the board-to-board connector 314. A 3.7V 180mAh rechargeable lithium polymer battery (DTP 551430, DTP Battery) (not shown) is placed in the space between the top and bottom layers. A filtered signal at the bottom face of the top layer is input to the MCU 315 to perform noise cancellation on the measured nine PPG signals. A description of the noise cancellation method will be described in more detail with reference to FIG. The SD card 316 and the low voltage drop regulator 317 are also disposed on the lower surface. The top face of the top layer is provided with a Bluetooth module 318 for real-time monitoring of wireless communication, a PPG signal and HR (heart rate), a measurement start / stop process, an organic LED 113 OLED 113) screen 319 and a switch button are respectively arranged. A board-to-board connector 320 is also disposed on the bottom face of the top layer to connect to the board-to-board connector 314 of the top face of the bottom layer , And receives the filtered signal.

The noise removing unit 120 may remove noise by applying a singular value decomposition (SVD) to the plurality of PPG signals collected through the sensing unit 110.

The heart rate measuring unit 130 may measure the heart rate in the noise-canceled signal through the noise removing unit 120. At this time, the method of measuring the heart rate by the heart rate measuring unit 130 is the same as that of the known photoplethysmography method, so that detailed description thereof will be omitted in the present embodiment.

FIG. 4 is a schematic flowchart of a method for removing noise from a signal using a wearable multichannel photoelectrochemical pulse wave measuring apparatus according to an embodiment of the present invention. FIG. 5 is a flowchart illustrating a method for removing noise from a cut SVD according to an exemplary embodiment of the present invention. FIG. 6 is a view for explaining noise removing effect of a cut SVD according to another embodiment of the present invention. FIG. 7 is a view for explaining the effect of each exercise intensity and each method RMSE distribution in the second embodiment.

Referring to FIG. 4, the wearable multi-channel opto-electrical pulse-wave measuring apparatus 100 simultaneously collects a plurality of PPG signals measured at different measurement sites (S410). The wearable multichannel photoelectrochemical pulse wave measuring apparatus 100 collects a plurality of PPG signals simultaneously through nine photosensors that are worn on the subject's wrists and at different positions.

Thereafter, the wearable multi-channel optoelectronic pulse wave measuring apparatus 100 removes noise by applying cut singular value decomposition (SVD) to the collected plurality of PPG signals (S420). Hereinafter, a process of eliminating noise by applying the cut singular value decomposition (SVD) to a plurality of collected PPG signals according to the present embodiment will be described in detail.

≪ Process of removing noise by applying cut singular value decomposition (SVD) to a plurality of collected PPG signals >

)는 아래의 수학식 1과 같이 2차원 행렬(P)로 표현한다. In describing the present embodiment, all channel signals (

) Is expressed by a two-dimensional matrix P as shown in the following equation (1).

에 나타난다. 여기서, n은 샘플 시간이고, 채널수는 k(k=1,2, …, K)이다. 수학식 1에서, 각 행은 각 채널 k로부터의 펄스 신호에 해당하고, N은 샘플들의 수이다. 본 실시예에 따르면, 멀티 채널 PPG 센서에서 잡음을 제거하기 위해 9 채널에 각각 샘플링 속도 100(K=9 및 N=500)의 5초 창을 사용한다. 측정된 다중 펄스 신호에 특이값 분해(SVD)를 적용한다. 특이값 분해(SVD)의 주요 특징은 주기적 또는 준 주기적 프로세스를 분석하는데 사용할 수 있는 시스템을 구성하는 기본 구조 모드를 분리하는 것이다. At this time, each channel pulse signal

Lt; / RTI > Here, n is the sample time, and the number of channels is k (k = 1, 2, ..., K). In Equation (1), each row corresponds to a pulse signal from each channel k, and N is the number of samples. According to the present embodiment, a 5 second window having a sampling rate of 100 (K = 9 and N = 500) is used for 9 channels in order to remove noise from the multi-channel PPG sensor. Apply singular value decomposition (SVD) to the measured multi-pulse signals. A key feature of singular value decomposition (SVD) is the separation of the fundamental structure modes that make up a system that can be used to analyze periodic or quasi-periodic processes.

Using singular value decomposition (SVD), the matrix P expressed by Equation (1) can be decomposed as shown in Equation (2) below.

)는 행렬 P의 특이값에 해당하며, U와 V는 시그마(

)의 각각 왼쪽과 오른쪽의 특이 벡터이다. 보다 구체적으로,

의 고유 벡터는 U의 열을 구성하고,

의 고유 벡터는 V의 열을 구성한다. 여기서, 본 실시예에 따른 멀티 채널 PPG 센서의 경우, U와 V는 각각 500x500과 9x9 행렬이다. 시그마(

)의 특이값

는

또는

로부터의 고유값의 제곱근이며, 대각 행렬은 아래의 수학식 3과 같이 나타낼 수 있다.Here, sigma (

) Corresponds to the singular value of the matrix P, and U and V are sigma

) Are the left and right singular vectors, respectively. More specifically,

≪ / RTI > eigenvectors of < RTI ID = 0.0 &

≪ / RTI > constitutes a column of V. < RTI ID = In the case of the multi-channel PPG sensor according to the present embodiment, U and V are 500x500 and 9x9 matrices, respectively. Sigma

)

The

or

And the diagonal matrix can be expressed by Equation (3) below. &Quot; (3) "

)의 전체 랭크(full rank)에서 랭크(rank)를 감소시키고 P를 근사화한다. 감소된 랭크(rank)는 첫번째 가장 큰 특이값과 관련된 U와 V의 열을 사용한다. 절삭된(truncated) 특이값과 벡터를

,

및

로 나타낸다. 그러면, 행렬(P)을 아래의 수학식 4와 같이 근사화할 수 있다.U, V, and sigma

) ≪ / RTI > and approximates P at the full rank of < RTI ID = 0.0 > The reduced rank uses the columns of U and V associated with the first largest singular value. The truncated singular values and vector

,

And

Respectively. Then, the matrix P can be approximated as shown in Equation (4) below.

는 근사 또는 재구성된 P이며, 아래의 수학식 5에 의해 표현될 수 있다.here,

Is an approximate or reconstructed P, and can be expressed by the following equation (5).

에 대응하는 각각의 행 벡터는 각각의 k번째 채널에서 잡음이 제거된 신호가 된다. 상술한 수학식 4에서 절삭된 SVD의 효과를 설명하기 위해, 휴식 중에 본 실시예에 따른 멀티 채널 PPG 센서로 5초 동안 9개의 채널을 통해 손목의 펄스 신호를 동시에 측정했다. 도 5의 (a)는 9개의 채널 신호 중 하나를 나타낸다. 다음으로, 도 5의 (b)와 같이 5dB의 신호대 잡음비(SNR)로 각 채널 신호에 AWGN(additive white Gaussian noise)을 추가했다. SVD를 이용하여, U, V 및 시그마(

)를 얻었으며, 도 5의 (c) 에서

, 도 5 (d)에서

, 도 5 (e)에서

의 분해된 신호를 나타내었다. 이것을 통해, 랭크(rank) 1의 특이값과 특이 벡터가 깨끗한 신호를 주요 구성 요소로 제공함을 보여준다. 결과적인 특이값과 정보 에너지에 상응하는 값은 아래의 표 1에 나타내었다. At this time,

Each of the row vectors corresponding to the k < th > In order to illustrate the effect of the cut SVD in the above equation (4), the pulse signal of the wrist was simultaneously measured through 9 channels for 5 seconds with the multi-channel PPG sensor according to the present embodiment during resting. 5 (a) shows one of nine channel signals. Next, AWGN (additive white Gaussian noise) is added to each channel signal with a signal-to-noise ratio (SNR) of 5 dB as shown in FIG. 5 (b). Using SVD, U, V, and sigma (

), And in FIG. 5 (c)

, Fig. 5 (d)

, Fig. 5 (e)

And the decomposed signal of FIG. This shows that the singular values and singular vectors of rank 1 provide a clean signal as a key component. The resulting singular values and values corresponding to the information energies are shown in Table 1 below.

은

(n=1, 2,… 9)의 합이고, ER(energy ratio)은 각 정보 에너지(

)를 총 에너지(

)로 나눈값이다. 첫번째 특이값(

)의 정보 에너지 비율은 총 정보 에너지(

)의 78.94%였다. 이를 통해 볼 때, 결과적으로 잡음 제거가

에서만 효율적이라는 것을 의미한다. 지배적인 특이값(

)과 그 좌우의 특이 벡터

과

은 모든 채널을 통해 주성분을 형성한다. 다른 나머지 분해 신호(

내지

)도 순전히 잡음 신호이다. 따라서, 잡음 제거는 아래의 수학식 6과 같이 랭크(rank)가 1인 절삭된 SVD로 수행되는 것이 바람직하다.here,

silver

(n = 1, 2, ... 9), ER (energy ratio) is the sum of each information energy

) To total energy (

). The first singular value (

) Is the total information energy (

). As a result, noise cancellation

It is only effective. The dominant singular value (

) And its left and right singular vectors

and

Form the main component through all channels. Other remaining decomposition signals (

To

) Is also purely a noise signal. Therefore, it is preferable that the noise cancellation is performed with the cut SVD having a rank of 1 as shown in Equation (6) below.

은 스케일링 벡터

과 연관된다. 또한,

,

,

의 크기는 각각 500x1, 1x1, 500x1로 감소된다. 잡음 제거 성능을 비교하기 위해, 수동으로 최상의 품질 신호를 선택하고, 도 5의 (f)와 (g)에 도시된 바와 같이 밴드 패스 필터와 경험적 모델 분해(EMD)를 적용했다. 밴드 패스 필터는 0.5와 10Hz의 차단 주파수로 적용되었다. 경험적 모델 분해(EMD) 결과에 대해, 먼저 첫번째 고유모드를 얻고, 측정된 신호에서 그것을 뺐다. 절삭된 SVD의 더 나은 성분은 다중 소스에서 비롯된다. SVD는 한 쌍의 고유 벡터 U와 V및 관련된 특이값 시그마(

)로 분해한다. 시그마(

)는 스케일링 벡터(

)와 연관된 주성분(

)을 제공한다. 각 채널에서 다른 SNR 레벨에 대한 잡음 제거 효과를 더 자세히 조사하기 위해, 9 채널의 클린 펄스 신호를 유사하게 얻었으며, AWGN의 다른 레벨을 추가(

에서 -30dB,

에서 -25dB,

에서 -20dB,

에서 -15dB,

에서 -10dB,

에서 -5dB,

에서 0dB,

에서 5dB,

에서 10dB)했다. 상술한 수학식 5의 분해에 기초하여,

을 근사화하고, 도 6에 도시된 바와 같이 잡음이 손상된

과 비교했다. 도 5와 유사하게, 랭크(rank) 1의 절삭된 SVD는 모든 채널을 통해 잡음 제거된 신호를 제공했다. Here,

Is a scaling vector

Lt; / RTI > Also,

,

,

Are reduced to 500x1, 1x1 and 500x1, respectively. To compare the noise cancellation performance, we manually selected the best quality signal and applied a bandpass filter and empirical model decomposition (EMD) as shown in Figures 5 (f) and 5 (g). Bandpass filters were applied at cutoff frequencies of 0.5 and 10 Hz. For empirical model decomposition (EMD) results, we first get the first eigenmode, and we subtract it from the measured signal. The better components of the cut SVD originate from multiple sources. SVD is a pair of eigenvectors U and V and associated singular value sigma

). Sigma

) Is a scaling vector (

) ≪ / RTI >

). To further investigate the effect of noise reduction on different SNR levels in each channel, we obtained a similar 9-channel clean pulse signal and added another level of AWGN

To -30 dB,

To -25 dB,

To -20 dB,

To -15 dB,

To -10 dB,

-5dB,

0 dB,

5dB,

10dB). Based on the decomposition of the above-mentioned equation (5)

And as shown in FIG. 6,

. Similar to FIG. 5, the cut SVD of rank 1 provided a noise canceled signal over all channels.

You can also consider whether the motion induces false pulses in several channels, and add false pulses to the clean signal in only one of the nine channels to investigate the false pulse reduction. We changed the false pulse amplitude to 20%, 50%, and 100% of the clean pulse signal, respectively. The reconstructed signal generated from the cut SVD is shown in Figures 6 (b) to 6 (d), which illustrates the use of the decomposition feature from multiple sources and SVD to reduce the false pulse irrespective of the false pulse amplitude . Also, it can be confirmed that the number of simultaneously damaged channels is increased by the false pulse. For investigation, the false pulse amplitude was set to the same level (100%) as the clean signal and the damaged channel was increased to 3, as shown in Figure 6 (d). As a result of the cut SVD, the signal shows that the amplitude of the false pulse was reduced to 1% with one damaged channel, 7% with two damaged channels and 19% with three damaged channels.

Finally, the heart rate is measured in the noise-free signal (S430). In order to evaluate the heart rate measurement using the cut SVD based on the multi-channel PPG sensor according to this embodiment, an experiment was conducted on a modified version of the Bruce protocol. The Bruce protocol includes 5 minutes of walking for warm-up on a treadmill, 10 minutes of jogging at medium strength, 5 minutes of rest (walking), 10 minutes of intense jogging and 5 minutes of walking for cooling down. For the first session of medium strength jogging, the slope was 12 degrees and the speed was 4.0 km / h. In the second session of intense jogging, the slope and speed increased to 20 and 6.4 km / h, respectively. Since a reconstructed pulse signal generated from the cut SVD can be obtained from each channel, we chose one channel that provides a percent root mean square difference (PRD). PRD can be obtained as shown in Equation (7) below.

Random selection is also allowed since the reconstructed pulse signals of each channel are nearly superposed regardless of signal quality. With a reconstructed single pulse signal, pulse peaks were successively found by incorporating a filter bank with a variable cut-off frequency, a spectral estimate of the initial dominant frequency corresponding to HR, a rank-order nonlinear filter and decision logic. Electrocardiographic data were recorded simultaneously using a 24 hour Holter monitor (SEER Light, GE Healthcare, Milwaukee, WI, USA) to evaluate the measured heart rate (HR) values. For error analysis, root mean squared error (RMSE) was used for each subject defined as follows:

는 i번째 5초 데이터 세그먼트에서 본 실시예에 따른 센서 및 알고리즘으로부터 추정된 심박수(HR)(bpm)이고,

는 i번째 홀터의 심박수(HR) 5초 데이터 세그먼트에서 홀터로부터의 심박수(HR)(bpm)이다. 본 실시예에 따른 방법을 단일 채널 접근 방식(A), 다중 채널 가중 방식(B) 및 다중 채널 기반 최상의 채널 선택 방식(C) 등의 세가지 접근 방식과 비교했다. 단일 채널 방식(A)의 경우, 중간 포토 센서를 선택하고 분석을 수행했다. 다중 채널 가중 방식(B)의 경우, 교차 분석을 통해 9개의 채널 신호 모두의 신호 품질을 평가했으며, 각 채널은 각 펄스 템플릿을 사용하여 매 5초 데이터 세그먼트마다 계산되었다. 템플릿은 각 데이터 세그먼트에 대한 평균 비트 수를 사용하여 각 채널 신호에서 생성되었다. 그런 다음, 상관값을 정규화하고 각 값에 해당 신호를 곱한다. 다중 채널 기반 최상의 채널 선택 방식(C)의 경우, 다중 채널 잡음 레벨(MCNL)을 양자화하고, 5초 데이터 세그먼트마다 최소 모션 아티팩트(motion artifacts)가 있는 채널을 선택하기 위해 다중 채널 템플릿 매칭 알고리즘에 적용했다. 통계적 차이는 t-테스트(P<0.05)를 사용하였다. 도 7을 참조하면, 운동강도는 걷기, 중간 강도 운동 및 강렬한 운동으로 분류되며, 이에 대한 결과는 각각 도 7의 (a), (b), (c)에 나타내었다. 즉, 도 7의 (a) 내지 (c)에는 단일 채널 방법, 다중 채널 가중 방법, 다중 채널 최상의 신호 선택 방법 및 제안된 방법에서의 RMSE 분포가 도시되어 있다. 걷기는 워밍업, 휴식 및 쿨 다운 단계를 포함한다. 그래프의 위쪽과 아래쪽에 있는 다이아몬드는 5번째와 95번째 백분위 수를 나타내며, 위쪽과 아래쪽의 사각형은 10번째와 90번째 백분위 수를 나타낸다. 상단과 하단의 수염은 75번째와 25번째 백분위 수를 나타내며 원은 중간값을 나타낸다. 걷기에서 RMSE의 중앙값은 모든 방법에서 0.6bpm의 범위 이내였다. 단일 채널의 경우 5.10bpm, 다중 채널 가중 방법의 경우 4.88bpm, 다중 채널 최상의 신호 선택의 경우 5.23bpm 및 제안된 방법의 경우 4.63bpm이었다. 제안된 방법은 중간값이 가장 낮지만, 단일 방법을 제외하고는 통계적으로 유의하지 않다. 중간 강도 운동에서 제안된 방법은 단일 채널의 경우 7.35bpm, 다중 채널 가중 방법의 경우 6.12bpm, 다중 채널 최상의 신호 선택의 경우 7.14bpm, 제안된 방법의 경우 5.29bpm의 중간값을 제공한다. 제안된 방법의 결과는 모든 방법에 대해 통계적으로 중요하다. 강도 높은 운동에서 단일 채널의 경우 10.05bpm, 다중 채널 웨이팅 방법의 경우 6.79bpm, 다중 채널 최상의 신호 선택의 경우 7.41bpm를 나타내고, 제안 방법의 경우 5.79bpm의 최저 중간값도 제공한다. 평가를 위해 ECG Lead-I를 사용했다. 비록 제안된 방법을 이용하는 경우 ECG(심전도) 신호의 모션 아티팩트(motion artifacts)를 최소화했지만, 모션 아티팩트(motion artifacts)는 발견되었다. 불완전한 기준 신호는 비교적 높은 오차를 야기했다. 표 II는 각 운동 강도와 각 방법에 대한 RMSE의 중앙값과 표준 편차를 요약한 것이다.here,

(HR) (bpm) estimated from the sensor and algorithm according to the present embodiment in the i-th 5-second data segment,

HR is the heart rate HR of the i-th Holter (HR) (bpm) from the Holter in the 5-sec data segment. The method according to the present embodiment is compared with three approaches such as a single channel approach (A), a multi-channel weighting scheme (B) and a multi-channel based best channel selection scheme (C). For the single channel method (A), an intermediate photosensor was selected and analyzed. For the multi-channel weighting scheme (B), the signal quality of all nine channel signals was evaluated through a crosstalk analysis, and each channel was calculated for each 5 second data segment using each pulse template. The template was generated in each channel signal using the average number of bits for each data segment. Then, the correlation value is normalized and each value is multiplied by the corresponding signal. For multi-channel based best channel selection method (C), it is applied to multi-channel template matching algorithm to quantize multi-channel noise level (MCNL) and select the channel with minimum motion artifacts per 5 second data segment did. The statistical difference was the t-test (P <0.05). Referring to FIG. 7, the exercise intensity is classified into walking, middle intensity exercise, and intense exercise, and the results are shown in FIGS. 7 (a), 7 (b) and 7 (c). That is, FIGS. 7A to 7C show the RMSE distribution in the single channel method, the multi-channel weighting method, the multi-channel best signal selection method, and the proposed method. Walking includes warm-up, rest, and cooldown stages. The diamonds at the top and bottom of the graph represent the 5th and 95th percentiles, and the top and bottom squares represent the 10th and 90th percentiles. The beard at the top and bottom represents the 75th and 25th percentiles, and the circle represents the median. The median RMSE in walking was within 0.6 bpm for all methods. 5.10 bpm for single channel, 4.88 bpm for multi-channel weighted method, 5.23 bpm for multi-channel best signal selection and 4.63 bpm for the proposed method. The proposed method has the lowest median but is not statistically significant except for the single method. The proposed method for medium intensity motion provides a median of 7.35 bpm for a single channel, 6.12 bpm for a multi-channel weighting method, 7.14 bpm for a multi-channel best-case signaling, and 5.29 bpm for a proposed method. The results of the proposed method are statistically significant for all methods. For intensive motion, we show 10.05 bpm for a single channel, 6.79 bpm for a multi-channel weighting method, 7.41 bpm for a multi-channel best signal selection, and the lowest median of 5.79 bpm for the proposed method. I used ECG Lead-I for evaluation. Although the proposed method minimizes the motion artifacts of the ECG signal, motion artifacts have been found. Incomplete reference signals caused relatively high errors. Table II summarizes the median and standard deviation of each exercise intensity and RMSE for each method.

Table 2 above shows the average and standard deviation of the RMSE for walking, medium intensity and strong intensity motion as a result of comparing the performance of the single channel method, the multi channel weight method, the multi channel optimum channel selection method and the proposed method Is shown.

Figure 8 shows an example segment of the results for the various methods described above.

8 (a) to (i) show respective channel signals measured in the multi-channel PPG sensor according to the present embodiment. From this, it can be seen that the measured signal varies from channel to channel. 8 (j) to (l) show the results of the multi-channel weighting method, the multi-channel best signal selection method, and the proposed method, respectively. As a result, it can be seen that the proposed method provides a lower false pulse amplitude than the multi-channel weighted method. On the other hand, the multi-channel best signal selection method selected the channel incorrectly. This causes (i) of FIG. 8 that the ninth channel signal provides the highest correlation with the template signal. The template signal must take into account the pulse width. However, the pulse shape of amplitude and pulse width changes dynamically during motion. Thus, if only the optimal signal is selected, the time-varying pulse shape may result in less accurate motion artifacts reduction.

We also investigated the false pulse amplitude reduction in each method to evaluate the proposed method. Each false pulse amplitude and previous pure pulse amplitude was measured and the false pulse amplitude divided by the pure pulse amplitude. The average rate of the original signal is 0.24. For the multi-channel signal, the average ratio of the multi-channel weighting method, multi-channel best signal selection method, and proposed method was 0.23, 0.24 and 0.18. The proposed method reduced the false pulse amplitude by 25%.

In addition, the influence of the number of channels was investigated. 9 shows the RMSE distribution for the medium intensity exercise and the hard intensity exercise according to the number of used channels. As shown in FIG. 9, the increased channel decreased the RMSE value. In the medium intensity motion, when the number of channels increases to four, the error decreases (see Fig. 9 (a)). There was no statistical significance when using more than 4 channels. On the other hand, in the case of a high-intensity motion, the error continuously decreased as the number of channels increased to 9 (see Fig. 9 (b)).

According to the present embodiment as described above, a plurality of PPG signals measured at slightly different measurement sites through a homogeneous multi-channel PPG sensor are collected at the same time, and then, truncated singular value decomposition (SVD) was applied. As a result, it was confirmed that the heart rate evaluation accuracy was increased and the false pulse amplitude was decreased. In other words, according to the present embodiment, the performance of the PPG signal can be improved by effectively removing motion artifacts from the signal by using singular value decomposition (SVD), and accordingly, It is possible to measure the heart rate accurately.

The methods according to embodiments of the present invention may be implemented in an application or implemented in the form of program instructions that may be executed through various computer components and recorded on a computer readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, and the like, alone or in combination. The program instructions recorded on the computer-readable recording medium may be ones that are specially designed and configured for the present invention and are known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules for performing the processing according to the present invention, and vice versa.

While the specification contains many features, such features should not be construed as limiting the scope of the invention or the scope of the claims. In addition, the features described in the individual embodiments herein may be combined and implemented in a single embodiment. On the contrary, the various features described in the singular embodiments may be individually implemented in various embodiments or properly combined.

Although the operations are described in a particular order in the figures, it should be understood that such operations are performed in a particular order as shown, or that all described operations are performed in a series of sequential orders, or to obtain the desired result. In certain circumstances, multitasking and parallel processing may be advantageous. It should also be understood that the division of various system components in the above embodiments does not require such distinction in all embodiments. The above-described application components and systems can generally be packaged into a single software product or multiple software products.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention. The present invention is not limited to the drawings.

100: Wearable multichannel photoelectric pulse wave measuring device
110: sensing unit
111: Photodetector
113: LED
120: Noise canceling
130: heart rate measuring unit
311: Multi-channel PPG sensor
312: LED driving circuit
313: Active filter
314, 320: Board-to-board connector
315: MCU
316: SD card
317: Low Voltage Enhanced Regulator
318: Bluetooth module
319: OLED

Claims (11)

A method for removing noise from a signal in a wearable multi-channel photoelectrical pulse wave measuring apparatus,
Simultaneously collecting a plurality of PPG signals measured at different measurement sites by the sensing unit;
Removing noise by applying cut singular value decomposition (SVD) to the collected plurality of PPG signals; And
And measuring a heart rate in the noise canceled signal. The method according to claim 1,
The step of removing noise by applying cut singular value decomposition (SVD) to the collected plurality of PPG signals includes:
Expressing all the collected channel signals by a two-dimensional matrix;
Decomposing the two-dimensional matrix using SVD; And
And extracting the noise-removed signal by approximating the decomposed matrix to remove noise. 3. The method of claim 2,
In the step of representing all the collected channel signals by a two-dimensional matrix,
Wherein all the collected channel signals are represented by a two-dimensional matrix expressed by Equation (1) below, and each row corresponds to a pulse signal from each channel k.
[Equation 1]

Where k is the number of channels and N is the number of samples. The method of claim 3,
In the step of decomposing the two-dimensional matrix using SVD,
Wherein the two-dimensional matrix is decomposed using an SVD according to Equation (2) below.
&Quot; (2) &quot;

Where U and V are the left and right singular vectors, respectively, and sigma

) Is the singular value of the matrix P. 5. The method of claim 4,
In the step of approximating the decomposed matrix and extracting the noise-free signal,
The decomposed matrix is expressed as a singular value and a vector, and approximated as shown in Equation (3) below, and a noise-removed signal is extracted from a matrix expressed by Equation (4) below. How to remove.
&Quot; (3) &quot;

here,

Lt; / RTI &gt;

,

Are specific left and right specific vectors, respectively.
&Quot; (4) &quot;

Where k is the number of channels and N is the number of samples. 6. The method of claim 5,
remind

&Lt; / RTI &gt; wherein each row vector corresponding to the kth channel is a noise canceled signal at each kth channel. The method according to claim 1,
The sensing unit includes:
And a reflectance type PPG sensor in which a pair of LEDs and a photodetector are disposed on the same side. A sensing unit for simultaneously collecting a plurality of PPG signals measured at different measurement sites;
A noise eliminator for removing noise by applying cut singular value decomposition (SVD) to the collected plurality of PPG signals; And
And a heart rate measuring unit for measuring heart rate in the noise canceled signal. The wearable multi-channel photoelectrical pulse wave measuring apparatus using singular value decomposition. 9. The method of claim 8,
The sensing unit may include a reflectance type PPG sensor in which pairs of LEDs and photodetectors are disposed on the same side;
An LED drive circuit including a metal oxide silicon field effect transistor and a digital-to-analog converter for controlling a current supplied to the LED;
A transimpedance amplifier for converting the PPG signal obtained from the photodetector into a voltage signal; And
And an active filter for filtering the converted voltage signal,
Wherein the filtered signal is converted into digital data by an analog-to-digital converter built in a microcontroller unit. 10. The method of claim 9,
The sensing unit is provided as a two-sided printed circuit board (PCB)
The bottom layer includes analog circuitry on the top face and includes a sensing element on the bottom face,
Characterized in that the top layer comprises a user interaction component on the top face and the digital signal processing signal component is on the bottom face. Multichannel photoelectric pulse wave measuring device. 11. The method of claim 10,
In the bottom layer, nine multichannel PPG sensors are arranged on the bottom face, each consisting of a pair of photodetectors and LEDs for measuring the PPG signal in direct contact with the wrist. Wearable multichannel photoelectrical pulse wave measuring device using decomposition.

Images