Signal / noise detection device, method, program, recording medium
The signal/noise determination device uses machine learning and sensor data assumptions to simplify training data acquisition and enhance signal/noise separation accuracy, addressing the challenge of large data requirements in neural networks.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- ADVANTEST CORP
- Filing Date
- 2022-06-28
- Publication Date
- 2026-05-26
AI Technical Summary
Existing neural network-based signal/noise separation methods require large amounts of training data, which are difficult to obtain.
A signal/noise determination device and method using a determination model generated by machine learning, which determines whether each component of the measurement result originates from a signal or noise source, utilizing a plurality of sensors and assuming signal and noise positions as training data, and employing techniques like independent component analysis, FFT, and convolutional neural networks.
Enables easy acquisition of training data and accurate determination of signal and noise components, improving the efficiency and accuracy of signal/noise separation.
Smart Images
Figure 0007865806000007 
Figure 0007865806000008 
Figure 0007865806000009
Abstract
Description
Technical Field
[0001] The present invention relates to the measurement of signals and noise.
Background Art
[0002] Conventionally, it has been known to separate a measured signal into a signal component from a signal source of interest and a signal component from an environmental magnetic noise source (see, for example, Patent Documents 1 to 5 and Non-Patent Document 1). Among them, a method using a neural network is known (see, for example, Patent Documents 3 to 5 and Non-Patent Document 1).
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Patent Document 2
Patent Document 3
Patent Document 4
Patent Document 5
Non-Patent Documents
[0004]
Non-Patent Document 1
[0005] However, using neural networks requires a large amount of training data, which is difficult to obtain.
[0006] Therefore, the present invention aims to easily acquire training data when machine learning is used to determine whether each component of the signal measurement result originates from a signal source or a noise source. [Means for solving the problem]
[0007] The signal / noise determination device according to the present invention comprises: a plurality of sensors for measuring signals and noise; a determination model recording unit that records a determination model generated by machine learning, which determines whether each component of the measurement result originates from a signal source or a noise source, using the measurement results of the sensors assumed when signal information and noise information are assumed, the assumed signal information, and the assumed noise information as training data; and a signal / noise determination unit that determines whether each component of the measurement result originates from a signal source or a noise source based on the measurement result and the determination model, wherein the signal information is the position of the signal source and the signal, and the noise information is the position of the noise source and the noise.
[0008] According to the signal / noise determination device configured as described above, multiple sensors measure signals and noise. The determination model recording unit records a determination model generated by machine learning, which determines whether each component of the measurement result originates from a signal source or a noise source, using the assumed measurement results of the sensors, the assumed signal information, and the assumed noise information as training data. Based on the measurement result and the determination model, the signal / noise determination unit determines whether each component of the measurement result originates from a signal source or a noise source. The signal information is the position of the signal source and the signal. The noise information is the position of the noise source and the noise.
[0009] Furthermore, the signal / noise determination device according to the present invention may be configured such that the signal source is assumed to be located within a predetermined single region, and the noise source is assumed to be located around the region.
[0010] Furthermore, the signal / noise determination device according to the present invention may be configured to assume that the signal source or the noise source is randomly arranged.
[0011] Furthermore, the signal / noise determination device according to the present invention may be configured such that the components are obtained by performing independent component analysis on the measurement results.
[0012] Furthermore, the signal / noise determination device according to the present invention may be configured such that the components are obtained by performing principal component analysis on the measurement results.
[0013] Furthermore, the signal / noise determination device according to the present invention may be configured such that the aforementioned components are obtained by performing an FFT on the measurement result, and then performing an IFFT on the result of the FFT.
[0014] Furthermore, the signal / noise determination device according to the present invention may use a convolutional neural network for machine learning.
[0015] Furthermore, the signal / noise determination device according to the present invention may use a sigmoid function as the activation function of the output layer of the convolutional neural network.
[0016] Furthermore, the signal / noise detection device according to the present invention may use the Softmax function as the activation function for the output layer of the convolutional neural network.
[0017] Furthermore, the signal / noise determination device according to the present invention may be configured such that the signal is expressed as a magnetic dipole moment or a current dipole moment.
[0018] The present invention relates to a signal / noise determination method for determining the origin of components of measurement results of a sensor using a signal / noise determination device having a plurality of sensors for measuring signals and noise, comprising: a determination model recording step of recording a determination model generated by machine learning, which determines whether each component of the measurement result originates from a signal source or a noise source, using the assumed measurement results of the sensor when signal information and noise information are assumed, the assumed signal information, and the assumed noise information as training data; and a signal / noise determination step of determining whether each component of the measurement result originates from a signal source or a noise source based on the measurement result and the determination model, wherein the signal information is the position of the signal source and the signal, and the noise information is the position of the noise source and the noise.
[0019] The present invention is a program for causing a computer to execute a signal / noise determination process for determining the origin of components of measurement results of sensors, using a signal / noise determination device having a plurality of sensors for measuring signals and noise, the signal / noise determination process including: a determination model recording step of recording, as teacher data, a determination model generated by machine learning for determining whether each of the components of the measurement results is derived from a signal source or a noise source, the determination model being based on the measurement results of the sensors assumed when signal information and noise information are assumed, the assumed signal information, and the assumed noise information; and a signal / noise determination step of determining whether each of the components of the measurement results is derived from the signal source or the noise source, based on the measurement results and the determination model, wherein the signal information includes the position of the signal source and the signal, and the noise information includes the position of the noise source and the noise.
[0020] The present invention is a computer-readable recording medium recording a program for causing a computer to execute a signal / noise determination process for determining the origin of components of measurement results of sensors, using a signal / noise determination device having a plurality of sensors for measuring signals and noise, the signal / noise determination process including: a determination model recording step of recording, as teacher data, a determination model generated by machine learning for determining whether each of the components of the measurement results is derived from a signal source or a noise source, the determination model being based on the measurement results of the sensors assumed when signal information and noise information are assumed, the assumed signal information, and the assumed noise information; and a signal / noise determination step of determining whether each of the components of the measurement results is derived from the signal source or the noise source, based on the measurement results and the determination model, wherein the signal information includes the position of the signal source and the signal, and the noise information includes the position of the noise source and the noise.
Brief Description of the Drawings
[0021] [Figure 1] It is a functional block diagram showing the configuration of a signal / noise determination device 1 according to an embodiment of the present invention. [Figure 2] It is a plan view of the sensor group 12. [Figure 3] It is a diagram showing the positional relationship between the sensor group 12, the signal source region S, and the noise source region N, and is a plan view (Fig. 3(a)) and a front view (Fig. 3(b)) of the sensor group 12 and the like.
Embodiments for Carrying Out the Invention
[0022] Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0023] Fig. 1 is a functional block diagram showing the configuration of a signal / noise determination device 1 according to an embodiment of the present invention. Fig. 2 is a plan view of the sensor group 12. The signal / noise determination device 1 according to an embodiment of the present invention includes a sensor group 12, a determination model recording unit 14, and a signal / noise determination unit 16.
[0024] The sensor group 12 is a plurality of sensors 12a. The plurality of sensors 12a measure signals and noise. The signals and noise are represented, for example, as magnetic dipole moments or current dipole moments. Referring to Fig. 2, the plurality of sensors 12a are two-dimensionally arranged at equal intervals, for example, vertically and horizontally.
[0025] Fig. 3 is a diagram showing the positional relationship between the sensor group 12, the signal source region S, and the noise source region N, and is a plan view (Fig. 3(a)) and a front view (Fig. 3(b)) of the sensor group 12 and the like.
[0026] The signal sources S1 and S2 output signals. The noise sources N1 and N2 output noise. In the example shown in Fig. 3, there are two signal sources and two noise sources each, but there may be three or more signal sources, and there may be one or more noise sources. However, the total number of signal sources and noise sources needs to be less than the number of sensors 12a.
[0027] The signal source region S is a single, predetermined region located almost directly above the sensor group 12. Signal sources S1 and S2 are present in the signal source region S. The noise source region N is located around the signal source region S. Noise sources N1 and N2 are present in the noise source region N.
[0028] The determination model recording unit 14 records a determination model that determines whether each component of the measurement results of the sensor group 12 (multiple sensors 12a) originates from signal sources S1 and S2 or noise sources N1 and N2. The method for generating the determination model will be described later.
[0029] The signal / noise determination unit 16 determines, based on the measurement results of the sensor group 12 and the determination model recorded in the determination model recording unit 14, whether each component of the measurement result originates from the signal sources S1 and S2 or from the noise sources N1 and N2.
[0030] The signal / noise determination unit 16 receives measurement results from the sensor group 12 and obtains the components of the measurement results by performing independent component analysis or principal component analysis. Alternatively, the signal / noise determination unit 16 may obtain the components of the measurement results by performing an FFT on the measurement results and then an IFFT on the result of the FFT.
[0031] In order to perform independent component analysis, the number of components in the measurement result (i.e., the total number of signal sources and noise sources) must be known. If the number of components in the measurement result is unknown, it is necessary to estimate the number of components in the measurement result. Here, if eigenvalues (singular values) are calculated from the covariance matrix of the measurement result and arranged in descending order, the number of eigenvalues will be as large as the number of components in the measurement result. Therefore, in the signal / noise determination unit 16, by calculating eigenvalues (singular values) from the covariance matrix of the measurement result and determining the number of eigenvalues that are above a predetermined threshold, the number of components in the measurement result can be estimated.
[0032] The decision model is generated by machine learning (for example, using a convolutional neural network) using the assumed measurement results (so-called simulation) of sensor 12a, which are assumed to be based on the assumption of signal information (i.e., the location and signal of the signal source) and noise information (i.e., the location and noise of the noise source), along with the assumed signal information and assumed noise information, as training data.
[0033] In generating the decision model, it is assumed that the signal sources are randomly placed within the signal source region S, and that the noise sources are randomly placed within the noise source region N.
[0034] Here, when the signal and noise are represented as magnetic dipole moments (vector m), the magnetic flux density B (a function of vector r) generated at sensor 12a is expressed by the Biot-Savart law as shown in equation (1) below. Here, μ0 is the magnetic constant. Also, vector r is the direction vector from the signal source and noise source (magnetic dipole) to sensor 12a.
[0035]
number
[0036]
number
[0037] Furthermore, when the signal and noise are expressed as current dipole moments (vector p), the magnetic flux density B (a function of vector r) generated in sensor 12a is expressed by the Biot-Savart law as shown in equation (5) below.
[0038]
number
[0039]
number
[0040] Furthermore, the magnetic field generated by a coil with radius a, number of turns n, and current I can be expressed in cylindrical coordinates as follows: equations (9), (10), and (11).
[0041]
number
[0042]
number
[0043] Up until now, the sensor group 12 has been assumed to consist of sensors 12a capable of measuring components in three axes, arranged in two dimensions. However, it is also possible that the sensors 12a are arranged in a line in one dimension. Furthermore, it is also possible that the sensors 12a are arranged three-dimensionally (as a sensor array). In addition, the sensors 12a may be capable of measuring only components in one axis, or even only components in two axes.
[0044] Furthermore, when a convolutional neural network is used as a machine learning method, the decision model comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. When the measurement result of sensor 12a is provided to the input layer, the output of the input layer is provided to the convolutional layer, the output of the convolutional layer is provided to the pooling layer, the output of the pooling layer is provided to the fully connected layer, and the output of the fully connected layer is provided to the output layer. The output layer outputs whether each component of the measurement result originates from a signal source (output 1) or from a noise source (output 0).
[0045] As described above, one convolutional layer and one pooling layer may be provided, but depending on the number of sensors 12a, two or more convolutional layers and pooling layers may be provided.
[0046] Furthermore, a sigmoid function is used as the activation function for the output layer. However, a Softmax function may also be used as the activation function for the output layer. In this case, since a multi-level output is obtained, the approximate distances of the signal source and noise source from sensor 12a can also be determined. This improves the accuracy of estimating the positions of signal sources S1 and S2 using the measurement results of sensor 12a.
[0047] Next, the operation of the embodiment of the present invention will be described.
[0048] First, a decision model is created and recorded in the decision model recording unit 14. The decision model is created using a convolutional neural network. As training data, the measurement results of sensor 12a (so-called simulation) assumed when signal information (i.e., the position and signal of the signal source) and noise information (i.e., the position and noise of the noise source) are assumed, along with assumed signal information and assumed noise information.
[0049] Subsequently, the actual signal sources S1 and S2 and the actual noise sources N1 and N2 (see Figure 3) are measured by the sensor group 12.
[0050] Measurement results from sensor 12a of sensor group 12 are provided to signal / noise determination unit 16. The components of the measurement results are obtained by independent component analysis or the like. Furthermore, based on the measurement results of sensor group 12 and the determination model recorded in determination model recording unit 14, it is determined whether each component of the measurement results originates from signal sources S1 and S2 or from noise sources N1 and N2.
[0051] According to an embodiment of the present invention, when machine learning is used to determine whether each component of the signal measurement result originates from signal sources S1 and S2 or noise sources N1 and N2, the measurement results of the sensor 12a assumed when signal information (i.e., the location and signal of the signal source) and noise information (i.e., the location and noise of the noise source) are assumed (so-called simulation), and the assumed signal information and assumed noise information are used as training data. Therefore, according to an embodiment of the present invention, training data can be acquired more easily than when actual measurement results are used as training data.
[0052] Furthermore, the above embodiment can be realized as follows: A computer equipped with a CPU, a hard disk, and a media (USB memory, CD-ROM, etc.) reader is made to read a media containing a program that implements each of the above parts, for example, the judgment model recording unit 14 and the signal / noise judgment unit 16, and install it on the hard disk. The above functions can also be realized by this method. [Explanation of Symbols]
[0053] 1. Signal / Noise Detection Device 12 sensor group 12a sensor 14. Judgment Model Recording Unit 16 Signal / Noise Detection Unit S1, S2 signal source N1, N2 Noise Sources S signal source area N Noise source region
Claims
1. Multiple sensors for measuring signals and noise, A determination model recording unit records a determination model generated by machine learning, which determines whether each component of the measurement result originates from a signal source or a noise source, using the assumed measurement result of the sensor, the assumed signal information, and the assumed noise information as training data. A signal / noise determination unit that determines whether each component of the measurement result originates from the signal source or the noise source, based on the measurement result and the determination model, Equipped with, The signal information is the position of the signal source and the signal, The noise information is the location of the noise source and the noise, Signal and noise detection device.
2. A signal / noise determination device according to claim 1, It is assumed that the signal source is located within a predetermined single region, A signal-noise detection device in which the noise source is assumed to be located around the region.
3. A signal / noise determination device according to claim 2, A signal / noise determination device in which the signal source or the noise source is assumed to be randomly arranged.
4. A signal / noise determination device according to claim 1, A signal / noise determination device in which the aforementioned components are obtained by performing independent component analysis on the measurement results.
5. A signal / noise determination device according to claim 1, A signal / noise determination device in which the aforementioned components are obtained by performing principal component analysis on the measurement results.
6. A signal / noise determination device according to claim 1, A signal / noise determination device in which the aforementioned components are obtained by performing an FFT on the measurement result and then performing an IFFT on the result of the FFT.
7. A signal / noise determination device according to claim 1, The aforementioned machine learning is a signal / noise detection device that uses a convolutional neural network.
8. A signal / noise determination device according to claim 7, A signal-to-noise detection device that uses a sigmoid function as the activation function for the output layer of the aforementioned convolutional neural network.
9. A signal / noise determination device according to claim 7, A signal-to-noise detection device that uses the Softmax function as the activation function for the output layer of the aforementioned convolutional neural network.
10. A signal / noise determination device according to any one of claims 1 to 9, A signal / noise determination device in which the aforementioned signal is represented as a magnetic dipole moment or an electric dipole moment.
11. A signal / noise determination method for determining the origin of components of measurement results from multiple sensors, using a signal / noise determination device having multiple sensors for measuring signals and noise, A determination model recording step involves recording a determination model, which is generated by machine learning using the assumed measurement results of the sensor, the assumed signal information, and the assumed noise information as training data, to determine whether each component of the measurement results originates from a signal source or a noise source. A signal / noise determination step, based on the measurement results and the determination model, determines whether each component of the measurement results originates from the signal source or the noise source. Equipped with, The signal information is the position of the signal source and the signal, The noise information is the location of the noise source and the noise, Signal / noise detection method.
12. A program for causing a computer to perform a signal / noise determination process to determine the origin of components of the measurement results of a signal / noise determination device having multiple sensors for measuring signals and noise, The aforementioned signal / noise determination process, A determination model recording step involves recording a determination model, which is generated by machine learning using the assumed measurement results of the sensor, the assumed signal information, and the assumed noise information as training data, to determine whether each component of the measurement results originates from a signal source or a noise source. A signal / noise determination step, based on the measurement results and the determination model, determines whether each component of the measurement results originates from the signal source or the noise source. Equipped with, The signal information is the position of the signal source and the signal, The noise information is the location of the noise source and the noise, program.
13. A computer-readable recording medium containing a program for causing a computer to perform a signal / noise determination process to determine the origin of components of the measurement results of a signal / noise determination device having multiple sensors for measuring signals and noise, The aforementioned signal / noise determination process, A determination model recording step involves recording a determination model, which is generated by machine learning using the assumed measurement results of the sensor, the assumed signal information, and the assumed noise information as training data, to determine whether each component of the measurement results originates from a signal source or a noise source. A signal / noise determination step, based on the measurement results and the determination model, determines whether each component of the measurement results originates from the signal source or the noise source. Equipped with, The signal information is the position of the signal source and the signal, The noise information is the location of the noise source and the noise, Recording medium.