KR102274563B1 - Method and apparatus for signal classification - Google Patents

Abstract

Obtaining a plurality of cumulant values from a modulated signal, generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values, and inputting the two-dimensional image to a neural network A signal classification method for determining a modulation method is disclosed.

Description

Method and apparatus for signal classification

It relates to a signal classification method and apparatus. Specifically, it relates to a method and apparatus for classifying a modulation scheme of a signal according to a signal-to-noise ratio (SNR) of a received signal.

Automatic Modulation Classification (AMC) technology is a technology for predicting a modulation method of a received signal without prior information of any system in wireless communication.

As an approach of the AMC technology, there is a method of extracting a feature value from already obtained data and performing machine learning using the extracted feature value for classification of a modulation method.

In this case, the method of performing machine learning using the constellation image has a problem in that the classification performance is deteriorated because the constellation images are similar in a low SNR environment.

In addition, in the conventional method of performing machine learning using a cumulant, an absolute value is taken from the cumulant and used as an input of a signal modulation method classification device. Although the cumulant values are different, the absolute values are different. There was a problem that the values were the same when taking , so that an error could occur in the classification of the modulation method.

To provide a signal classification method and apparatus.

The technical problem to be achieved by the present embodiment is not limited to the technical problems as described above, and other technical problems may be inferred from the following embodiments.

As a means for solving the above technical problem, according to one aspect, a signal classification method includes: obtaining a plurality of cumulant values from a modulated signal; generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values; and inputting the 2D image into a neural network to determine a modulation method of the modulated signal.

In the 2D image, a real value and an imaginary value of each of the plurality of cumulant values may be displayed on a complex plane.

The generating of the two-dimensional image may include limiting the real and imaginary axes included in the two-dimensional image to a predetermined range, respectively, based on the distribution characteristics of the plurality of cumulant values displayed on the two-dimensional image. can

The determining of the modulation scheme of the modulated signal may include classifying the modulated signal into at least one modulation scheme among a phase shift keying (PSK) series, a pulse amplitude modulation (PAM) series, and a quadrature amplitude modulation (QAM) series.

According to another aspect, there is provided a computer-readable recording medium having recorded thereon a program for implementing a signal classification method, the method comprising: obtaining a plurality of cumulant values from a modulated signal; generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values; and inputting the 2D image into a neural network to determine a modulation method of the modulated signal.

According to another aspect, in a computer program stored in a medium for executing a signal classification method in combination with hardware, the method comprising: obtaining a plurality of cumulant values from a modulated signal; generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values; and inputting the 2D image into a neural network to determine a modulation method of the modulated signal.

According to another aspect, a signal classification apparatus includes a memory for storing one or more instructions; and by executing the one or more instructions, obtaining a plurality of cumulant values from a modulated signal, generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values, and generating the two-dimensional image and a processor for determining a modulation method of the modulated signal by input to the neural network.

In the 2D image, a real value and an imaginary value of each of the plurality of cumulant values may be displayed on a complex plane.

The processor may limit a range of a real axis and an imaginary axis included in the 2D image to a predetermined range, respectively, based on a distribution characteristic of the plurality of cumulant values displayed on the 2D image.

The processor may classify the modulated signal into at least one modulation scheme among a phase shift keying (PSK) sequence, a pulse amplitude modulation (PAM) sequence, and a quadrature amplitude modulation (QAM) sequence.

1 is a block diagram illustrating an example of a hardware configuration of a signal classification apparatus.
2 is a diagram for explaining an example of an architecture of a convolutional neural network.
3 is a diagram for describing an example of a convolution operation in a convolutional neural network.
4 is a diagram for explaining an example of a method of generating a two-dimensional image based on a cumulant value of a modulated signal.
5 is a diagram for explaining an example of training data for training a convolutional neural network.
6 is a flowchart illustrating an example of a process in which a signal modulation scheme is classified.

The terms used in the present embodiments were selected as currently widely used general terms as possible while considering the functions in the present embodiments, which may vary depending on the intention or precedent of a person skilled in the art, the emergence of new technology, etc. can In addition, there are also arbitrarily selected terms in a specific case, and in this case, the meaning will be described in detail in the description of the embodiment. Therefore, the terms used in the present embodiments should be defined based on the meaning of the term and the overall contents of the present embodiments, rather than the simple name of the term.

In the descriptions of the embodiments, when a certain part is connected to another part, this includes not only a case in which it is directly connected, but also a case in which it is electrically connected with another component interposed therebetween. . Also, when it is said that a part includes a certain component, this means that other components may be further included, rather than excluding other components, unless otherwise stated.

Terms such as “consisting of” or “comprising” used in the present embodiments should not be construed as necessarily including all of the various components or various steps described in the specification, and some components or It should be construed that some steps may not be included, or may further include additional components or steps.

The description of the following embodiments should not be construed as limiting the scope of rights, and what can be easily inferred by those skilled in the art should be construed as belonging to the scope of the embodiments. Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an example of a hardware configuration of a signal classification apparatus.

The signal classification apparatus 100 may be implemented as various types of devices such as a personal computer (PC), a server device, a mobile device, and an embedded device. Furthermore, the signal classification apparatus 100 may correspond to a dedicated hardware accelerator mounted on the device as described above, and the signal classification apparatus 100 is an NPU (neural processing unit), which is a dedicated module for driving a neural network, It may be a hardware accelerator such as a Tensor Processing Unit (TPU), a Neural Engine, or the like, but is not limited thereto.

Referring to FIG. 1 , the signal classification apparatus 100 includes a processor 110 and a memory 120 . In the signal classification apparatus 100 shown in FIG. 1, only the components related to the present embodiments are shown. Accordingly, it is apparent to those skilled in the art that the signal classification apparatus 100 may further include other general-purpose components in addition to the components shown in FIG. 1 .

The processor 110 serves to control overall functions for executing the signal classification apparatus 100 . For example, the processor 110 generally controls the signal classification apparatus 100 by executing one or more instructions or programs stored in the memory 120 in the signal classification apparatus 100 . The processor 110 may be implemented as a central processing unit (CPU), a graphics processing unit (GPU), an application processor (AP), etc. provided in the signal classification apparatus 100 , but is not limited thereto.

The memory 120 is hardware for storing various data processed in the signal classification apparatus 100 , and for example, the memory 120 may store the data processed by the signal classification apparatus 100 and data to be processed. have. In addition, the memory 120 may store applications, drivers, etc. to be driven by the signal classification apparatus 100 . The memory 120 includes random access memory (RAM), such as dynamic random access memory (DRAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), CD- It may include ROM, Blu-ray or other optical disk storage, a hard disk drive (HDD), a solid state drive (SSD), or flash memory 120 .

The processor 110 may classify the modulation method of the received modulated signal using a neural network. A neural network refers to an overall model that has problem-solving ability by changing the strength of synaptic bonding through learning, in which artificial neurons that form a network by combining synapses.

Specifically, the processor 110 may effectively classify the modulation scheme of the modulated signal using a deep neural network, which is a neural network structure having a plurality of hidden layers.

The processor 110 may input a two-dimensional image representing characteristics of the received modulated signal to the neural network. The 2D image may be generated by the processor 110 obtaining a plurality of cumulant values representing characteristics of the modulated signal from the modulated signal, and based on complex plane coordinate information of each of the plurality of cumulant values.

The processor 110 may train the neural network using learning data generated for each modulation scheme in various SNR environments so that the neural network can classify the modulation scheme of the modulated signal based on the two-dimensional image received by the neural network. The training data may be composed of a two-dimensional image corresponding to a specific modulation scheme in a corresponding SNR environment and a label value indicating a specific modulation scheme.

When the 2D image representing the characteristics of the modulated signal is input to the neural network, the processor 110 may use convolutional neural networks, which are algorithms suitable for image classification. Hereinafter, an example of a convolutional neural network will be described with reference to FIG. 2 .

2 is a diagram for explaining an example of an architecture of a convolutional neural network.

Referring to FIG. 2 , the convolutional neural network 2 may include a pooling layer and a fully connected layer in addition to the convolutional layer.

The convolutional neural network 2 may be implemented as an architecture having multiple layers including an input image, feature maps and an output. In the convolutional neural network 2, a convolution operation is performed on an input image with a filter called a kernel, and as a result, feature maps are output. In this case, the generated output feature maps are input feature maps, and a convolution operation with the kernel is performed again, and new feature maps are output. As a result of repeatedly performing such a convolution operation, a result of recognizing the features of the input image through the convolutional neural network 2 may be finally output.

For example, when an image with a size of 32x32 pixels is input to the convolutional neural network 2 of FIG. 2, the input image is output as 6 feature maps with a size of 28x28 pixels through a convolution operation with a kernel having a size of 5x5. can be Thereafter, 6 feature maps having a size of 28×28 pixels may be output as 6 feature maps having a size of 14×14 through sub-sampling (pooling) in the pooling layer. Thereafter, the size of the 14x14 feature maps may be reduced through iterative convolution with the kernel, and finally features of 1x1 pixel size may be output.

The convolutional neural network 2 filters and outputs robust features that can represent the entire image from the input image by repeatedly performing convolutional operations and sub-sampling (or pooling) operations in several layers, and outputs the output final features. Through this, the recognition result of the input image can be derived.

In the convolutional neural network (2), when a meaningful feature is extracted by a convolution operation in the convolutional layer and a sub-sampling (or pooling) operation in the pooling layer, the recognition or classification operation for the feature is performed in the fully connected layer can be

Meanwhile, an output of each layer constituting the convolutional neural network 2 may be a result value of an activation function, and the activation function may be, for example, a rectified linear unit (ReLU) function or a soft max function.

In the present invention, a two-dimensional image generated from a modulated signal may be input as an input image of the convolutional neural network 2, and a soft max function is used as the activation function of the last layer constituting the convolutional neural network 2 to be two-dimensional. A modulation scheme corresponding to an image may be classified, but the configuration of the convolutional neural network 2 may be variously modified and is not limited thereto.

3 is a diagram for describing an example of a convolution operation in a convolutional neural network.

In the example of FIG. 3 , it is assumed that the input feature map 310 has a size of 4x4 pixels, the kernel 320 has a size of 2x2 pixels, and it is assumed that the output feature map 330 has a size of 3x3 pixels, but is not limited thereto. may be implemented with feature maps and kernels 320 of various sizes. In addition, the values defined in the input feature map 310 , the kernel 320 , and the output feature map 330 are all exemplary values, and the present embodiments are not limited thereto.

The kernel 320 performs a convolution operation while sliding on the input feature map 310 in units of a window (or tile) having a size of 2x2 pixels. The convolution operation sums up all the values obtained by multiplying each pixel value of a certain window of the input feature map 310 and the weight of each element at a corresponding position in the kernel 320, and each pixel of the output feature map 330 An operation that returns a value. Specifically, the kernel 320 first performs a convolution operation with the first window 311 of the input feature map 310 . That is, each pixel value 1, 2, 0, and 1 of the first window 311 is multiplied by the weights 1, 2, 3, and 4 of each element of the kernel 320, respectively, and all values obtained as a result are In addition, 9 is calculated, and the pixel value of the first row and first column 331 of the output feature map 330 is determined to be 9. Here, the pixel value of the first row, first column 331 of the output feature map 330 corresponds to the first window 311 . In the same manner, a convolution operation between the second window 312 of the input feature map 310 and the kernel 320 is performed to determine the pixel value 19 of the first row, second column 332 of the output feature map 330 . . Finally, a convolution operation is performed between the ninth window 313, which is the last window of the input feature map 310, and the kernel 320, so that 9, the pixel value of the third row and third column 333 of the output feature map 330, is it is decided

That is, the convolution operation between one input feature map 310 and one kernel 320 is performed by repeatedly multiplying values of elements corresponding to each other in the input feature map 310 and the kernel 320 and summing the multiplication results. It can be processed by performing, and an output feature map 330 is generated as a result of the convolution operation.

Meanwhile, a pooling layer may be located to reduce the size of data of the output feature map 330 that is the result of the convolution operation. The pooling layer simplifies the output feature map 330 of the convolutional layer through a pooling operation. In the pooling operation, the kernel 320 may be applied, and a specific value may be extracted from values in the output feature map 330 in which the kernel 320 is located. An example of pooling to simplify information is max-pooling. In this case, the maximum value may be extracted from values in the output feature map 330 in which the kernel 320 is located during the pooling operation.

In the example of FIG. 3 , the kernel 320 having a size of 2x2 pixels moves by 1 movement interval (Stride) in the output feature map 330, which is the result of the convolution operation between the input feature map 310 and the kernel 320, and the kernel The maximum value may be taken from the values in the output feature map 330 where 320 is located. As a result, the output feature map 330 having a size of 3x3 pixels can be simplified to a size of 2x2 pixels.

4 is a diagram for explaining an example of a method of generating a two-dimensional image based on a cumulant value of a modulated signal.

Referring to FIG. 4, when SNR　 is -10dB in an additive white Gaussian noise (AWGN) channel environment, a two-dimensional image generated from a signal modulated by an 8-PSK (Phase Shift Keying) method is shown. have. However, in various SNR environments, a two-dimensional image of a signal modulated by a modulation method such as a phase shift keying (PSK) series, a pulse amplitude modulation (PAM) series, and a quadrature amplitude modulation (QAM) series is generated by the processor 110 can be The two-dimensional image is input to the convolutional neural network, and a modulation method of the modulation signal represented by the two-dimensional image may be classified based on the input.

The processor 110 may obtain a plurality of cumulant values from the modulated signal, and generate a two-dimensional image based on complex plane coordinate information of each of the obtained plurality of cumulant values. For example, the processor 110 obtains second-order cumulant and fourth-order cumulant values from the modulated signal, and two-dimensionally based on the complex plane coordinate information of the second-order and fourth-order cumulant values. You can create an image.

The second-order cumulant value and the fourth-order cumulant value of the modulated signal may be theoretically obtained by Equations 1 to 6 below.

C ₂₀ and C ₂₁ may represent a second cumulant value of the modulated signal, and C _40, C ₄₁ and C ₄₂ may represent a fourth cumulant value of the modulated signal. In Equations 1 to 5, y(n) represents a modulated signal. Since the modulated signal y(n) contains noise in the AWGN environment, it may be formed of a complex value. Therefore, in the second cumulant value, C ₂₀ may have a complex value, and C ₂₁ may have a real value. Similarly, in the quaternary cumulant value, C ₄₀ may be a real value, and C ₄₁ and C ₄₂ may be a complex value.

However, in the case of obtaining the second cumulant value and the fourth cumulant value of the modulated signal in the real environment, the second cumulant value and the fourth cumulant value can be estimated by Equations 7 to 11 below. have.

The values derived by Equations 7 to 11 are estimates of the second and fourth cumulant values of the modulated signal in an environment free from white Gaussian noise. In Equations 7 to 11, N denotes the number of symbols of the modulated signal used to derive one cumulant value. A symbol is a meaningful data bundle in a modulated signal, and the number of bits representing the symbol may vary according to a modulation method.

For example, N may be 100, and each of the second cumulant value and the fourth cumulant value may be derived by Equations 7 through 11 using 100 symbols in the modulated signal. If the processor 110 obtains 2000 symbols from the modulated signal _{, 200 each of C 20} , C ₂₀ , C _{40 ,} C ₄₁ and C ₄₂ may be obtained.

Meanwhile, in order to obtain the second and fourth cumulant values of the modulated signal in the AWGN channel environment, Equations 12 and 13 below should be considered.

은 AWGN 의 분산을 추정한 값이며,

는 AWGN 환경에서의 2차 큐뮬런트 값을 나타낸다. In Equation 12 above

is the estimated variance of AWGN,

denotes the second-order cumulant value in the AWGN environment.

는 AWGN 환경에서 변조 신호의 4차 큐뮬런트 값들을 나타낸다.

각각 수학식 9내지 11을 통해 추정한 백색 가우시안 잡음이 없는 환경에서의 변조 신호의 4차 큐뮬런트 값을 수학식 12를 통해 획득한

의 제곱으로 나누어 획득할 수 있다.Referring to Equation 13 above,

denotes the fourth-order cumulant values of the modulated signal in the AWGN environment.

The fourth-order cumulant value of the modulated signal in an environment without white Gaussian noise estimated through Equation 9 to

It can be obtained by dividing by the square of

Since the second and fourth cumulant values obtained from the modulated signal in the AWGN environment in the above manner are complex numbers or real numbers, the cumulant values C ₂₀ , C ₂₀ , C _{40 ,} C ₄₁ and C _{42 are} Each may have coordinate information on a complex plane including a real axis and an imaginary axis.

The processor 110 generates a second cumulant value and a second cumulant value on a two-dimensional plane including a real axis and an imaginary axis based on the coordinate information on the complex plane of each of the second cumulant value and the fourth cumulant value; The fourth-order cumulant values can be shown. In FIG. 4 , a horizontal axis R denotes a real axis, and a vertical axis I denotes an imaginary axis.

Referring to FIG. 4 , C ₂₀ , C ₂₀ , C _{40 ,} C ₄₁ and C ₄₂ are respectively blue, orange, yellow green, purple, and yellow, but the colors representing the respective cumulant values are not limited thereto. In addition, C ₂₀ , C ₂₀ , C _{40 ,} C _{41 ,} and C ₄₂ are respectively displayed on a two-dimensional plane including a real axis and an imaginary axis by 200, but specific cumulant values displayed on the two-dimensional plane The number of is not limited thereto.

In addition, the processor 110 sets the range of the real axis and the imaginary axis of the two-dimensional image to a certain range so that the distribution characteristics _{of the C 20} , C ₂₀ , C _{40 ,} C ₄₁ and C _{42 values displayed on the two-dimensional image are well revealed.} can be limited This is to improve the modulation scheme classification rate of signals in the convolutional neural network. For example, referring to FIG. 4 , the real axis of the 2D image may have a range of -2.5 to 2 and the imaginary axis may be limited by the processor 110 to have a range of -1.5 to 1.5. However, the range in which the real axis and the imaginary axis of the 2D image are limited is not limited thereto, and may be limited to various ranges so that the distribution characteristics of the cumulant values displayed on the 2D image can be clearly displayed.

In addition, the processor 110 may reduce the pixel size of the 2D image generated from the modulated signal to a predetermined size in order to improve the modulation method classification speed of the signal in the convolutional neural network. For example, the pixel size of the 2D image may be reduced to a size of 32x32 pixels horizontally and vertically, but is not limited thereto, and may be reduced to various pixel sizes.

On the other hand, in order to classify the modulation method of the modulated signal by inputting the two-dimensional image generated based on the cumulant value of the modulated signal received by the processor 110 to the convolutional neural network, a training process of the convolutional neural network is required. Do. Hereinafter, an example of training data used in a training process of a convolutional neural network will be described with reference to FIG. 5 .

5 is a diagram for explaining an example of training data for training a convolutional neural network.

The processor 110 may input training data so that the convolutional neural network can learn a very large number of neural network coefficients to exhibit optimized classification performance.

For example, the processor 110 puts input data and output data corresponding thereto into the convolutional neural network, and updates the connection strength of the connection lines constituting the convolutional neural network so that output data corresponding to the input data is output. By supervised learning Convolutional neural networks can be trained.

Specifically, the processor 110 may use an error backpropagation algorithm, and the error backpropagation algorithm decreases the error between the expected value and the actual output value in the output layer of the neural network in a direction (Gradient Descent Rule). The connection strength of a convolutional neural network can be adjusted. However, the algorithm applied by the processor 110 for training the convolutional neural network is not limited thereto, and various algorithms may be applied.

Referring to FIG. 5 , one piece of training data may be composed of a two-dimensional image and a label value generated from a signal modulated by a specific modulation method in a specific SNR environment. The two-dimensional image is generated based on the complex plane coordinate information of the cumulant value of the modulated signal as described above in FIG. 4, and the label value may be any number indicating the modulation method of the modulated signal.

In supervised learning, since input data and corresponding output data can be input as training data together, the 2D image corresponds to the input data constituting the training data, and the label value indicates the modulation method indicated by the 2D image as the input data. It may correspond to output data.

Specifically, in FIG. 5, when the SNR is any one of -10 dB, -5 dB, 0 dB, 5 dB, and 10 dB, modulation is performed by any one of 8-PSK, 4-PAM, BPSK, and 16-QAM. It shows the form of a plurality of training data including a two-dimensional image generated from the obtained signal and a label value indicating a corresponding modulation method. The label value may be set to 0, 1, 2, and 3 for 8-PSK, 4-PAM, BPSK, and 16-QAM, respectively.

5 shows 20 pieces of training data generated from each modulation scheme in each SNR. However, when actually training a convolutional neural network, for example, 300 or more pieces of training data generated from modulated signals modulated by a modulation method of 8-PSK at -10 dB may be input to the convolutional neural network.

However, FIG. 5 is only an example of training data, and the training data may include two-dimensional images of modulated signals according to various modulation schemes at various SNRs, and the label value is set to an arbitrary value so that it can be distinguished for each modulation scheme. can be set.

Meanwhile, after learning the convolutional neural network as described above, the processor 110 may input a two-dimensional image generated from an arbitrary modulated signal to the convolutional neural network in order to classify the modulation method of the arbitrary modulated signal.

Table 1 below shows the signal classification rate (%) of the convolutional neural network trained with training data according to the present invention.

SNR

modulation method -10dB -5dB 0dB 5dB 10dB 8-PSK 87 100 100 100 100 4-PAM 100 100 100 100 100 BPSK 100 100 100 100 100 16-QAM 72 100 100 100 100

Table 1 shows that when the processor 110 trains the convolutional neural network by inputting 300 training data according to each modulation scheme in each SNR shown in FIG. 5 as training data of the convolutional neural network, the trained convolutional neural network The signal classification rate (%) was measured by inputting 100 pieces of test data for each modulation method in each SNR. Referring to Table 1, it can be seen that 8-PSK, 4-PAM, BPSK, and 16-QAM can be accurately classified in an environment of SNR -5dB or higher. 6 is a flowchart illustrating an example of a process in which a signal modulation scheme is classified.

In operation 610, the processor 110 may obtain a plurality of cumulant values from the modulated signal. Since the modulated signal may have complex values in an additive white Gaussian noise environment, a plurality of cumulant values representing characteristics of the modulated signal may be real or complex values.

In operation 620, the processor 110 may generate a two-dimensional image based on the complex plane coordinate information of each of the plurality of cumulant values. Since the cumulant values may be real or complex values, each cumulant value may have coordinate information on a complex plane including a real axis and an imaginary axis. The processor 110 may generate a two-dimensional image by plotting a plurality of cumulant values on a two-dimensional plane including a real axis and an imaginary axis. The processor 110 sets the range of the real axis and the imaginary axis of the two-dimensional image so that the distribution characteristics of the plurality of cumulants displayed on the two-dimensional image are well revealed in order to improve the modulation method classification rate of the signal in the convolutional neural network. range can be limited. In addition, the processor 110 may reduce a pixel size of a 2D image generated from a modulated signal to a predetermined size in order to improve a signal modulation method classification speed in the convolutional neural network.

In step 630, the processor 110 may determine a modulation method of the modulated signal by inputting the 2D image to the neural network. When the 2D image representing the characteristics of the modulated signal is input to the neural network, the processor 110 may use convolutional neural networks, which are algorithms suitable for image classification. The processor 110 may train the convolutional neural network in a specific SNR environment using training data composed of a two-dimensional image and a label value generated from a signal modulated by a specific modulation method.

Meanwhile, the above-described embodiments of the present invention can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. In addition, the structure of the data used in the above-described embodiment of the present invention may be recorded in a computer-readable recording medium through various means. The computer-readable recording medium includes a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optically readable medium (eg, a CD-ROM, a DVD, etc.).

So far, with respect to the present invention, the preferred embodiments have been looked at. Those of ordinary skill in the art to which the present invention pertains will understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments are to be considered in an illustrative rather than a restrictive sense. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the scope equivalent thereto should be construed as being included in the present invention.

Claims (10)

obtaining a plurality of cumulant values from the modulated signal;
generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values; and
and determining a modulation method of the modulated signal by inputting the two-dimensional image into a neural network. The method of claim 1,
The 2D image is a signal classification method in which a real value and an imaginary value of each of the plurality of cumulant values are displayed on a complex plane. The method of claim 1,
The step of generating the two-dimensional image,
A signal classification method for limiting ranges of a real axis and an imaginary axis included in the two-dimensional image to predetermined ranges, respectively, based on distribution characteristics of the plurality of cumulant values displayed on the two-dimensional image. The method of claim 1,
Determining a modulation method of the modulated signal comprises:
A signal classification method for classifying the modulated signal into at least one modulation scheme of a phase shift keying (PSK) series, a pulse amplitude modulation (PAM) series, and a quadrature amplitude modulation (QAM) series. In the computer-readable recording medium in which a program for implementing a signal classification method is recorded,
The method is
obtaining a plurality of cumulant values from the modulated signal;
generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values; and
and determining a modulation method of the modulated signal by inputting the two-dimensional image into a neural network. In combination with hardware, a computer program stored in a computer-readable recording medium to execute a signal classification method,
The method is
obtaining a plurality of cumulant values from the modulated signal;
generating a two-dimensional image based on complex plane coordinate information of each of the plurality of cumulant values; and
and determining a modulation scheme of the modulated signal by inputting the two-dimensional image into a neural network. a memory that stores one or more instructions; and
By executing the one or more instructions, a plurality of cumulant values are obtained from a modulated signal, a two-dimensional image is generated based on complex plane coordinate information of each of the plurality of cumulant values, and the two-dimensional image is neural A signal classification apparatus comprising a processor input to a network and determining a modulation method of the modulated signal. 8. The method of claim 7,
The 2D image is a signal classification apparatus in which real and imaginary values of each of the plurality of cumulant values are displayed on a complex plane. 8. The method of claim 7,
The processor is
A signal classification apparatus for limiting a range of a real axis and an imaginary axis included in the two-dimensional image to a predetermined range, respectively, based on distribution characteristics of the plurality of cumulant values displayed on the two-dimensional image. 8. The method of claim 7,
The processor is
A signal classification apparatus for classifying the modulated signal into at least one modulation scheme of a phase shift keying (PSK) series, a pulse amplitude modulation (PAM) series, and a quadrature amplitude modulation (QAM) series.

Images