CN118296302A - Electromagnetic signal modulation classification method and device based on diffusion model - Google Patents
Electromagnetic signal modulation classification method and device based on diffusion model Download PDFInfo
- Publication number
- CN118296302A CN118296302A CN202410540600.0A CN202410540600A CN118296302A CN 118296302 A CN118296302 A CN 118296302A CN 202410540600 A CN202410540600 A CN 202410540600A CN 118296302 A CN118296302 A CN 118296302A
- Authority
- CN
- China
- Prior art keywords
- data set
- diffusion model
- denoising
- electromagnetic signal
- mask
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 93
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000009826 distribution Methods 0.000 claims abstract description 13
- 238000012545 processing Methods 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 8
- 230000009191 jumping Effects 0.000 claims description 3
- 230000007306 turnover Effects 0.000 abstract description 6
- 230000008569 process Effects 0.000 description 21
- 238000004891 communication Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 239000011159 matrix material Substances 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 3
- 229910003460 diamond Inorganic materials 0.000 description 3
- 239000010432 diamond Substances 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000000873 masking effect Effects 0.000 description 2
- 230000002441 reversible effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
Landscapes
- Complex Calculations (AREA)
Abstract
The invention discloses an electromagnetic signal modulation classification method and device based on a diffusion model, which comprises the steps of firstly acquiring an electromagnetic signal data set, and performing rotation and turnover operation to obtain a first data set; then copying each sample in the first data set into at least two samples to obtain a second data set, then randomly selecting a continuous sequence of each sample in the second data set as a mask segment, taking the rest part as a non-mask segment, and converting the mask segment into random pure noise by using standard Gaussian distribution to obtain a mask data set; and constructing a denoising diffusion model, training by adopting a second data set, and denoising the samples in the mask data set by using the trained denoising diffusion model to obtain an enhanced data set. And finally, training an automatic modulation recognition model by adopting the enhanced data set, and detecting the electromagnetic signal to be detected. The invention utilizes the denoising diffusion model to reconstruct data to strengthen the data set, and better solves the problems of insufficient data set and weak model generalization capability.
Description
Technical Field
The application belongs to the technical field of electromagnetic signal modulation classification, and particularly relates to an electromagnetic signal modulation classification method and device based on a diffusion model.
Background
With the rapid development of modern wireless communication technology, a large number of communication devices are widely used in military, commercial and civil applications, so that information acquisition in production and life of people becomes more accurate and rapid. Under such a background, the amount of communication data increases dramatically, spectrum resources become tense, and interference between spectrums causes degradation of communication quality and efficiency. Moreover, in early wireless communication systems, the modulation scheme used is relatively few and the identification of the modulation type is relatively simple due to technical limitations. However, with the progress of communication technology, especially the development of digital modulation technology, modulation schemes are becoming more and more diversified and complex, and identification of modulation types becomes difficult. In the fields of military and security, a modulation mode capable of rapidly identifying enemy or unknown signals is very important for electronic reconnaissance and signal information collection. Therefore, based on the improvement of the spectrum utilization efficiency, the method is flexibly suitable for various different modulation modes, the military information capacity is enhanced, and the like, and the accurate and rapid identification of the modulation modes becomes an important requirement in the field of electromagnetic signal processing. Therefore, an automatic modulation and classification system (AMC, automatic Modulation Classification) capable of automatically recognizing the modulation scheme of a received electromagnetic signal has become a key issue in the field of electromagnetic signal processing. The identification speed, compatibility and accuracy of the AMC system are improved, and the method is important to the improvement of the intelligence and efficiency of the wireless communication system.
More of the current AMC systems use deep learning models to conduct experiments, however, the performance of deep learning models depends largely on the data set quality. Lack of training data, or too much noise in the training data, often results in overfitting, thereby significantly affecting classification accuracy. In the field of deep learning, however, a high quality and sufficiently large data set is difficult to acquire in a real situation. Therefore, the electromagnetic signal data is subjected to data enhancement and the like, so that the data set is more sufficient, and the method has great practical significance for manufacturing the electromagnetic signal AMC data set based on deep learning. Data enhancement techniques also include a variety of conventional or modern methods, not necessarily all applicable to the electromagnetic signal arts. Under the background, the method for automatically modulating and classifying the electromagnetic signals by effectively combining the data enhancement technology and the deep learning method is researched, and has important practical significance and wide application prospect.
Disclosure of Invention
The application aims to provide an electromagnetic signal modulation classification method and device based on a diffusion model, so as to enhance effective data of electromagnetic signal data and improve accuracy of automatic modulation classification of electromagnetic signals.
In order to achieve the above purpose, the technical scheme of the application is as follows:
An electromagnetic signal modulation classification method based on a diffusion model comprises the following steps:
Acquiring an electromagnetic signal data set, and performing rotation and overturning operations to obtain a first data set;
Copying each sample in the first data set into at least two samples to obtain a second data set, randomly selecting a continuous sequence of each sample in the second data set as a mask segment, using the rest as a non-mask segment, and converting the mask segment into random pure noise by using standard Gaussian distribution to obtain a mask data set;
constructing a denoising diffusion model taking a transducer as a core, and training the denoising diffusion model by adopting a second data set;
Denoising the samples in the mask data set by using the trained denoising diffusion model to obtain an enhanced data set;
And training an automatic modulation recognition model by adopting the enhanced data set, and detecting the electromagnetic signal to be detected by adopting the trained automatic modulation recognition model to obtain a modulation type corresponding to the electromagnetic signal to be detected.
Further, the randomly selecting a continuous sequence as the mask segment includes:
when selecting the mask segments, the starting positions of the selected sequences are different, but the sequences are identical in length.
Further, the denoising diffusion model with the transducer as a core performs the following operations:
Taking a sample in the data set as input data, obtaining a first characteristic after convolution processing and ReLU activation, and inputting the first characteristic into a first residual error layer;
Embedding the diffusion model containing the time step t after full connection processing and ReLU, and adding the diffusion model with the first characteristic to obtain a second characteristic;
inputting the second characteristic into a double-layer transducer encoder to capture time and characteristic dependence, so as to obtain a third characteristic;
After the third feature is subjected to convolution processing, adding the third feature with other auxiliary information of the first feature subjected to expansion splicing and convolution processing to obtain a fourth feature;
the fourth feature is subjected to convolution treatment to obtain a fifth feature;
embedding a diffusion model containing a prompt word c to perform convolution expansion operation, and performing dot multiplication processing on the diffusion model and the fifth feature to obtain a sixth feature;
the sixth feature is used for controlling input information through a gating activation unit to obtain a seventh feature;
The seventh feature is input to the next residual layer as new input data after being convolved with the first feature and added;
all residual layers are connected through jumping, and then the output of the denoising diffusion model is obtained after convolution and ReLU processing.
Further, the prompt word is a single-heat code corresponding to the modulation type of the electromagnetic signal.
Further, the denoising the samples in the mask dataset by using the trained denoising diffusion model includes:
inputting the samples in the mask data set into a trained denoising diffusion model, and denoising the mask segments to obtain denoising sequence segments corresponding to the mask segments;
And splicing the denoising sequence segment and the non-mask segment to obtain a sample of the enhanced data set.
The application also provides an electromagnetic signal modulation classification device based on the diffusion model, which comprises a processor and a memory storing a plurality of computer instructions, wherein the computer instructions realize the steps of the electromagnetic signal modulation classification method based on the diffusion model when being executed by the processor.
The application provides an electromagnetic signal modulation classification method and device based on a diffusion model, which has the remarkable advantages that compared with the prior art:
The data enhancement effect is strong: the application firstly uses a rotation and turnover method to enhance the original data by N times, then integrally copies K parts and enhances the data by a denoising diffusion model, simultaneously introduces the capturing time and characteristic dependence of a double-layer transducer structure into the denoising diffusion model, expands the original data set into the previous N times and K times, has very high upper limit of the overall enhancement multiple, and can better generalize the model in the subsequent training process.
The model is convenient to manage: if multiple different modulation types of data are used in the training process, DDPM may confuse the different modulation types of signal characteristics, generating a chaotic signal. The application takes the original modulation category of the signal as a prompt word to be merged into the construction of the diffusion model, can easily divide the signal of each modulation category, does not need a plurality of diffusion models to train the signals of different modulation categories, reduces the number of models and provides convenience for model management.
Minimization of loss function optimization: the method combines the minimized loss function and introduces the original modulation category of the signal into the loss function as the prompt word, thus not only being capable of generating samples of specific category, but also improving the quality and diversity of the generated samples, and having better effect and higher classification accuracy in the subsequent training process.
Drawings
FIG. 1 is a flow chart of an electromagnetic signal modulation classification method based on a diffusion model.
FIG. 2 is a schematic view of a diffusion model structure according to the present application.
FIG. 3 is a graph of accuracy of experimental results on a small dataset.
FIG. 4 is a graph of accuracy of experimental results on a full dataset.
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
In one embodiment, as shown in fig. 1, an electromagnetic signal modulation classification method based on a diffusion model is provided, which includes:
Step S1, acquiring an electromagnetic signal data set, and performing rotation and overturning operations to obtain a first data set.
The electromagnetic signal data set of this embodiment may use some public data set, such as signal automatic modulation identification public data set, for example rml2016.10a, to collect a total of 220000 samples, each sample having 128 sampling points and two channels. At the same time, the data set contains an array of 220000 size for holding a class index of each sample, the class index is not more than 11, indicating that the total number of classes of the data set is 11.
The step performs rotation and turnover operation on each sample in the data set, and expands the rotation and turnover operation to N times (N < = 6) of the original rotation and turnover operation.
Specifically, the I/Q signal is realized by rotating the matrix dot-multiplying I/Q signal matrix by 90 degrees, 180 degrees and 270 degrees clockwise, and the rotated I/Q signal matrix is connected with the original signal matrix to realize triple enhancement of data, wherein the specific formula is as follows:
Wherein θ represents a rotation angle, I represents real data, Q represents imaginary data, I 'represents real data after rotating θ, and Q' represents imaginary data after rotating θ.
In addition, three-fold enhancement of data is also realized by a turnover mode, and a specific formula is expressed as follows:
Resulting in a first data set that is initially data enhanced.
And S2, copying each sample in the first data set into at least two samples to obtain a second data set, randomly selecting a continuous sequence of each sample in the second data set as a mask segment, taking the rest part as a non-mask segment, and converting the mask segment into random pure noise by using standard Gaussian distribution to obtain a mask data set.
The method comprises the steps of determining a coefficient K, copying K copies of each sample in a first data set, and obtaining a second data set.
And then randomly selecting a continuous sequence from each sample in the second data set as a mask segment, taking the rest part as a non-mask segment, and converting the mask segment into random pure noise by using standard Gaussian distribution, thereby obtaining a mask data set.
Preferably, when selecting the mask segments, the starting positions of the selected sequences are different, but the sequences are of uniform length.
And S3, constructing a denoising diffusion model taking a transducer as a core, and training the denoising diffusion model by adopting a second data set.
Fig. 2 is a denoising diffusion model with a transducer as a core, and the capturing time and characteristic dependence of the transducer encoder are introduced on the basis of DDPM (Denoising Diffusion Probabilistic Models) diffusion models in the embodiment. In this model, the data processing procedure is as follows:
Taking a sample in the data set as input data, obtaining a first characteristic after convolution processing and ReLU activation, and inputting the first characteristic into a first residual error layer;
Embedding (Diffusion Embedding) a diffusion model containing a time step t, performing full connection processing and ReLU, and adding the diffusion model with the first characteristic to obtain a second characteristic;
inputting the second characteristic into a double-layer transducer encoder to capture time and characteristic dependence, so as to obtain a third characteristic;
After the third feature is subjected to convolution processing, adding the third feature with other auxiliary information of the first feature subjected to expansion splicing and convolution processing to obtain a fourth feature;
the fourth feature is subjected to convolution treatment to obtain a fifth feature;
embedding a diffusion model containing a prompt word c to perform convolution expansion operation, and performing dot multiplication processing on the diffusion model and the fifth feature to obtain a sixth feature;
The sixth feature controls input information through a gating activation unit, so that long-term dependency of the sequence is better captured, and the principle of the gating activation unit can refer to a gate unit of the LSTM to obtain a seventh feature;
The seventh feature is input to the next residual layer as new input data after being convolved with the first feature and added;
all residual layers are connected through jumping, and then the output of the denoising diffusion model is obtained after convolution and ReLU processing.
Wherein the other auxiliary information vectors include time embedding and feature embedding, obtained by extracting the first feature. The diffusion model embedding comprising t and the diffusion model embedding comprising c are one type of diffusion model embedding, and comprise a time step t and a prompt word c respectively. Regarding diffusion model embedding, the diffusion model embedding is a parameter at the beginning, then the size of the model array is adapted through convolution and deformation, and the diffusion model embedding is convenient to be added with input data to be used as additional information, and is a mature technology in the technical field, and the description is omitted here.
The embodiment is different from the existing denoising diffusion model in that a denoising diffusion model taking a Transform as a core is introduced, and the purpose of the embodiment is to generate electromagnetic signal fragments from pure noise by utilizing a DDPM trainable noise estimation method E θ through the capturing performance of the Transform on time and characteristic dependence.
DDPM is implemented by the model through step-by-step denoising, so that the step-by-step denoising can be reversely and reversely pushed through the forward process of denoising in the denoising stage. During the DDPM training process, each sample in the second data set, assuming electromagnetic signal data { x 0, m }, is input, and electromagnetic signal sample x 0 is divided into mask segments according to mask mNon-masking segmentsMeanwhile, the single thermal code corresponding to the modulation type is also input as additional information to help the model identify the shape of the signal corresponding to the modulation type.
In each step T (i.e. the sum of the added noise is taken as the running T step), a Gaussian noise E is generated and added to the mask segmentGenerating a noise mask segment(Will be used as the input of the next stepThe noise e θ is estimated for the single-hot integration of the modulation class).
Then, willAndIntegrated intoAs an input of e θ. After input, estimate noise using e θ By minimizingAnd e losses, e θ can be trained to update network parameters. In the next step of the training process,And a diffusion model embedding including t+1, the diffusion model embedding including c being used as an input for the next step (the diffusion model embedding including c is the same as that described above), the diffusion model embedding including c being unchanged, and then regenerating a Gaussian noise E to be added to the noise mask segment of the previous stepThe step T is repeated until training is completed.
In order to overcome the defect that the existing denoising diffusion model is used for independently training one model for each modulation type, the model is convenient to manage, the training cost of the model is reduced, and the original modulation type of an electromagnetic signal is introduced into the diffusion model as a prompt word. Specifically:
the sample class index in the enhanced dataset is converted to a format of one-hot encoding, e.g., class index 5 is one-hot encoded into an array [0,0,0,0,0,1,0,0,0,0,0], where the length of the array represents the number of classes, the portion of array 1 represents that the sample belongs to class 5, and the portion of array 0 indicates that the sample does not belong to the remaining classes. And then, inputting the single-heat codes of the sample types as parameters, namely prompt word c, into a denoising diffusion model to participate in the training process of the model. The diffusion model containing the prompt word c is embedded into vectors after convolution expansion processing and before the vector is subjected to gating activation unit to carry out convolution point multiplication processing.
The forward process was modeled as follows:
Wherein q (x 1:T|x0) represents the forward direction process of the denoising diffusion model, The training noise adding process can be represented, that is, the data is added step by step from t=1 to T. q (x t|xt-1) represents the data distribution of x t at time T given the potential variable x t-1 at the last time, i.e. given the noise state of x t-1 at the last time, how x t should noise at time T, x 0 represents the original sample, x t represents the state at the noise T time, β t is a small normal number representing the noise level, T represents the total step size, I represents the identity matrix, and N () represents the normal distribution.
Because the original modulation category of the training set data is required to be used as a prompt word to participate in the reverse process of the model, the definition of the reverse denoising process is as follows:
Wherein, Representing the inverse denoising process, i.e. the one-hot encoding c with the non-masked segments of the original samples in a given modulation classBy noise masking segmentsObtaining the denoising mask segment of the last stepIs a conditional probability distribution of (a),Signal mask segment representing time TIs a function of the probability distribution of (1),Inverse process representing denoising diffusion model, i.e. one-hot encoding c of given modulation class and non-masking segment of original sampleIn the case of (a), denoising step by step to obtain a final denoising mask segmentIn the process of (a),A conditional denoising function is represented as such,Represents the non-masked segment of the original sample under category c, μ θ represents the conditional mean, σ θ represents the conditional standard deviation.
The method can not only generate samples of specific categories, but also improve the quality and diversity of the generated samples. In particular minimization using a minimization loss functionAnd E, thereby training the noise estimation functionSimulating the noise addition process to the mask segment so that it can be derived from pure noiseIs inferred from (a)
Wherein L (θ) represents a loss function, θ is a parameter to be adjusted of the denoising diffusion model, q (x 0) represents a probability distribution, x 0 represents an original sample, N (0,I) represents a standard normal distribution,Representing the sampling of the original samples x 0 from the probability distribution q (x 0), the sampling of e from the normal distribution N (0,I) with mean 0 and covariance as the identity matrix, the expected operation over time t, e representing random noise, e θ representing the noise estimation function,A signal mask segment representing the time t,A non-masked segment representing the original samples under category c, c representing the one-hot encoding of a given modulation category,Representing the square of the euclidean norm.
It should be noted that, the improvement of the present application is to add a transducer encoder and introduce a diffusion model embedding containing the hint word c, and the training of the diffusion model is a relatively mature technology in the art, which is not described herein.
And S4, denoising the samples in the mask data set by using the trained denoising diffusion model to obtain an enhanced data set.
The method comprises the steps of inputting samples in a mask data set into a trained denoising diffusion model, and denoising mask segments to obtain denoising sequence segments corresponding to the mask segments; and splicing the denoising sequence segment and the non-mask segment to obtain a sample of the enhanced data set.
Specifically by totaling T denoising steps from pure noiseTo infer a new mask segmentMasking the new segmentWith previous non-mask segmentsAnd forming a new complete electromagnetic signal, namely, the obtained enhanced data, and reconstructing a new electromagnetic signal data set.
In the denoising process DDPM inputs are samples in the mask dataset, each sample is composed of a mask segment of pure noiseNon-mask segmentsAnd the composition, while inputting the one-hot coding corresponding to the modulation class as additional information. Assume thatFrom electromagnetic signal samples by step-wise increase of gaussian noise in step TIn order to make the subsequent process clearer and more understandable, the process will beIs marked asIn the DDPM denoising process, the goal is to step from the total of T stepsReduction toIn each step T (t=t, T-1, …, 1) (i.e., the total denoising takes the step T), a gaussian noise is estimated using e θ output by the denoising diffusion model, and the estimated noise is usedTo be combined withDenoising is thatWill outputWith the diffusion model embedding containing T-1, the diffusion model embedding containing c (the diffusion model embedding containing c does not change) is input to DDPM again, the next step is carried out, and the denoising sequence segment of the final output signal of the T steps is repeated(Reduced)Unlike the original signal, which also enables data enhancement), the denoised sequence segmentNAND mask segmentAnd splicing to form a new electromagnetic signal, thereby obtaining an enhanced data set. In this process, the unmasked segmentWill participate in the calculation as a condition variable.
And S5, training an automatic modulation recognition model by adopting the enhanced data set, and detecting the electromagnetic signal to be detected by adopting the trained automatic modulation recognition model to obtain a modulation type corresponding to the electromagnetic signal to be detected.
In the step, the data of which the N is multiplied by K is adopted to train an automatic modulation recognition model, such as an LSTM model, and the trained automatic modulation recognition model can be used for detecting the electromagnetic signal to be detected to obtain the modulation type corresponding to the electromagnetic signal to be detected.
It should be noted that, training the automatic modulation recognition model and predicting by using the trained automatic modulation recognition model is a relatively mature technology in the field, and will not be described herein.
In another embodiment, the application also provides an electromagnetic signal modulation classification device based on a diffusion model, which comprises a processor and a memory storing a plurality of computer instructions, wherein the computer instructions realize the steps of the electromagnetic signal modulation classification method based on the diffusion model when being executed by the processor.
For specific limitations of the electromagnetic signal modulation classification device based on the diffusion model, reference may be made to the above limitation of the electromagnetic signal modulation classification method based on the diffusion model, which is not described herein. The electromagnetic signal modulation classification device based on the diffusion model can be fully or partially realized by software, hardware and a combination thereof. The method can be embedded in hardware or independent of a processor in the computer equipment, or can be stored in a memory in the computer equipment in a software mode, so that the processor can call and execute the corresponding operations of the modules.
The memory and the processor are electrically connected directly or indirectly to each other for data transmission or interaction. For example, the components may be electrically connected to each other by one or more communication buses or signal lines. The memory stores a computer program executable on a processor that implements the method of the embodiments of the present application by running the computer program stored in the memory.
The Memory may be, but is not limited to, random access Memory (Random Access Memory, RAM), read Only Memory (ROM), programmable Read Only Memory (Programmable Read-Only Memory, PROM), erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), etc. The memory is used for storing a program, and the processor executes the program after receiving an execution instruction.
The processor may be an integrated circuit chip having data processing capabilities. The processor may be a general-purpose processor including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), and the like. The methods, steps and logic blocks disclosed in the embodiments of the present invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
In order to verify the technical scheme of the application, an experiment is carried out by adopting a small data set of RML2016.10a and a full data set of RML 2016.10a. Fig. 3 shows a linear graph of accuracy on rml2016.10a small data set according to the technical scheme of the present application, where the abscissa corresponds to the signal-to-noise ratio of the signal, the ordinate corresponds to the accuracy of modulation classification recognition, the square curve corresponds to the accuracy of only rotation inversion enhancement, the diamond curve corresponds to the accuracy of combination of diffusion model denoising and rotation inversion enhancement, the circular curve corresponds to the accuracy of not performing any data enhancement, and under the conditions of high signal-to-noise ratio and insufficient data quantity, diffusion model denoising and rotation inversion enhancement (diamond curve) can greatly enhance the accuracy of modulation classification recognition of the model compared with the two.
Fig. 4 shows a linear graph of accuracy on the rml2016.10a full dataset, where the correspondence in the graph is consistent with that in fig. 3, and in case of sufficient data volume, the accuracy of model modulation classification recognition can be better enhanced by diffusion model denoising and rotation inversion enhancement (diamond curve) compared with the two.
The above examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.
Claims (6)
1. The electromagnetic signal modulation classification method based on the diffusion model is characterized by comprising the following steps of:
Acquiring an electromagnetic signal data set, and performing rotation and overturning operations to obtain a first data set;
Copying each sample in the first data set into at least two samples to obtain a second data set, randomly selecting a continuous sequence of each sample in the second data set as a mask segment, using the rest as a non-mask segment, and converting the mask segment into random pure noise by using standard Gaussian distribution to obtain a mask data set;
constructing a denoising diffusion model taking a transducer as a core, and training the denoising diffusion model by adopting a second data set;
Denoising the samples in the mask data set by using the trained denoising diffusion model to obtain an enhanced data set;
And training an automatic modulation recognition model by adopting the enhanced data set, and detecting the electromagnetic signal to be detected by adopting the trained automatic modulation recognition model to obtain a modulation type corresponding to the electromagnetic signal to be detected.
2. The electromagnetic signal modulation classification method based on the diffusion model according to claim 1, wherein randomly selecting a continuous sequence as the mask segment comprises:
when selecting the mask segments, the starting positions of the selected sequences are different, but the sequences are identical in length.
3. The electromagnetic signal modulation classification method based on the diffusion model according to claim 1, wherein the denoising diffusion model with a transducer as a core performs the following operations:
Taking a sample in the data set as input data, obtaining a first characteristic after convolution processing and ReLU activation, and inputting the first characteristic into a first residual error layer;
Embedding the diffusion model containing the time step t after full connection processing and ReLU, and adding the diffusion model with the first characteristic to obtain a second characteristic;
inputting the second characteristic into a double-layer transducer encoder to capture time and characteristic dependence, so as to obtain a third characteristic;
After the third feature is subjected to convolution processing, adding the third feature with other auxiliary information of the first feature subjected to expansion splicing and convolution processing to obtain a fourth feature;
the fourth feature is subjected to convolution treatment to obtain a fifth feature;
embedding a diffusion model containing a prompt word c to perform convolution expansion operation, and performing dot multiplication processing on the diffusion model and the fifth feature to obtain a sixth feature;
the sixth feature is used for controlling input information through a gating activation unit to obtain a seventh feature;
The seventh feature is input to the next residual layer as new input data after being convolved with the first feature and added;
all residual layers are connected through jumping, and then the output of the denoising diffusion model is obtained after convolution and ReLU processing.
4. The electromagnetic signal modulation classification method based on the diffusion model according to claim 3, wherein the prompt word is a single-heat code corresponding to a modulation class of the electromagnetic signal.
5. The electromagnetic signal modulation classification method based on the diffusion model according to claim 1, wherein denoising the samples in the mask dataset using the trained denoising diffusion model comprises:
inputting the samples in the mask data set into a trained denoising diffusion model, and denoising the mask segments to obtain denoising sequence segments corresponding to the mask segments;
And splicing the denoising sequence segment and the non-mask segment to obtain a sample of the enhanced data set.
6. An electromagnetic signal modulation classification device based on a diffusion model, comprising a processor and a memory storing a number of computer instructions, wherein the computer instructions when executed by the processor implement the steps of the method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410540600.0A CN118296302A (en) | 2024-04-30 | 2024-04-30 | Electromagnetic signal modulation classification method and device based on diffusion model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410540600.0A CN118296302A (en) | 2024-04-30 | 2024-04-30 | Electromagnetic signal modulation classification method and device based on diffusion model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118296302A true CN118296302A (en) | 2024-07-05 |
Family
ID=91680221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410540600.0A Pending CN118296302A (en) | 2024-04-30 | 2024-04-30 | Electromagnetic signal modulation classification method and device based on diffusion model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118296302A (en) |
-
2024
- 2024-04-30 CN CN202410540600.0A patent/CN118296302A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111797122B (en) | Method and device for predicting change trend of high-dimensional reappearance concept drift stream data | |
| WO2022105125A1 (en) | Image segmentation method and apparatus, computer device, and storage medium | |
| CN113902926A (en) | General image target detection method and device based on self-attention mechanism | |
| CN110458084B (en) | Face age estimation method based on inverted residual error network | |
| CN113095370B (en) | Image recognition method, device, electronic equipment and storage medium | |
| CN111460936A (en) | Remote sensing image building extraction method, system and electronic equipment based on U-Net network | |
| CN111553215A (en) | Personnel association method and device, and graph convolution network training method and device | |
| CN115147598B (en) | Target detection segmentation method and device, intelligent terminal and storage medium | |
| CN114638994B (en) | Multi-modal image classification system and method based on attention multi-interaction network | |
| CN110379491B (en) | Method, device, equipment and storage medium for identifying brain glioma | |
| CN111310743B (en) | Face recognition method and device, electronic equipment and readable storage medium | |
| CN111653275A (en) | Method and device for constructing voice recognition model based on LSTM-CTC tail convolution and voice recognition method | |
| CN116310425B (en) | Fine-grained image retrieval method, system, equipment and storage medium | |
| CN114358216B (en) | Quantum clustering method based on machine learning framework and related device | |
| CN109145738B (en) | Dynamic video segmentation method based on weighted non-convex regularization and iterative re-constrained low-rank representation | |
| CN111783688B (en) | Remote sensing image scene classification method based on convolutional neural network | |
| CN118296302A (en) | Electromagnetic signal modulation classification method and device based on diffusion model | |
| CN116704433A (en) | Self-supervision group behavior recognition method based on context-aware relationship predictive coding | |
| CN116958809A (en) | Remote sensing small sample target detection method for feature library migration | |
| CN116229580A (en) | Pedestrian re-identification method based on multi-granularity pyramid intersection network | |
| CN113743593B (en) | Neural network quantization method, system, storage medium and terminal | |
| CN115905702A (en) | Data recommendation method and system based on user demand analysis | |
| CN112633285B (en) | Domain adaptation method, domain adaptation device, electronic equipment and storage medium | |
| CN112766465A (en) | Training method of neural network for intelligent rotation performance detection | |
| CN115512693A (en) | Audio recognition method, acoustic model training method, device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |