Self-adaptive neural modulation recognition model construction method based on binary sampling operator and application

Abstract

The invention belongs to the technical field of signal automatic modulation classification, and discloses a binary sampling operator-based adaptive neural modulation recognition model construction method and application, and the method comprises the steps: employing a training set to preliminarily train a neural modulation recognition model for automatic modulation classification; initializing a binary sampling operator in each signal-to-noise ratio environment, sampling each piece of data in the verification set by adopting the corresponding binary sampling operator, inputting the sampled data into the neural modulation recognition model after preliminary training, and taking the minimum difference between an obtained model prediction result and a label as an optimization target; obtaining a binary sampling operator in each signal-to-noise ratio environment; and sampling each data in the training set through a binary sampling operator in a corresponding signal-to-noise ratio environment, and retraining the preliminarily trained neural modulation recognition model by using the sampled data to update model parameters so as to obtain a corresponding neural modulation recognition model. According to the invention, the automatic modulation classification capability of the NMR model in a complex noise environment can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal automatic modulation classification, and more specifically, relates to a method for constructing an adaptive neural modulation recognition model based on a binary sampling operator and its application. Background Art

[0002] Automatic Modulation Classification (AMC) is of great significance in the field of communication because the premise of demodulating a signal is to identify the modulation format of the signal. Therefore, it is crucial to develop an effective AMC recognition model. In recent years, inspired by the remarkable progress of deep learning technology in various fields, Neural Modulation Recognition (NMR), as an AMC model that integrates deep learning technology for modulation classification, has attracted increasing attention and achieved rapid development. O'Shea et al. developed NMR networks such as CLDNN, MCLDNN, and AMC-Net for automatic modulation classification.

[0003] The current NMR methods have insufficient recognition accuracy for actual wireless signals in complex noise environments. Feature selection methods, such as SubsamplerNet (SN), Holistic Subsampler (HS), etc., aim to eliminate redundant or noise features that may hinder modeling, and screen out samples in the dataset with low contribution rates to the classification results through feature selection. To a certain extent, the recognition accuracy of the classification model can be improved. This method has a high recognition accuracy in high signal-to-noise ratio environments, but in low signal-to-noise ratio environments, it is easy to eliminate low signal-to-noise ratio samples in the dataset, thereby reducing the recognition accuracy in low signal-to-noise ratio environments. How to improve the recognition accuracy of the NMR model in complex signal-to-noise ratio environments is a difficult problem! Summary of the Invention

[0004] In view of the above-mentioned defects or improvement requirements of the prior art, the present invention provides a method for constructing an adaptive neural modulation recognition model based on a binary sampling operator and its application, aiming to improve the recognition accuracy of the NMR model in complex signal-to-noise ratio environments.

[0005] To achieve the above object, the present invention provides a method for constructing an adaptive neural modulation recognition model based on a binary sampling operator, including:

[0006] Given a dataset containing different signal-to-noise ratio environments, the dataset includes a training set and a validation set; wherein, the samples in the dataset are multi-bit data obtained by sampling signals in a signal-to-noise ratio environment s, s ∈ {1, 2,..., S}, S is the total number of signal-to-noise ratio environments; the label is the modulation format of the signal.

[0007] Use the training set to preliminarily train a neural modulation recognition model for automatic modulation classification;

[0008] Initialize the binary sampling operator for each signal-to-noise ratio environment s, sample each data in the validation set using the binary sampling operator corresponding to the signal-to-noise ratio environment s, and take the minimum difference between the prediction result obtained by inputting the sampled data into the preliminarily trained neural modulation recognition model and the label as the optimization goal, and optimize the binary sampling operator for each signal-to-noise ratio environment s;

[0009] Sample each data in the training set using the optimized binary sampling operator corresponding to the signal-to-noise ratio environment s, and retrain the preliminarily trained neural modulation recognition model with the sampled data to update the model parameter θ and obtain the corresponding neural modulation recognition model.

[0010] Further, the sampling operation of the binary sampling operator is:

[0011]

[0012] where m s is the binary sampling operator for the signal-to-noise ratio environment s, m s ∈{0,1} L , L represents the length of the input sample x s , represents the data obtained by sampling x s using m s , and ⊙ represents dot product.

[0013] Further, use the particle swarm optimization algorithm to optimize the binary sampling operator for each signal-to-noise ratio environment s, and the optimization goal is:

[0014]

[0015] where, represents inputting into the preliminarily trained neural modulation recognition model, and the obtained model prediction result, θ0 represents the parameters of the preliminarily trained neural modulation recognition model; y is the label of x s ; J(·) represents the loss function of the model, which is used to characterize the difference between the model prediction result and the label.

[0016] Further, the optimization goal of the model parameter θ is:

[0017] Further, it also includes using the method of multiple iterative training to optimize the binary sampling operator for each signal-to-noise ratio environment s and the corresponding neural modulation recognition model parameter θ again, and the method of multiple iterative training includes:

[0018] Step 1: At the current iteration number t, initialize a set of binary sampling operators in S signal-to-noise ratio (SNR) environments. Denote it as an empty set;

[0019] Step 2: Sample the corresponding data in the validation set using the binary sampling operator in the SNR environment s. Input the sampled data into the preliminarily trained neural modulation recognition model. Take the minimum difference between the obtained model prediction result and the label as the optimization objective, and optimize the binary sampling operator at the current iteration number t and SNR environment s. And update m t ,

[0020] Step 3: Calculate, at the current iteration number t, the data obtained by sampling each data in the training set through Sampled data where ⊙ represents dot product;

[0021] Step 4: Let s = s + 1. Determine whether the current SNR environment s is less than the total number of SNR environments S. If yes, jump to Step 2; if not, jump to Step 5;

[0022] Step 5: Use Retrain the current neural modulation recognition model to obtain the corresponding model parameters θ t ;

[0023] Step 6: Let t = t + 1. Determine whether the current iteration number t is less than the preset iteration number T. If yes, jump to Step 1; if not, output the set of binary sampling operators m in S SNR environments at the iteration number T T , model parameters θ T , and obtain the binary sampling operator and neural modulation recognition model in each SNR environment s after iterative optimization.

[0024] Furthermore, the optimization algorithm is a particle swarm optimization algorithm.

[0025] The present invention also provides an adaptive neural modulation recognition method based on a binary sampling operator, including:

[0026] Input the signals with unknown modulation formats collected in different SNR environments into the optimized binary sampling operator for sampling, and input the sampled data into the trained neural modulation recognition model to obtain the modulation format of the signals;

[0027] where the optimized binary sampling operator and the trained neural modulation recognition model are constructed by using the adaptive neural modulation recognition model construction method described in any one of the above.

[0028] The present invention also provides an electronic device, including a computer-readable storage medium and a processor;

[0029] The computer-readable storage medium is used to store executable instructions;

[0030] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the adaptive neural modulation recognition model construction method described in any one of the above, or execute the adaptive neural modulation recognition method described above.

[0031] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the adaptive neural modulation recognition model construction method described in any one of the above, or implements the adaptive neural modulation recognition method described above.

[0032] The present invention also provides a computer program product, including a computer program, and when the computer program runs on a computer, it causes the computer to execute the adaptive neural modulation recognition model construction method described in any one of the above, or execute the adaptive neural modulation recognition method described above.

[0033] Generally speaking, through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0034] (1) In the method of the present invention, the neural modulation recognition model is initially trained by using a training set, and then the verification set and the initially trained neural modulation recognition model are used to optimize the binary sampling operator in each signal-to-noise ratio environment s, and then the training set is sampled by using the binary sampling operator; on the one hand, the data after sampling is used to retrain the initially trained neural modulation recognition model, adding new training samples to the neural modulation recognition model, so that the recognition accuracy of the model obtained after retraining and optimization is higher; on the other hand, the present invention combines the binary sampling operator with the NMR model (neural modulation recognition model) to perform binary sampling on each data in the sample. The process of binary sampling is equivalent to identifying the influence of each bit of data in each sample (a sample contains multiple bits of data after sampling a section of signal) on the classification result. The binary sampling operators in different signal-to-noise ratio environments are different, and the signals in complex signal-to-noise ratio environments are sampled adaptively. The higher the signal-to-noise ratio, the more useful data is retained on each sample, and the lower the signal-to-noise ratio, the more data is masked on each sample; that is to say, the present invention combines the binary sampling operator with the NMR model, which is equivalent to performing more fine-grained sampling on each sample. For high signal-to-noise ratio samples containing multiple bits of data, it plays a role in retaining useful information, and for low signal-to-noise ratio samples containing multiple bits of data, it plays a role in masking noise signals. Whether it is for high signal-to-noise ratio samples or low signal-to-noise ratio samples, the recognition accuracy of the model can be improved.

[0035] (2) Preferably, the particle swarm optimization algorithm is used to optimize the binary sampling operator in each signal-to-noise ratio environment. During the optimization process, the sampling ratio of the binary sampling operator is not fixed, which enhances the ability of NMR to adaptively and flexibly select data features, is more applicable to signal modulation format recognition in complex signal-to-noise ratio environments, and thus effectively improves the NMR automatic modulation classification ability.

[0036] (3) Preferably, the NMR model is retrained based on the data sampled by the currently optimized binary sampling operator, and the optimization objective is to minimize the difference between the prediction result of the NMR model and the label, and the model parameter θ is optimized, which can make the coordination between the NMR model and the binary sampling operator better.

[0037] (4) Preferably, the present invention generates a binary sampling operator through multiple iterative optimizations, which can avoid overfitting of the binary sampling operator, reduce underfitting of the binary sampling operator, and more accurately select the features of data in different signal-to-noise ratio environments. Through multi-step iterative training, the recognition accuracy of the NMR model is further enhanced.

[0038] In summary, the present invention adaptively subsamples the input signal by using a binary sampling operator to reduce the noise impact, so as to adapt to different signal-to-noise ratio environments and improve the NMR model automatic modulation classification ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of a method for constructing an adaptive neural modulation recognition model based on a binary sampling operator in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0041] Embodiment 1

[0042] As Figure 1 shown, an embodiment of the present invention provides a method for constructing an adaptive neural modulation recognition model based on a binary sampling operator, including:

[0043] S1. Divide a given data set containing different signal-to-noise ratio environments of signals into a training set, a validation set and a test set; a sample in the data set is data obtained by sampling a wireless signal in a signal-to-noise ratio environment s, s ∈ {1, 2,..., S}, S is the total number of signal-to-noise ratio environments, and the label is the modulation format corresponding to the wireless signal.

[0044] S2. Use the training set to preliminarily train a neural modulation recognition model (NMR model) for automatic modulation classification.

[0045] S3. Initialize the Binary Sampling Operator (BSO) for each signal-to-noise ratio environment s, sample each data in the validation set using the binary sampling operator corresponding to the signal-to-noise ratio environment s, and input the sampled data into the preliminarily trained neural modulation recognition model. Taking the highest recognition accuracy of this neural modulation recognition model (the smallest difference between the result predicted by the model and the label) as the optimization goal, use an optimization algorithm to obtain the binary sampling operator for each signal-to-noise ratio environment s. Among them, during the optimization process of the binary sampling operator, the model parameters θ0 of the preliminarily trained neural modulation recognition model remain fixed.

[0046] S4. Sample each data in the training set using the binary sampling operator corresponding to the signal-to-noise ratio environment s, input the sampled data into the preliminarily trained neural modulation recognition model, and retrain the preliminarily trained neural modulation recognition model to update the NMR model parameters θ of the preliminarily trained model to obtain the corresponding neural modulation recognition model.

[0047] In an alternative embodiment, in S1, use the typical wireless communication signal datasets RML2016.10a and RML2016.10b (in other embodiments, other wireless signal datasets can also be used), and divide the training set, validation set, and test set according to 3:1:1. The number of signal-to-noise ratio environments S = 20, and the input signal (input sample) step size L = 128, that is, each sample is divided into 128-bit data after being sampled.

[0048] In an alternative embodiment, in S2, use the currently most advanced NMR model, including any one of the network structures of AMC-Net, AWN, DualNet, MCLDNN, or ResNet.

[0049] In an alternative embodiment, in S3, use the Particle Swarm Optimization (PSO) algorithm to optimize the BSO. The number of particles and the number of iterations of the implemented PSO algorithm are 20 and 10 respectively. In the embodiments of the present invention, the binary sampling operator for each signal-to-noise ratio environment s is initialized as an empty set.

[0050] In an alternative embodiment, when using the Particle Swarm Optimization (PSO) algorithm to optimize the BSO, the sampling ratio of the BSO is not fixed.

[0051] In an alternative embodiment, for a given input signal x s , the corresponding SNR environment is s, and the binary sampling operation can be expressed as: where \(s\in\{1,2,\ldots,S\}\), \(s\) is the total number of SNR environments, and \(m\) s is the BSO (binary sampling operator) in the SNR environment \(s\), and \(m\) s \(\in\{0,1\}\) L , \(L\) represents the length of an input sample, denotes \(x\) s the data after being sampled by the binary sampling operator, and \(\odot\) represents dot product. In the embodiments of the present invention, the input signal \(x\) s is used as an input sample and is 2 * 128 bits.

[0052] In an alternative embodiment, the optimization of \(m\) s can be expressed as: where denotes the prediction result of the NMR model after inputting the input into the preliminarily trained NMR model, \(\theta_0\) represents the parameters of the preliminarily trained NMR model; \(y\) is the label of \(x\) s ; \(J(\cdot)\) represents the loss function of the NMR model. In the embodiments of the present invention, the loss function is the minimum of the feature loss between the result predicted by the NMR model and the label.

[0053] In an alternative embodiment, in S4, the preliminarily trained neural modulation recognition model is retrained to update the parameters \(\theta\) of the preliminarily trained NMR model. The optimization of the NMR model parameters \(\theta\) can be expressed as: Using the validation set, solve for \(\theta\) through the optimization algorithm.

[0054] In an alternative embodiment, it further includes further optimizing the binary sampling operator and the corresponding neural modulation recognition model \(\theta\) in each signal-to-noise ratio environment \(s\) by adopting a method of multiple iterative trainings, specifically including:

[0055] Step 1, at the current iteration number \(t\), initialize the set of binary sampling operators in \(S\) signal-to-noise ratio environments denotes the empty set;

[0056] Step 2, use the validation set \(D\) val , and through the PSO algorithm, solve for the binary sampling operator at the current iteration number \(t\) and the signal-to-noise ratio environment \(s\) and update \(m\) t ,

[0057] Step 3, calculate, at the current iteration number \(t\), the data train in the training set \(D\) after being binary sampled by

[0058] Step 4. Let \(s = s + 1\), and determine whether the current signal-to-noise ratio environment \(s\) is less than the total number of signal-to-noise ratio environments \(S\). If so, jump to Step 2; if not, jump to Step 5.

[0059] Step 5. Use the data after binary sampling to retrain the current neural modulation recognition model to obtain the NMR model parameter \(\theta\) t ; in the embodiment of the present invention, when \(t = 1\), the obtained NMR model parameter \(\theta\) t is the above \(\theta\).

[0060] Step 6. Let \(t = t + 1\), and determine whether the current iteration number \(t\) is less than the preset iteration number \(T\). If so, jump to Step 1; if not, output the binary sampling operator set \(m\) in \(S\) signal-to-noise ratio environments under the current iteration number \(t\) T , the NMR model parameter \(\theta\) T , \(\theta\) T is the finally trained NMR model parameter, and the binary sampling operator and the neural modulation recognition model under each signal-to-noise ratio environment \(s\) after optimization are obtained; in the embodiment of the present invention, the preset iteration number \(T = 6\).

[0061] The algorithm implementation of the method for constructing an adaptive neural modulation recognition model based on a binary sampling operator in the embodiment of the present invention is shown in Table 1:

[0062] Table 1 Algorithm implementation process

[0063]

[0064]

[0065] To further illustrate the ability of the neural modulation recognition method provided by the present invention in automatic modulation classification, the following is an analysis in combination with specific experimental examples:

[0066] The RML2016.10a and RML2016.10b generated by the commonly used GNU Radio are used in this experimental example. RML2016.10a contains a total of 220,000 examples, covering 11 modulation categories, with 20 different signal-to-noise ratio values ranging from -20 dB to +18 dB, at intervals of +2 dB. Each modulation category includes 1000 samples per signal-to-noise ratio, and each sample contains 128 sampling time steps. The RML2016.10b dataset is an extended version of RML2016.10a, containing 1.2 million samples and covering 10 modulation categories.

[0067] In this experimental example, six feature selection methods were selected as comparison baselines, namely: SubsamplerNet (SN), Holistic Subsampler (HS), Laplacian Score (LS), Fisher Score (FS), Robust Feature Selection (RFS), and Feature Quality Index (FQI). Denote the adaptive neural modulation recognition model based on the binary sampling operator in the embodiments of the present invention as SAFS. Tests were conducted on five types of networks: AMC-Net, AWN, DualNet, MCLDNN, and ResNet. Table 2 below shows the average accuracies of all methods on two datasets.

[0068] Table 2 Average accuracies of each method on two datasets

[0069]

[0070] The results show that SAFS in the embodiments of the present invention is superior to all feature selection baselines, proving the effectiveness and superiority of the neural modulation recognition method provided by the present invention.

[0071] For a given training set containing signal-to-noise ratio, the method of the present invention trains a neural modulation recognition model for automatic modulation classification; proposes a binary sampling operator (BSO) with a non-fixed sampling ratio, uses the validation set, and generates the BSO under each signal-to-noise ratio environment through an optimization algorithm; then retrains the NMR model using the training set and the current BSO to update the model parameters; uses the above iterative optimization algorithm to perform single-step or multi-step iteration and training on the BSO and NMR model parameters to generate the final BSO and the final NMR model under each SNR environment. By adaptively subsampling the input signal using the binary sampling operator to reduce the noise impact and adapt to different signal-to-noise ratio environments; by retraining the NMR to fine-tune the model parameters to enhance the coordination between the NMR model and the BSO, ultimately achieving the effect of improving the AMC accuracy.

[0072] The present invention applies the binary sampling operator to the NMR model, which can effectively cope with the severe noise of communication signals, enhance the NMR model, and improve the AMC accuracy.

[0073] Embodiment 2

[0074] An embodiment of the present invention provides an adaptive neural modulation recognition method based on a binary sampling operator, including:

[0075] Sampling the signals with unknown modulation methods and different signal-to-noise ratio environments through the trained binary sampling operator, and inputting the sampled data into the trained neural modulation recognition model to obtain the modulation method of the signal. The trained binary sampling operator and the neural modulation recognition model are constructed by the method for constructing an adaptive neural modulation recognition model based on a binary sampling operator in Embodiment 1.

[0076] For related technical solutions, refer to the description of the corresponding method in Embodiment 1, which will not be elaborated here.

[0077] Embodiment 3

[0078] An embodiment of the present invention provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method in Embodiment 1 or 2 above are implemented.

[0079] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application-Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory, various functions of the electronic device can be realized.

[0080] For related technical solutions, they are the same as above and will not be elaborated here.

[0081] Embodiment 4

[0082] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in Embodiment 1 or 2 above are implemented.

[0083] Specifically, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0084] The related technical solutions are the same as above and will not be elaborated here.

[0085] Embodiment 5

[0086] The embodiment of the present application provides a computer program product, including a computer program. When the computer program runs on a computer, it causes the computer to execute the steps of the method in the above Embodiment 1 or 2.

[0087] The related technical solutions are the same as above and will not be elaborated here.

[0088] Those skilled in the art can easily understand that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing an adaptive neural modulation recognition model based on a binary sampling operator, characterized in that: include: Given a data set containing different signal-to-noise ratio environments, the data set includes a training set and a validation set; wherein the samples in the data set are multi-bit data after the signal is sampled under the signal-to-noise ratio environment s, s∈{1,2,…,S}, S is the total number of signal-to-noise ratio environments; the label is the modulation format of the signal; Preliminarily training a neural modulation recognition model for automatic modulation classification using the training set; Initialize the binary sampling operator under each signal-to-noise ratio environment s, sample each data in the validation set using the binary sampling operator under the corresponding signal-to-noise ratio environment s, minimize the difference between the prediction result obtained by inputting the sampled data into the neural modulation recognition model after preliminary training and the label, and optimize the binary sampling operator under each signal-to-noise ratio environment s; The data in the training set are sampled by a binary sampling operator under an optimized corresponding signal-to-noise ratio environment s, and the neural modulation recognition model after the preliminary training is retrained with the sampled data to update the model parameters θ to obtain the corresponding neural modulation recognition model.

2. The method for constructing an adaptive neural modulation recognition model according to claim 1, characterized in that: The sampling operation of the binary sampling operator is: In the formula, m s is the binary sampling operator under the signal-to-noise ratio environment s, m s ∈{0,1} L , L represents the input sample x s Length, Represents x s By m s The sampled data, ⊙ represents the dot product.

3. The method for constructing an adaptive neural modulation recognition model according to claim 2, characterized in that: The particle swarm optimization algorithm is used to optimize the binary sampling operator under each signal-to-noise ratio environment s. The optimization goal is: in, Indicates that Input the neural modulation recognition model after preliminary training, and obtain the model prediction result, θ0 represents the neural modulation recognition model parameter after preliminary training; y is x s ; J(·) represents the loss function of the model, which is used to characterize the difference between the prediction result of the model and the label.

4. The method for constructing an adaptive neural modulation recognition model according to claim 3, characterized in that: The optimization objective of the model parameter θ is:

5. The method for constructing an adaptive neural modulation recognition model according to any one of claims 1 to 4, characterized in that: It also includes using a multiple iterative training method to optimize the binary sampling operator and the corresponding neural modulation recognition model parameter θ under each signal-to-noise ratio environment s, and the multiple iterative training method includes: Step 1: Initialize the binary sampling operator set under S signal-to-noise ratio environment at the current iteration number t represents the empty set; Step 2: Sample the corresponding data in the validation set using a binary sampling operator under a signal-to-noise ratio environment s, and input the sampled data into the neural modulation recognition model after preliminary training. The optimization goal is to minimize the difference between the model prediction result and the label, and optimize the current iteration number t and the binary sampling operator under the signal-to-noise ratio environment s. And update m t , Step 3: Calculate the number of iterations t for each data in the training set. The data after sampling Among them, ⊙ represents the dot product; Step 4, let s=s+1, determine whether the current signal-to-noise ratio environment s is less than the total number of signal-to-noise ratio environments S, if so, jump to step 2; if not, jump to step 5; Step 5, use Retrain the current neural modulation recognition model to obtain the corresponding model parameters θ t ; Step 6, let t = t + 1, determine whether the current iteration number t is less than the preset iteration number T, if so, jump to step 1; if not, output the binary sampling operator set m under the S signal-to-noise ratio environment under the iteration number T T , model parameters θ T , and obtain the binary sampling operator and neural modulation recognition model in each signal-to-noise ratio environment s after iterative optimization.

6. The method for constructing an adaptive neural modulation recognition model according to claim 5, characterized in that: The optimization algorithm is a particle swarm optimization algorithm.

7. An adaptive neural modulation recognition method based on a binary sampling operator, characterized in that: include: Inputting signals with unknown modulation formats collected under different signal-to-noise ratio environments into the optimized binary sampling operator for sampling, and inputting the sampled data into the trained neural modulation recognition model to obtain the modulation format of the signal; The optimized binary sampling operator and the trained neural modulation recognition model are constructed by the adaptive neural modulation recognition model construction method described in any one of claims 1 to 6.

8. An electronic device, characterized in that: comprising a computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the adaptive neural modulation recognition model construction method described in any one of claims 1-6, or to execute the adaptive neural modulation recognition method described in claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for constructing an adaptive neural modulation recognition model as described in any one of claims 1 to 6 is implemented, or the adaptive neural modulation recognition method as described in claim 7 is implemented.

10. A computer program product, characterized in that It includes a computer program, which, when running on a computer, enables the computer to execute the adaptive neural modulation recognition model construction method described in any one of claims 1 to 6, or execute the adaptive neural modulation recognition method described in claim 7.

Images