Lightweight Automatic Modulation Recognition Method and System for Signals Based on Deep Learning
By building a lightweight, dense convolutional long and short-time memory network structure, the problem of high complexity of deep learning automatic modulation recognition method on resource-constrained devices is solved, and efficient recognition on satellite platforms is achieved.
Patent Information
- Application Number
- CN202310291803.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-23
- Publication Date
- 2025-08-01
- Estimated Expiration
- 2043-03-23
AI Technical Summary
The existing deep learning automatic modulation recognition methods have problems with high model complexity and numerous parameters on resource-constrained devices, making it difficult to effectively deploy in satellite platforms and other environments.
A lightweight dense convolutional long and short-term memory network structure is adopted, and a sample data set is generated by simulating a real communication environment, and a lightweight convolutional network is used for training and optimization to build a signal modulation recognition model.
While ensuring recognition accuracy, the model size is reduced, suitable for resource-constrained equipment deployment, and improve application convenience on satellite platforms.
Smart Images
Figure CN116319210B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and particularly relates to a lightweight automatic modulation recognition method and system for signals based on deep learning. Background Art
[0002] Automatic modulation recognition (AMR) can obtain the modulation pattern of a signal and is a prerequisite for completing signal demodulation and then obtaining information. Automatic modulation recognition is a key technology in modern communication systems and has wide applications, such as spectrum interference detection, spectrum sensing, cognitive radio, etc. Researchers have carried out a large number of studies in the field of automatic modulation recognition and proposed various automatic modulation recognition methods. Currently, the automatic modulation recognition methods for signals can be roughly divided into the likelihood ratio-based automatic modulation recognition method (LB-AMR), the feature extraction-based automatic modulation recognition method (FB-AMR), and the deep learning-based automatic modulation recognition method (DL-AMR). LB-AMR is optimal in the sense of Bayesian estimation, but it highly depends on prior knowledge and parameter estimation, and the algorithm complexity is high. FB-AMR extracts various handcrafted features through an expert system, such as the instantaneous signal amplitude, phase, and frequency, constellation diagram, time-frequency distribution features, high-order cumulants, cyclic spectrum, etc., and then applies algorithms such as artificial neural structures, support vector machines, and decision trees to the classification process to improve the AMR performance. However, the extraction of the differentiated handcrafted features in FB-AMR requires extensive domain knowledge, and the quality of feature extraction directly determines the performance of FB-AMR. DL-AMR is somewhat superior to FB-AMR and LB-AMR.
[0003] However, the current deep learning algorithms are not designed for AMR. Some algorithm frameworks directly borrowed from fields such as image recognition and speech recognition have achieved certain effects in the AMR field, but they often have numerous algorithm parameters and large and complex models. On platforms with limited available resources or valuable resources, such as satellite platforms, due to the constraints of volume, mass, power consumption, and the influence of environmental factors such as space radiation, extreme temperature, and difficult maintenance, the computing power and storage space of on-board computers are very different from those of ground computers. Therefore, lightweight and low-complexity DL-AMR methods have begun to be studied. How to make DL-AMR reduce the model size or accelerate the calculation time while ensuring the recognition accuracy so that it can be deployed on resource-constrained devices has become a research direction in the field of signal modulation recognition. Summary of the Invention
[0004] Therefore, the present invention provides a lightweight automatic modulation recognition method and system for signals based on deep learning, which solves the problems of high complexity, numerous parameters, and large models in existing automatic modulation recognition, and meets the deployment requirements on resource-constrained devices through lightweight model design.
[0005] According to the design scheme provided by the present invention, a lightweight automatic modulation recognition method for signals based on deep learning is provided, including:
[0006] Obtain a sample data set containing several digital modulation methods and several analog modulation methods by modulating an information sequence in a simulated real communication environment;
[0007] Construct a lightweight dense convolutional long short-term memory network structure based on a lightweight convolutional network, and use the sample data set to train and optimize the lightweight dense convolutional long short-term memory network structure; and establish a signal modulation recognition model based on the trained and optimized lightweight dense convolutional long short-term memory network structure;
[0008] Input the target signal to be recognized into the signal modulation recognition model, and use the signal modulation recognition model to obtain the modulation method of the target signal to be recognized.
[0009] As the lightweight automatic modulation recognition method for signals based on deep learning of the present invention, further, obtaining a sample data set containing several digital modulation methods and several analog modulation methods by modulating an information sequence in a simulated real communication environment includes:
[0010] First, use a random function to randomly generate a 01-bit information sequence of radio;
[0011] Then, use the 01-bit information sequence to modulate the information sequence in a simulated real communication environment, and generate N digital modulation signals and M analog modulation signals, so as to use the N digital modulation signals and M analog modulation signals to form a sample data set, where M and N are integers greater than 1 respectively.
[0012] As the lightweight automatic modulation recognition method for signals based on deep learning of the present invention, further, in the process of modulating and sampling the information sequence by using the 01-bit information sequence to simulate the real communication environment, add real communication environment influence parameters to obtain the IQ sampling sequence of the simulated modulation signal, where the real communication environment influence parameters include but are not limited to: additive Gaussian noise, multipath fading, sampling rate offset, and center frequency offset.
[0013] As the lightweight automatic modulation recognition method for signals based on deep learning of the present invention, further, in the process of forming a sample data set by using N digital modulation signals and M analog modulation signals, for the information points of each modulation signal, use the information points with a preset threshold number as the sampling interval, and continuously collect K information points each time to form a signal sample; each modulation signal samples a preset number of samples, and combine the samples sampled by all modulation signals to construct a signal sample data set, where the preset number is set in units of ten thousand.
[0014] As the signal lightweight automatic modulation recognition method based on deep learning of the present invention, further, a lightweight dense convolutional long short-term memory network structure based on a lightweight convolutional network includes: a data input batch normalization processing unit for normalizing an input signal sequence, a dense link convolutional unit for performing convolution and splicing operations on the normalized data, and a feature extraction and classification unit for extracting temporal features of the spliced data using a long short-term memory network and classifying the temporal features through an activation function.
[0015] As the signal lightweight automatic modulation recognition method based on deep learning of the present invention, further, the feature extraction and classification unit is formed by connecting two long short-term memory network layers and a fully connected layer. The two long short-term memory network layers adopt different numbers of hidden units, and the activation function of the fully connected layer adopts the Softmax function.
[0016] As the signal lightweight automatic modulation recognition method based on deep learning of the present invention, further, when training and optimizing the lightweight dense convolutional long short-term memory network structure using a sample data set, an absolute cross-entropy loss function is set as the target loss function during the training process, an Adam optimizer is selected to optimize the network, and a maximum number of iterations or an early stopping mechanism is set as the iteration termination condition for training and optimization.
[0017] Further, based on the above method, the present invention also provides a signal lightweight automatic modulation recognition system based on deep learning, including: a sample construction module, a model training module, and a target recognition module, where
[0018] The sample construction module is used to obtain a sample data set including several digital modulation methods and several analog modulation methods by simulating a real communication environment to modulate an information sequence;
[0019] The model training module is used to build a lightweight dense convolutional long short-term memory network structure based on a lightweight convolutional network, train and optimize the lightweight dense convolutional long short-term memory network structure using the sample data set; and establish a signal modulation recognition model based on the trained and optimized lightweight dense convolutional long short-term memory network structure;
[0020] The target recognition module is used to input a target signal to be recognized into the signal modulation recognition model and obtain the modulation method of the target signal to be recognized using the signal modulation recognition model.
[0021] The beneficial effects of the present invention:
[0022] The present invention generates eight existing digital modulation signals and two analog modulation signals by modulating an information sequence in a simulated real communication environment to construct a sample data set for model training and optimization; constructs a lightweight dense convolutional long short-term memory network model for signal modulation recognition based on a lightweight convolutional neural network structure, and uses the sample data set for training and optimization, and uses the trained and optimized model to recognize target signals, solving the problems of high complexity, numerous parameters, and large model size in existing automatic modulation recognition. Through the lightweight design of the model, while ensuring the recognition rate, it can meet the deployment on resource-constrained devices such as satellite platforms and is convenient for practical scenario applications. Description of the Drawings
[0023] Figure 1 Schematic diagram of the signal lightweight automatic modulation recognition process based on deep learning in the embodiment;
[0024] Figure 2 Schematic diagram of the signal modulation recognition model construction process in the embodiment;
[0025] Figure 3 Schematic diagram of the verification loss change during the training process of the network model in the embodiment;
[0026] Figure 4 Schematic diagram of the signal modulation recognition rate result in the embodiment. Detailed Embodiment
[0027] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the drawings and technical solutions.
[0028] In an embodiment of the present invention, as shown in Figure 1 a signal lightweight automatic modulation recognition method based on deep learning is provided, including:
[0029] S101. Obtain a sample data set including several digital modulation methods and several analog modulation methods by modulating an information sequence in a simulated real communication environment.
[0030] Specifically, first, a 01-bit information sequence of radio can be randomly generated by using a random function; then, the 01-bit information sequence is used to simulate the modulation of the information sequence in a real communication environment, and N digital modulation signals and M analog modulation signals are generated, and the N digital modulation signals and M analog modulation signals are used to form a sample data set, where M and N are integers greater than 1 respectively.
[0031] Among them, a 01-bit information sequence is used to simulate a real communication environment to modulate an information sequence. During the modulation and sampling of the information sequence, real communication environment impact parameters are added to obtain an IQ sampling sequence of the simulated modulation signal. The real communication environment impact parameters include, but are not limited to: additive Gaussian noise, multipath fading, sampling rate offset, and center frequency offset.
[0032] A 01-bit sequence is randomly generated using a random function to ensure the randomness of the information content and eliminate the influence of the signal content on signal modulation recognition. The real communication environment is simulated to modulate the information sequence, and an in-phase and quadrature (IQ) sequence is sampled. During the modulation and sampling of the information sequence, the influences of additive Gaussian noise, multipath fading, sampling rate offset, and center frequency offset are added, which is similar to the actual environment. Some parameters are as follows: modulation rate 25 KBaud / s, sampling rate 200 KHz, standard deviation of each sample point during the sampling rate drift process 1 Hz, maximum sampling rate offset 50 Hz, maximum Doppler frequency used in fading simulation 1 Hz, time delay vector [1, 0.8, 0.3], signal-to-noise ratio at intervals of 2 dB between -20 dB and 18 dB, and the number of sine waves used in frequency-selective fading simulation 8. Finally, an IQ sampling sequence of the modulation signal can be obtained, that is, two values for each information point, and 8 information points for one symbol.
[0033] It should be noted that the specific values of the above partial parameter settings are not within the protection scope of the solution in this case, and the specific values of the parameters can be adjusted according to the actual experimental environment. The above M / N can respectively represent 8 digital modulation signals and 2 analog modulation signals, namely 8 digital modulation methods of 8PSK, BPSK, CPFSK, GFSK, PAM4, 16QAM, 64QAM, and 2 analog modulation methods of AM-DSB and WBFM.
[0034] As a preferred embodiment, further, N digital modulation signals and M analog modulation signals are used to form a sample data set. For the information points of each modulation signal, with a preset threshold number of information points as the sampling interval, K information points are continuously collected each time to form a signal sample; each modulation signal samples a preset number of samples, and the samples sampled by all modulation signals are combined to construct a signal sample data set, where the preset number is set in units of ten thousand.
[0035] The information points of each modulation signal use 8 information points as the sampling interval, and 128 information points are continuously collected each time to form a signal sample. Each modulation signal samples 120,000 samples, and all the signals are combined into a signal sample set. From the generated sample set, 60% of each type of modulation signal is extracted to form a training sample set, 20% is extracted from the remaining 40% to form a verification sample set, and the last 20% of the entire sample set is used as a test sample set.
[0036] Each modulated signal can collect 120,000 samples, and all the signals are combined to form a signal sample set. For each modulated signal, signals are collected at intervals of 2 dB from -20 dB to 18 dB. 6,000 samples are collected at each signal-to-noise ratio, and a total of 120,000 samples are collected for each modulated signal. All the collected signals are combined to form a signal sample set, with a total of 1.2 million samples.
[0037] From the generated sample set, 60% of each type of modulated signal can be randomly selected to form a training sample set, 20% is randomly selected from the remaining 40% to form a validation sample set, and the last remaining 20% of the entire sample set is used as a test sample set. For the 6,000 signal samples at each signal-to-noise ratio under each modulation pattern, first, 60% is randomly selected and added to the training set. Then, 20% is randomly selected from the remaining 40% and added to the validation sample set. Finally, the remaining 20% is added to the test sample set. See Figure 2 As shown, the training sample set, validation sample set, and test sample set are respectively used to train, validate, and test the network model, so as to obtain the final model structure for modulating and identifying the target signal through training optimization.
[0038] It should be noted that the above specific values can be set according to empirical values and / or actual usage environments. For those skilled in the art, other specific values can also be selected.
[0039] S102. Build a lightweight dense convolutional long short-term memory network structure based on a lightweight convolutional network, and use the sample data set to train and optimize the lightweight dense convolutional long short-term memory network structure; and establish a signal modulation recognition model based on the trained and optimized lightweight dense convolutional long short-term memory network structure.
[0040] Specifically, the lightweight dense convolutional long short-term memory network structure constructed based on a lightweight convolutional network includes: a data input batch normalization processing unit for normalizing the input signal sequence, a dense link convolutional unit for performing convolution and splicing operations on the normalized data, and a feature extraction and classification unit for using a long short-term memory network to extract the temporal features of the spliced data and classifying the temporal features through an activation function.
[0041] In the data input batch normalization unit, the IQ sequence is input into the network in a one-dimensional time series format, and the batch normalization (BN) layer is placed at the head of the network to normalize the input signal, which can solve the difficulty of model training, prevent the vanishing gradient and internal data distribution offset phenomena, and enhance the robustness of the network to various received signals; the BN layer first calculates the mean and variance of the training data, and then transforms the data to conform to the standard normal distribution with a mean of 0 and a standard deviation of 1.
[0042] The densely connected convolutional unit can be designed to include 5 convolutional units (Convolutional Unit 1, Convolutional Unit 2, Convolutional Unit 3, Convolutional Unit 4, Convolutional Unit 5) and 5 concatenation layers (Concatenate1, Concatenate2, Concatenate3, Concatenate4, Concatenate5). Among them, the convolutional unit is composed of a convolutional layer, a BN layer, and a ReLU activation function in sequence to enhance the non-linear mapping ability. The role of the concatenation layer is to concatenate the inputs according to the dimensions.
[0043] The connection structure of the densely connected convolutional unit can be designed as: BN layer → Convolutional Unit 1 → Concatenate1, BN layer → Concatenate1; Concatenate1 → Convolutional Unit 2 → Concatenate2, Concatenate1 → Concatenate2;
[0044] Concatenate2 → Convolutional Unit 3 → Concatenate3, Concatenate2 → Concatenate3; Concatenate3 → Convolutional Unit 4 → Concatenate4, Concatenate3 → Concatenate4; Concatenate4 → Convolutional Unit 5 →
[0045] Concatenate5; BN layer → Concatenate5
[0046] The feature extraction and classification unit can be sequentially connected by two long short-term memory layers (LSTM1, LSTM2) and a fully connected output layer. Among them, the activation function of the fully connected output layer is Softmax.
[0047] First, the parameters of the 6 convolutional layers and the 2 long short-term memory layers in the lightweight densely connected convolutional long short-term memory network can be set, as well as the target loss function, optimizer, initial learning rate, batch training size, number of training epochs, and early stopping mechanism; the network model with initialized parameters is trained and optimized using the sample data set.
[0048] The specific parameters of the 6 convolutional layers and the 2 long short-term memory layers can be set as follows:
[0049] The number of convolutional kernels of the one-dimensional convolutional layer in Convolutional Unit 1 is 4, and the convolutional kernel size is 3.
[0050] The number of convolutional kernels of the one-dimensional convolutional layer in Convolutional Unit 2 is 4, and the convolutional kernel size is 3.
[0051] The number of convolutional kernels of the one-dimensional convolutional layer in Convolutional Unit 3 is 4, and the convolutional kernel size is 5.
[0052] The number of convolutional kernels in the one-dimensional convolutional layer in convolutional unit 4 is 4, and the size of the convolutional kernel is 7.
[0053] The number of convolutional kernels in the one-dimensional convolutional layer in convolutional unit 5 is 8, and the size of the convolutional kernel is 1.
[0054] The number of hidden units in LSTM1 is 32.
[0055] The number of hidden units in LSTM2 is 10.
[0056] The number of convolutional kernels in the fully connected layer is 11, corresponding to the number of output categories.
[0057] When training the lightweight dense convolutional long short-term memory network with the training set, the Adam optimizer is selected to optimize the network. The initial learning rate is set to 0.001, 512 samples are trained in each batch, and the maximum number of training epochs for the entire training samples is 200. The model is verified with the validation set after each training epoch. Taking the validation loss as a reference, when the validation loss does not decrease after 50 epochs, the training of the model is stopped.
[0058] The test sample set is input into the trained lightweight dense convolutional long short-term memory network to obtain the recognition results, which are compared with the true categories, and the recognition accuracy rate is statistically calculated. At the same time, the inference time of the lightweight dense convolutional long short-term memory network is recorded to obtain the sample inference speed, so as to optimize the network model parameters.
[0059] In the model training optimization, the arrangement order of all samples in the training samples can be shuffled. The training samples and validation samples are input into the lightweight dense convolutional long short-term memory network model to train the lightweight dense convolutional long short-term memory network. When the maximum number of network training epochs is reached or the early stopping mechanism condition is met, the training process of the neural network is completed, and the trained lightweight dense convolutional long short-term memory network model is obtained.
[0060] S103. Input the target signal to be recognized into the signal modulation recognition model, and use the signal modulation recognition model to obtain the modulation method of the target signal to be recognized.
[0061] Use the signal modulation recognition model optimized by training to complete the efficient and high-precision recognition of various modulation signals, improve the overall performance of signal modulation recognition, and facilitate the deployment of the solution on the application device.
[0062] Furthermore, based on the above method, the present invention also provides a signal lightweight automatic modulation recognition system based on deep learning, including: a sample construction module, a model training module, and a target recognition module, where,
[0063] A sample construction module for obtaining a sample data set containing several digital modulation methods and several analog modulation methods by modulating an information sequence in a simulated real communication environment;
[0064] A model training module for constructing a lightweight dense convolutional long short-term memory network structure based on a lightweight convolutional network, training and optimizing the lightweight dense convolutional long short-term memory network structure using the sample data set; and establishing a signal modulation recognition model based on the trained and optimized lightweight dense convolutional long short-term memory network structure;
[0065] A target recognition module for inputting a target signal to be recognized into the signal modulation recognition model and using the signal modulation recognition model to obtain the modulation method of the target signal to be recognized.
[0066] To verify the effectiveness of the solution in this case, the following further explanation is made in combination with experimental data:
[0067] The simulation experiment is implemented on the NVIDIA Quadro RTX 6000 and Keras 2.6.0 TensorFlow-GPU 2.4.0 platforms to complete the simulation experiments of the present invention and the generation of modulation signals and the lightweight dense convolutional long short-term memory network. Using Figure 2 The steps shown are used to complete the experiment, and the change trend of the validation loss during the training process of the lightweight dense convolutional long short-term memory network is obtained to verify the signal recognition rate and the model inference speed of the solution in this case.
[0068] Figure 3 Shows the change of the validation loss during the training process of the lightweight dense convolutional long short-term memory network; from Figure 3 It can be seen that the validation loss decreases, converges and stabilizes, indicating that the training effect of the simulation experiment gradually improves with the increase of the number of training times. Figure 4 This is the result graph of the recognition rate of the simulation experiment of the solution in this case. It can be seen from the graph that the recognition rate gradually increases and stabilizes as the signal-to-noise ratio increases, and can reach up to 93.7%; at the same time, the model inference speed is recorded as 0.024 milliseconds for each sample inference time.
[0069] From the above simulation experiments, it can be shown that for the automatic recognition of modulation signals, the solution in this case can efficiently and accurately complete the automatic modulation recognition task, and the feasibility of the solution is further verified.
[0070] Unless otherwise specifically stated, the relative steps, numerical expressions and values of the components and steps set forth in these embodiments do not limit the scope of the present invention.
[0071] In the present specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts among the embodiments can be referred to each other. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.
[0072] The units and method steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation is not considered to exceed the scope of the present invention.
[0073] Those of ordinary skill in the art can understand that all or part of the steps in the above methods can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disc, etc. Optionally, all or part of the steps of the above embodiments can also be implemented using one or more integrated circuits. Correspondingly, the various modules / units in the above embodiments can be implemented in the form of hardware or in the form of software function modules. The present invention is not limited to any specific form of the combination of hardware and software.
[0074] Finally, it should be noted that the above-described embodiments are only specific embodiments of the present invention, used to illustrate the technical solutions of the present invention, rather than to limit it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments, or can easily think of changes, or make equivalent replacements for some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.
Claims
1. A lightweight automatic modulation recognition method for signals based on deep learning, characterized in that Including: Modulating an information sequence by simulating a real communication environment to obtain a sample data set including several digital modulation methods and several analog modulation methods; Constructing a lightweight dense convolutional long short-term memory network structure based on a lightweight convolutional network, training and optimizing the lightweight dense convolutional long short-term memory network structure using the sample data set; and establishing a signal modulation recognition model based on the trained and optimized lightweight dense convolutional long short-term memory network structure; Inputting a target signal to be recognized into the signal modulation recognition model, and using the signal modulation recognition model to obtain the modulation method of the target signal to be recognized; Among them, the lightweight dense convolutional long short-term memory network structure constructed based on the lightweight convolutional network includes: a data input batch normalization processing unit for normalizing the input signal sequence, a dense link convolutional unit for performing convolution and splicing operations on the normalized data, and a feature extraction and classification unit for extracting the temporal features of the spliced data using a long short-term memory network and classifying the temporal features through an activation function; The feature extraction and classification unit is composed of two long short-term memory network layers and a fully connected layer. The two long short-term memory network layers adopt different numbers of hidden units, and the activation function of the fully connected layer adopts the Softmax function; In the data input batch normalization unit, the IQ sequence is input into the network in a one-dimensional time series format, and the batch normalization layer is placed at the head of the network to standardize the input signal, which can solve the difficulty of model training, prevent the disappearance of gradients and the offset of internal data distribution, and enhance the robustness of the network to various received signals; the batch normalization layer first calculates the average value and variance of the training data, and then transforms the data to conform to the standard normal distribution with an average value of 0 and a standard deviation of 1; The dense link convolutional unit includes 5 convolutional units and 5 splicing layers. Among them, the convolutional unit is composed of a convolutional layer, a BN layer, and a ReLU activation function in sequence to enhance the non-linear mapping ability; the role of the splicing layer is to splice the input according to the dimension.
2. The signal lightweight automatic modulation recognition method based on deep learning according to claim 1, characterized in that Modulating an information sequence by simulating a real communication environment to obtain a sample data set including several digital modulation methods and several analog modulation methods, including: First, randomly generate a 01-bit information sequence of a radio using a random function; Then, use the 01-bit information sequence to simulate a real communication environment to modulate the information sequence, and generate N digital modulation signals and M analog modulation signals, so as to use the N digital modulation signals and M analog modulation signals to form a sample data set, where M and N are integers greater than 1 respectively.
3. The signal lightweight automatic modulation recognition method based on deep learning according to claim 2, wherein When using the 01-bit information sequence to simulate a real communication environment to modulate the information sequence, real communication environment influence parameters are added during the modulation and sampling process of the information sequence to obtain the IQ sampling sequence of the analog modulation signal. Among them, the real communication environment influence parameters include but are not limited to: additive Gaussian noise, multipath fading, sampling rate offset, and center frequency offset.
4. The signal lightweight automatic modulation recognition method based on deep learning according to claim 2, wherein In the sample data set composed of N digital modulation signals and M analog modulation signals, for each information point of each modulation signal, with the information points of the preset threshold number as the sampling interval, K information points are continuously collected each time to form a signal sample; each modulation signal samples a preset number of samples, and the samples sampled by all modulation signals are combined to construct a signal sample data set, where the preset number is set in ten thousands.
5. The signal lightweight automatic modulation recognition method based on deep learning according to claim 1, characterized in that When using the sample data set to train and optimize the lightweight dense convolutional long short-term memory network structure, set the absolute cross-entropy loss function as the target loss function during the training process, select the Adam optimizer to optimize the network, and set the maximum number of iterations or the early stopping mechanism as the iteration termination condition for training optimization.
6. A lightweight automatic modulation recognition system for signals based on deep learning, characterized in that, It includes: a sample construction module, a model training module, and a target recognition module, where The sample construction module is used to obtain a sample data set containing several digital modulation methods and several analog modulation methods by modulating the information sequence in a simulated real communication environment; The model training module is used to construct a lightweight dense convolutional long short-term memory network structure based on the lightweight convolutional network, and use the sample data set to train and optimize the lightweight dense convolutional long short-term memory network structure; and establish a signal modulation recognition model based on the trained and optimized lightweight dense convolutional long short-term memory network structure; The target recognition module is used to input the target signal to be recognized into the signal modulation recognition model, and use the signal modulation recognition model to obtain the modulation method of the target signal to be recognized; Among them, the lightweight dense convolutional long short-term memory network structure constructed based on the lightweight convolutional network includes: a data input batch normalization processing unit for normalizing the input signal sequence, a dense link convolutional unit for performing convolution and splicing operations on the normalized data, and a feature extraction and classification unit for extracting the temporal features of the spliced data using the long short-term memory network and classifying the temporal features through an activation function; The feature extraction and classification unit is composed of two long short-term memory network layers and a fully connected layer. The two long short-term memory network layers use different numbers of hidden units, and the activation function of the fully connected layer uses the Softmax function; In the data input batch normalization unit, the IQ sequence is input into the network in a one-dimensional time series format, and the batch normalization layer is placed at the head of the network to standardize the input signal, which can solve the difficulty of model training, prevent the phenomenon of gradient disappearance and internal data distribution deviation, and enhance the robustness of the network to various received signals; the batch normalization layer first calculates the average value and variance of the training data, and then transforms the data to conform to the standard normal distribution with an average value of 0 and a standard deviation of 1; The dense link convolutional unit includes 5 convolutional units and 5 splicing layers. The convolutional unit is composed of a convolutional layer, a BN layer, and a ReLU activation function in sequence to enhance the non-linear mapping ability; the role of the splicing layer is to splice the input according to the dimension.
7. An electronic device, characterized in that, It includes a memory and a processor, and the processor and the memory communicate with each other through a bus; the memory stores program instructions executable by the processor, and the processor can execute the method steps described in any one of claims 1 to 5 by invoking the program instructions.
8. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 5 are implemented.