Modulation identification method and device based on multi-mode residual attention network

Abstract

The invention relates to a modulation identification method and device based on a multi-mode residual attention network, and the method comprises the steps: obtaining modulation signal data to be identified; performing energy normalization and amplitude and phase data calculation processing on the modulation signal data to be identified to obtain amplitude data and phase data of the modulation signal to be identified; inputting the data into an intelligent identification network model to obtain a modulation mode of the modulation signal data to be identified; wherein the intelligent identification network model comprises a multi-modal feature fusion module, an ECA-based residual network module, a feature pyramid network module and a classification module. According to the method, the identification of more modulation types can be realized, the length of the signal is not limited, the problems of low accuracy and poor anti-noise performance of a traditional modulation identification method are solved, and meanwhile, the workload and the uncertainty of manually designing characteristic parameters are reduced.

Description

Technical Field

[0001] The present invention relates to an intelligent recognition technology for communication modulation modes in the field of wireless communication technology, and in particular to a modulation recognition method and device based on a multimodal residual attention network. Background Art

[0002] In modern information technology, the impact of electromagnetic space on social development, people's activities, and national security is also increasing. Communication modulation signals are the main carriers of information in electromagnetic space, providing a basis for wireless information interaction. In the field of military communications, communication modulation mode recognition technology is an important part of radio reconnaissance and electronic countermeasures, which can help monitor enemy intelligence and restrict communications, thereby improving the combat effectiveness of the army. In the field of civil communications, communication modulation mode recognition technology has been widely used in the field of radio management, which can help monitor unknown interference signals and illegal signals in the communication process, and improve the level of social modernization. It can be seen that communication modulation mode recognition has become a basic function in communication signal processing without or without prior knowledge.

[0003] With the development of modern information technology, there are more and more types of wireless communication devices, and thus the modulation styles are also diverse. From the earliest analog modulation to digital modulation, multiple modulations are mixed together, resulting in an increasingly complex electromagnetic environment, and higher requirements for modulation recognition technology. Whether in military or civilian use, the recognition of modulation types is of great significance, and corresponding solutions are also needed.

[0004] After a long period of development, traditional modulation recognition technology can be roughly divided into the likelihood ratio recognition method based on decision theory, the statistical pattern recognition method based on feature extraction, and the knowledge-based method. These methods have the following three main disadvantages: 1. They rely on expert experience for feature extraction and model construction, which is difficult to complete quickly and with high quality when there is a lot of data to be processed; 2. It is difficult to adapt to the complex and changeable electromagnetic environment. When the noise, fading, and interference change, new processing solutions need to be designed; 3. It is difficult to extract effective features when processing short burst signals, and the recognition effect is not ideal. Therefore, in the current complex and changeable electromagnetic environment, it is extremely important to develop efficient, adaptive, highly robust, and intelligent modulation recognition technology.

[0005] With the development of hardware and artificial intelligence algorithms, deep learning has shown high adaptability and satisfactory results in many fields. Deep learning provides a new way to solve modulation recognition. Compared with traditional methods, deep learning can automatically extract data features from massive data, and can also overcome the problems of low recognition efficiency, low classification accuracy, poor robustness and poor scalability of traditional methods. Summary of the invention

[0006] The present invention provides a modulation recognition method based on a multimodal residual attention network to solve the problems of low accuracy and poor noise resistance of traditional modulation recognition methods.

[0007] The technical solution adopted by the present invention to solve the technical problem is: to provide a modulation recognition method based on a multimodal residual attention network, comprising the following steps:

[0008] Acquire modulation signal data to be identified;

[0009] Performing energy normalization and calculating amplitude and phase data processing on the modulation signal data to be identified, so as to obtain amplitude data and phase data of the modulation signal to be identified;

[0010] The energy-normalized modulation signal data to be identified, the amplitude data and the phase data are input into an intelligent recognition network model to obtain a modulation mode of the modulation signal data to be identified; wherein the intelligent recognition network model includes:

[0011] A multimodal feature fusion module, used to extract features of the modulation signal data to be identified, the amplitude data and the phase data after energy normalization, and fuse the extracted features together by splicing to obtain fused features;

[0012] The residual network module based on ECA is used to extract deep features of the fusion features to obtain multi-level features;

[0013] A feature pyramid network module is used to perform global average pooling processing on each layer of the multi-level features, and splice the processed results in the channel dimension to obtain multi-level fusion features;

[0014] The classification module is used to perform classification according to the multi-level fusion features to obtain the modulation mode of the modulation signal data to be identified.

[0015] The multimodal feature fusion module includes:

[0016] A first feature extraction unit, composed of a two-dimensional convolution block, is used to extract the features of the modulation signal data to be identified after energy normalization to obtain a first feature;

[0017] A second feature extraction unit, composed of a one-dimensional convolution block, is used to extract the feature of the amplitude data to obtain a second feature;

[0018] A third feature extraction unit, composed of a one-dimensional convolution block, is used to extract the features of the phase data to obtain a third feature;

[0019] The splicing unit is used to splice the first feature, the second feature and the third feature to obtain a fusion feature.

[0020] The ECA-based residual network module includes multiple stacked ECA residual blocks, which extract features through three one-dimensional convolution blocks, then calculate the average value on each feature channel through global pooling, perform convolution processing on the feature channel, and calculate the attention weight using the Sigmoid function to achieve feature weighting.

[0021] The classification module comprises:

[0022] A fully connected layer, used to map the multi-level fusion features to the dimension of the number of categories;

[0023] The Softmax activation layer is used to convert the output of the fully connected layer into a probability distribution through a Softmax activation function.

[0024] The intelligent recognition network model is trained in the following way:

[0025] Collect modulated signal data with modulation type labels;

[0026] Performing energy normalization and calculating amplitude and phase data processing on all modulated signal data to obtain amplitude data and phase data of each modulated signal data;

[0027] Constructing a training data set and a test data set based on the energy-normalized modulation signal data, amplitude data, and phase data;

[0028] The intelligent recognition network model is trained using the training data set. During training, data of a fixed length is randomly intercepted from each data sample in the training data set for training. During testing, data samples in a test data set of any length are used for testing.

[0029] The loss function of the intelligent recognition network model is: FL(p k )=-α k (1-p k ) γ log(p k ), where FL(p k ) is the loss function of the intelligent recognition network model, α k is the category weight coefficient, p k is the probability of the correct category predicted by the intelligent recognition network model, and γ is the aggregation parameter.

[0030] The technical solution adopted by the present invention to solve the technical problem is: to provide a modulation recognition device based on a multimodal residual attention network, comprising:

[0031] An acquisition module, used for acquiring the modulation signal data to be identified;

[0032] A processing module, used for performing energy normalization and calculating amplitude and phase data processing on the modulation signal data to be identified, so as to obtain amplitude data and phase data of the modulation signal to be identified;

[0033] The identification module is used to input the energy-normalized modulation signal data to be identified, the amplitude data and the phase data into an intelligent identification network model to obtain the modulation mode of the modulation signal data to be identified; wherein the intelligent identification network model includes:

[0034] A multimodal feature fusion module, used to extract features of the modulation signal data to be identified, the amplitude data and the phase data after energy normalization, and fuse the extracted features together by splicing to obtain fused features;

[0035] The residual network module based on ECA is used to extract deep features of the fusion features to obtain multi-level features;

[0036] A feature pyramid network module is used to perform global average pooling processing on each layer of the multi-level features, and splice the processed results in the channel dimension to obtain multi-level fusion features;

[0037] The classification module is used to perform classification according to the multi-level fusion features to obtain the modulation mode of the modulation signal data to be identified.

[0038] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned modulation recognition method based on the multimodal residual attention network when executing the computer program.

[0039] The technical solution adopted by the present invention to solve its technical problem is: providing a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned modulation recognition method based on the multimodal residual attention network are implemented.

[0040] Beneficial Effects

[0041] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art: the intelligent recognition network model in the present invention enhances the feature extraction capability of the communication modulation signal through the multimodal feature fusion module, improves the recognition performance when the feature difference between different modulation types is small, and at the same time, the intelligent recognition network model uses multi-level features, so that it can better adapt to signals of different SPS. In addition, the intelligent recognition network model uses the ECA-based residual network module as the backbone, and the ECA mechanism can screen the features that are effective for modulation recognition, thereby improving the performance of the intelligent recognition network model. The intelligent recognition network model of the present invention can realize the recognition of more modulation types, without limiting the length of the signal, solving the problems of low accuracy and poor noise resistance of traditional modulation recognition methods, while reducing the workload and uncertainty of manually designed feature parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a modulation recognition method based on a multimodal residual attention network according to a first embodiment of the present invention;

[0043] Figure 2 It is a structural diagram of an intelligent identification network model in the first embodiment of the present invention;

[0044] Figure 3 is a schematic diagram of the structure of a one-dimensional convolution block and a two-dimensional convolution block in the first embodiment of the present invention;

[0045] Figure 4 is a schematic diagram of the structure of the ECA residual block in the first embodiment of the present invention;

[0046] Figure 5 is a curve diagram showing the variation of the accuracy of the training set and the test set with the number of iterations in the first embodiment of the present invention;

[0047] Figure 6 is a confusion matrix diagram of a simulation test set in the first embodiment of the present invention;

[0048] Figure 7 is a graph showing changes in the accuracy of the simulation test set with the signal-to-noise ratio in the first embodiment of the present invention;

[0049] Figure 8 is a graph showing changes in the accuracy of a simulation test set with respect to input signal length in the first embodiment of the present invention;

[0050] Fig. 9 It is a confusion matrix diagram of the actual data set in the first embodiment of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0052] The first embodiment of the present invention relates to a modulation recognition method based on a multimodal residual attention network, such as Figure 1 As shown, the following steps are included:

[0053] Step 1: Obtain the modulated signal data to be identified.

[0054] Step 2: Perform energy normalization and calculate amplitude and phase data processing on the modulation signal data to be identified to obtain amplitude data and phase data of the modulation signal to be identified.

[0055] Step 3, inputting the energy-normalized modulation signal data to be identified, the amplitude data and the phase data into an intelligent recognition network model to obtain a modulation mode of the modulation signal data to be identified.

[0056] like Figure 2 As shown, the intelligent recognition network model in this embodiment includes a multimodal feature fusion module, a residual network module based on ECA (Efficient Channel Attention Network), a feature pyramid network (Feature Pyramid Networks, FPN) module and a classification module. The output of the intelligent recognition network model can be expressed as:

[0057] Y=S(G mean (f(X)))

[0058] Among them, f(·) is the feature extracted by the multimodal feature fusion module and the ECA-based residual network module, G mean (·) represents the feature pyramid network module, and S(·) represents the classification module.

[0059] For f(·), the output can be written as:

[0060] f(X)=[F1,…,F M ]

[0061] Where M is the number of outputs. f(·) is based on CNN, which has the characteristics of parameter sharing and local connection. f(·) also has the characteristics of locality and translation invariance, that is, each output F m It is only related to local input, that is:

[0062] Fm =f([x n ,…,x n+k ])

[0063] Among them, k is the size of the feature receptive field. At the same time, in any time segment [x n ,…,x n+k ] are independent and identically distributed, and their statistical properties are the same, so there will be:

[0064] mean(F i )=mean(F j )

[0065] Where i and j represent any time. mean (·) It mainly performs global average pooling on the feature F sampled in time, which can be understood as the mean estimation of the feature F, so the model can adapt to data of different lengths. The longer the data, the faster G mean The more accurate the estimation of mean(F) by (f(X)) is, the better the modulation recognition performance is.

[0066] The multimodal feature fusion module in this embodiment is used to extract the features of the modulation signal data to be identified, the amplitude data and the phase data after energy normalization, and fuse the extracted features together by splicing to obtain fused features.

[0067] The multimodal feature fusion module in this embodiment includes: a first feature extraction unit, which is composed of a two-dimensional convolution block, and is used to extract the features of the modulated signal data to be identified after energy normalization to obtain a first feature; a second feature extraction unit, which is composed of a one-dimensional convolution block, and is used to extract the features of the amplitude data to obtain a second feature; a third feature extraction unit, which is composed of a one-dimensional convolution block, and is used to extract the features of the phase data to obtain a third feature; and a splicing unit, which is used to splice the first feature, the second feature and the third feature to obtain a fusion feature.

[0068] Figure 3 The specific structure diagram of the two-dimensional convolution block and the one-dimensional convolution block includes: a convolutional network layer, a batch normalization layer, and a ReLu activation layer. The multimodal feature fusion module in this embodiment enhances the feature extraction capability of the communication modulation signal and improves the recognition performance when the feature difference between different modulation types is small.

[0069] The ECA-based residual network module in this embodiment is used to extract the deep features of the fusion features to obtain multi-level features. The ECA-based residual network module includes four stacked ECA residual blocks. Figure 4The specific structure diagram of the ECA residual block is shown in Figure 2. In the ECA residual block, features are extracted through three one-dimensional convolution blocks, and then the average value on each feature channel is calculated through global pooling, and convolution processing is performed on the feature channel. The attention weight is calculated using the Sigmoid function to achieve feature weighting.

[0070] This embodiment uses multi-level features to better adapt to signals of different SPS. In addition, by using the ECA-based residual network module as the backbone, the ECA mechanism can filter features that are effective for modulation recognition, thereby improving the performance of the intelligent recognition network model.

[0071] The feature pyramid network module in this embodiment is used to perform global average pooling processing on each layer of the multi-level features, and splice the processed results in the channel dimension to obtain multi-level fusion features. The feature pyramid network module performs global average pooling on the four feature maps extracted from each ECA residual block, and then splices them in the channel dimension to achieve the effect of multi-level feature fusion. Among them, the global average pooling used can not only ensure that features of different levels can be fused together, but also is a necessary means to ensure that signals of different lengths can be adapted during reasoning.

[0072] The classification module in this embodiment is used to classify according to the multi-level fusion features to obtain the modulation mode of the modulation signal data to be identified. The classification module includes: a fully connected layer, which is used to map the multi-level fusion features to the dimension of the number of categories; a Softmax activation layer, which is used to convert the output of the fully connected layer into a probability distribution through a Softmax activation function. Assuming that the input feature of the classification module is x, the probability p of the kth category obtained after calculation by the fully connected layer and the Softmax activation layer is k It is calculated by the following formula:

[0073]

[0074] Among them, w is the weight parameter in the fully connected layer, b is the bias parameter in the fully connected layer, and M is the total number of eigenvalues.

[0075] The intelligent recognition network model training process in this implementation is as follows:

[0076] Data preparation:

[0077] In this embodiment, a large amount of labeled simulation data is generated based on MATLAB, with a total of nineteen modulation types, {"CW":0,"AM":1,"AM-DSB":2,"AM-SSB":3,"FM":4,"ASK":5,"BPSK":6,"QPSK":7,"OQPSK":8,"PI_4QPSK":9,"8PSK":10,"APSK16":11,"APSK32":12,"16QAM":13,"32QAM":14,"64QAM":15,"MSK":16,"2FSK":17,"4FSK":18}, including five analog modulations and fourteen digital modulations. In order to better adapt to real signals, the effects of different signal-to-noise ratios, SPS, frequency deviation, and modulation parameters are also considered during simulation. The specific parameter information contained in the data set is shown in Table 1.

[0078] Table 1 Dataset parameter information table

[0079]

[0080]

[0081] The length of each data sample is 16384. There are 2280000 simulation data in total, of which 80% (18240000 samples) are divided into training sets, and 20% (456000 samples) are divided into test sets.

[0082] Data processing:

[0083] Normalize all data and calculate amplitude and phase data. Assume that each data sample Energy normalized data S i ' is obtained by the following formula:

[0084]

[0085] Amplitude data A i ′ and phase data It is obtained from the following formula:

[0086]

[0087] A i ′=A i / A mean

[0088] Model training:

[0089] The specific parameter settings for model training are shown in Table 2.

[0090] Table 2 Specific parameters of model training

[0091]

[0092] Adam (Adaptive MomentEstimation) optimizer is an adaptive optimization algorithm that can adjust the learning rate based on historical gradient information. It is a variant of the gradient descent algorithm and is used to update the weights of the neural network. Warmup (learning rate warm-up) uses a smaller learning rate to warm up at the beginning of training, and then uses the preset learning rate for training after it is relatively stable. Cosine decay is a commonly used learning rate adjustment strategy. The learning rate will first increase slightly and then decrease. It is linear when it increases, and the cosine function is used to reduce the learning rate when it decreases.

[0093] Batch size is the amount of data input to the model each time of back propagation, Epoch is the number of rounds of iteration over all data. Input shape is the input size of the model during training.

[0094] The loss function of the intelligent recognition network model in this implementation is Focal loss, which is obtained by modifying the standard cross entropy loss. It introduces an adjustment factor to reduce the loss weight of easy-to-classify samples and increase the loss weight of difficult-to-classify samples. It is calculated as follows:

[0095] FL(p k )=-α k (1-p k ) γ log(p k )

[0096] Among them, p k is the probability of the correct category predicted by the intelligent recognition network model, α k is the category weight coefficient, which is used to deal with the problem of imbalanced number of categories, and γ is the aggregation parameter, which is used to adjust the weight of difficult and easy samples. After calculating the loss, back propagation updates the model parameters to train the model.

[0097] Figure 5 It is the curve of the accuracy of training set and validation set changing with the number of iterations, Figure 5 It can be seen that as the number of iterations increases, the accuracy of the training set still maintains an upward trend, but the accuracy of the validation set has stabilized, indicating that it has converged. The experimental results of the simulation data set are shown in Table 3.

[0098] Table 3 Experimental results of simulation data set

[0099] Training Accuracy Test accuracy 99.4％ 99.1％

[0100] The confusion matrix of the simulation test set is shown in Figure 6 The overall accuracy is high, and the confusion of each modulation type is also less. The curve of accuracy changing with signal-to-noise ratio is shown in Figure 7 It can be seen that as the signal-to-noise ratio increases, the accuracy rate also increases. When the signal-to-noise ratio is 3dB, the accuracy rate is about 92%, which means that it can adapt well to the situation with low signal-to-noise ratio. The model can use data of any length as input, change the input signal length, and obtain the accuracy curve that changes with the input signal length. Figure 8 As the length of the input signal data increases, the accuracy increases rapidly at first, and then increases slowly. When the input data length is greater than 1024, the accuracy is greater than 92%, and the model can better adapt to short burst signals.

[0101] Furthermore, in order to better illustrate the advantages of the present invention, actual data is used for verification. The parameters of the actual data set are shown in Table 4.

[0102] Table 4 Parameters of the actual data set

[0103]

[0104]

[0105] The parameter settings of the actual data and the simulation data will be different, in order to verify the generalization performance of the model. The main differences are as follows: 1) The sampling rate and symbol rate are different, in order to verify that the model can adapt to signals with different symbol rates in real situations; 2) The SPS is different. The SPS of the simulation data is selected as an integer between 5 and 15, while the actual data is selected as 7.5 and 12.5, which are outside the sampling values ​​of the simulation data; 3) The modulation parameters of each modulation mode (including FM, AM, ASK, FSK) are also selected outside the values ​​selected by the simulation data when collecting data, thereby proving the generalization of the modulation parameters. 4) Signal-to-noise ratio. The signal-to-noise ratio of the measured data is related to the real channel. 5) Frequency deviation. The frequency deviation is also related to the real channel, and it will be randomly adjusted during data collection.

[0106] The model trained with simulation data is tested on measured data with an input length of 4096. The results are shown in Table 5.

[0107] Table 5 Actual data test results

[0108] Actual data test accuracy 97.5％

[0109] The confusion matrix is ​​shown in Fig. 9 ,Compared to the simulation data, the results have decreased to a certain extent, but still have a high accuracy rate, proving that the ,method of this implementation has good adaptability.

[0110] It is not difficult to find that the intelligent recognition network model of the present invention can realize the recognition of more modulation types without limiting the length of the signal, which solves the problems of low accuracy and poor noise resistance of traditional modulation recognition methods, while reducing the workload and uncertainty of manually designed feature parameters.

[0111] A second embodiment of the present invention relates to a modulation recognition device based on a multimodal residual attention network, comprising:

[0112] An acquisition module, used for acquiring the modulation signal data to be identified;

[0113] A processing module, used for performing energy normalization and calculating amplitude and phase data processing on the modulation signal data to be identified, so as to obtain amplitude data and phase data of the modulation signal to be identified;

[0114] The identification module is used to input the energy-normalized modulation signal data to be identified, the amplitude data and the phase data into an intelligent identification network model to obtain the modulation mode of the modulation signal data to be identified; wherein the intelligent identification network model includes:

[0115] A multimodal feature fusion module, used to extract features of the modulation signal data to be identified, the amplitude data and the phase data after energy normalization, and fuse the extracted features together by splicing to obtain fused features;

[0116] The residual network module based on ECA is used to extract deep features of the fusion features to obtain multi-level features;

[0117] A feature pyramid network module is used to perform global average pooling processing on each layer of the multi-level features, and splice the processed results in the channel dimension to obtain multi-level fusion features;

[0118] The classification module is used to perform classification according to the multi-level fusion features to obtain the modulation mode of the modulation signal data to be identified.

[0119] The multimodal feature fusion module includes:

[0120] A first feature extraction unit, composed of a two-dimensional convolution block, is used to extract the features of the modulation signal data to be identified after energy normalization to obtain a first feature;

[0121] A second feature extraction unit, composed of a one-dimensional convolution block, is used to extract the feature of the amplitude data to obtain a second feature;

[0122] A third feature extraction unit, composed of a one-dimensional convolution block, is used to extract the features of the phase data to obtain a third feature;

[0123] The splicing unit is used to splice the first feature, the second feature and the third feature to obtain a fusion feature.

[0124] The ECA-based residual network module includes multiple stacked ECA residual blocks, which extract features through three one-dimensional convolution blocks, then calculate the average value on each feature channel through global pooling, perform convolution processing on the feature channel, and calculate the attention weight using the Sigmoid function to achieve feature weighting.

[0125] The classification module comprises:

[0126] A fully connected layer, used to map the multi-level fusion features to the dimension of the number of categories;

[0127] The Softmax activation layer is used to convert the output of the fully connected layer into a probability distribution through a Softmax activation function.

[0128] The intelligent recognition network model is trained in the following way:

[0129] Collect modulated signal data with modulation type labels;

[0130] Performing energy normalization and calculating amplitude and phase data processing on all modulated signal data to obtain amplitude data and phase data of each modulated signal data;

[0131] Constructing a training data set and a test data set based on the energy-normalized modulation signal data, amplitude data, and phase data;

[0132] The intelligent recognition network model is trained using the training data set. During training, data of a fixed length is randomly intercepted from each data sample in the training data set for training. During testing, data samples in a test data set of any length are used for testing.

[0133] The loss function of the intelligent recognition network model is: FL(p k )=-α k (1-p k ) γ log(p k ), where FL(p k ) is the loss function of the intelligent recognition network model, α k is the category weight coefficient, p k is the probability of the correct category predicted by the intelligent recognition network model, and γ is the aggregation parameter.

[0134] The third embodiment of the present invention relates to an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the modulation recognition method based on a multimodal residual attention network of the first embodiment when executing the computer program.

[0135] The fourth embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the modulation recognition method based on a multimodal residual attention network of the first embodiment.

[0136] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0137] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0138] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction method, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0139] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0140] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A modulation recognition method based on multimodal residual attention network, characterized in that: The following steps are involved: Acquire modulation signal data to be identified; Performing energy normalization and calculating amplitude and phase data processing on the modulation signal data to be identified, so as to obtain amplitude data and phase data of the modulation signal to be identified; The energy-normalized modulation signal data to be identified, the amplitude data and the phase data are input into an intelligent recognition network model to obtain a modulation mode of the modulation signal data to be identified; wherein the intelligent recognition network model includes: A multimodal feature fusion module, used to extract features of the modulation signal data to be identified, the amplitude data and the phase data after energy normalization, and fuse the extracted features together by splicing to obtain fused features; The residual network module based on ECA is used to extract deep features of the fusion features to obtain multi-level features; A feature pyramid network module is used to perform global average pooling processing on each layer of the multi-level features, and splice the processed results in the channel dimension to obtain multi-level fusion features; The classification module is used to perform classification according to the multi-level fusion features to obtain the modulation mode of the modulation signal data to be identified.

2. The modulation recognition method based on multimodal residual attention network according to claim 1 is characterized in that: The multimodal feature fusion module includes: A first feature extraction unit, composed of a two-dimensional convolution block, is used to extract the features of the modulation signal data to be identified after energy normalization to obtain a first feature; A second feature extraction unit, composed of a one-dimensional convolution block, is used to extract the feature of the amplitude data to obtain a second feature; A third feature extraction unit, composed of a one-dimensional convolution block, is used to extract the features of the phase data to obtain a third feature; The splicing unit is used to splice the first feature, the second feature and the third feature to obtain a fusion feature.

3. The modulation recognition method based on multimodal residual attention network according to claim 1, characterized in that: The ECA-based residual network module includes multiple stacked ECA residual blocks, which extract features through three one-dimensional convolution blocks, then calculate the average value on each feature channel through global pooling, perform convolution processing on the feature channel, and calculate the attention weight using the Sigmoid function to achieve feature weighting.

4. The modulation recognition method based on multimodal residual attention network according to claim 1, characterized in that: The classification module comprises: A fully connected layer, used to map the multi-level fusion features to the dimension of the number of categories; The Softmax activation layer is used to convert the output of the fully connected layer into a probability distribution through a Softmax activation function.

5. The modulation recognition method based on multimodal residual attention network according to claim 1, characterized in that: The intelligent recognition network model is trained in the following way: Collect modulated signal data with modulation type labels; Performing energy normalization and calculating amplitude and phase data processing on all modulated signal data to obtain amplitude data and phase data of each modulated signal data; Constructing a training data set and a test data set based on the energy-normalized modulation signal data, amplitude data, and phase data; The intelligent recognition network model is trained using the training data set. During training, data of a fixed length is randomly intercepted from each data sample in the training data set for training. During testing, data samples in a test data set of any length are used for testing.

6. The modulation recognition method based on multimodal residual attention network according to claim 1, characterized in that: The loss function of the intelligent recognition network model is: FL(p k )=-α k (1-p k ) γ log(p k ), where FL(p k ) is the loss function of the intelligent recognition network model, α k is the category weight coefficient, p k is the probability of the correct category predicted by the intelligent recognition network model, and γ is the aggregation parameter.

7. A modulation recognition device based on a multimodal residual attention network, characterized in that: include: An acquisition module, used for acquiring the modulation signal data to be identified; A processing module, used for performing energy normalization and calculating amplitude and phase data processing on the modulation signal data to be identified, so as to obtain amplitude data and phase data of the modulation signal to be identified; The identification module is used to input the energy-normalized modulation signal data to be identified, the amplitude data and the phase data into an intelligent identification network model to obtain the modulation mode of the modulation signal data to be identified; wherein the intelligent identification network model includes: A multimodal feature fusion module, used to extract features of the modulation signal data to be identified, the amplitude data and the phase data after energy normalization, and fuse the extracted features together by splicing to obtain fused features; The residual network module based on ECA is used to extract deep features of the fusion features to obtain multi-level features; A feature pyramid network module is used to perform global average pooling processing on each layer of the multi-level features, and splice the processed results in the channel dimension to obtain multi-level fusion features; The classification module is used to perform classification according to the multi-level fusion features to obtain the modulation mode of the modulation signal data to be identified.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the modulation recognition method based on the multimodal residual attention network as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the modulation recognition method based on a multimodal residual attention network as described in any one of claims 1 to 6 are implemented.

Images