Low signal-to-noise ratio conditional modulation identification method based on wavelet transform and channel attention

Abstract

The invention discloses a low signal-to-noise ratio conditional modulation identification method based on wavelet transform and channel attention. The method comprises the following steps: acquiring a received signal; and inputting the received signal into an identification model of a pre-trained neural network, and outputting a classification identification result and classification accuracy. In the recognition model based on signal reconstruction and the neural network, a wavelet threshold estimation module and a wavelet reconstruction module form a signal reconstruction path, a multi-scale feature extraction module and a prediction classification module form a recognition path, and the two paths are mutually enhanced, so that the combination of the advantages of a digital signal processing technology and the neural network is realized; a wavelet threshold estimation module is introduced to predict a denoising threshold through a neural network, and network parameters of the wavelet threshold estimation module are updated according to back propagation; a signal reconstruction path is combined with a recognition path, the interpretability of the network is enhanced, and the accuracy of modulation recognition in a low signal-to-noise ratio scene is remarkably improved by using multi-scale modulation characteristics.

Description

Technical Field

[0001] The invention belongs to the technical field of radio signal modulation recognition, and in particular relates to a modulation recognition method under low signal-to-noise ratio conditions based on wavelet transform and channel attention. Background Art

[0002] Automatic Modulation Recognition (AMR) is a technology that identifies the modulation mode of radio signals, especially non-cooperative signals. It provides basic modulation information of radio signals and is widely used in modern mobile communications, cognitive radio, spectrum sensing, signal detection and other civil fields. It plays an important practical role in spectrum scarcity issues such as large-scale Internet of Things, self-driving cars, and drones. In addition, in non-cooperative communication scenarios, automatic modulation recognition is crucial for reverse communication, acting as a key intermediate signal processing step between signal detection and signal demodulation, which helps subsequent signal demodulation and intelligence analysis. In most communication scenarios, the prior information related to the user is not known, especially in non-cooperative communication scenarios. Therefore, seeking a method that does not rely on the prior knowledge of the target signal and has good modulation recognition performance at low signal-to-noise ratio is an urgent problem to be solved in signal recognition research.

[0003] Traditional signal recognition technology is mainly divided into two methods: likelihood-based (LB) and feature-based (FB). The LB method can obtain the best performance by considering the probability density function of all unknown quantities, and is often used as the theoretical upper limit in the performance comparison of pattern recognition. However, this method has high computational complexity and cannot be applied to the situation of unknown signals and transceiver mismatch. The FB method refers to extracting specific features from the received signal and then using a reasonable classifier to classify different modulated signals. The FB-based method has relatively low computational complexity and is very reliable in alleviating the negative impact of multipath fading, but the selection of features depends largely on prior information. In recent years, deep learning (DL) has been widely studied in the field of AMR. AMR based on deep learning captures the relationship between radio signals and modulation categories by training a neural network, and uses the network to automatically extract signal features that are easy to classify. Existing DL-AMR methods include convolutional neural network (CNN) models, recurrent neural network (RNN) models, and hybrid neural network models. CNN models have excellent capabilities in processing data with spatial characteristics. Using their spatial feature extraction capabilities to detect signal modulation schemes, existing CNN-based AMR models can be divided into two categories: CNN models with raw IQ inputs and CNN models with preprocessed inputs, such as combining traditional FB-AMR methods with CNN models. In addition to spatial correlation characteristics, radio signals also have strong temporal correlation characteristics, which can be learned by RNN models. Therefore, many scholars use RNN to achieve modulation recognition. Benefiting from the respective advantages of RNN and CNN, hybrid neural network models have emerged, that is, hybrid models that combine RNN and CNN to further improve the performance of AMR.

[0004] Many techniques in digital signal processing (DSP) rely heavily on prior information and expert experience, which is impractical in non-cooperative scenarios. In addition, the existing DL-AMR methods will see a sharp drop in accuracy when the signal-to-noise ratio is lower than 0dB. Therefore, it is necessary to improve and optimize the deep neural network to improve the performance of DL-AMR under low signal-to-noise conditions, such as using a network with stronger feature extraction capabilities. However, the performance of AMR under low signal-to-noise ratio conditions can hardly be improved by optimizing the neural network alone, because when the signal-to-noise ratio is lower than 0dB, the signal is almost completely submerged by noise, making it difficult for the network to extract signal features for classification. In addition, existing methods treat DSP and AMR as independent tasks, and cannot allow the two to benefit from an end-to-end joint framework. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention. The technical problem to be solved by the present invention is achieved by the following technical solutions:

[0006] The present invention provides a low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention, the method comprising:

[0007] Get the received signal;

[0008] Inputting the received signal into a pre-trained recognition model based on signal reconstruction and neural network, and outputting a modulation category corresponding to the received signal;

[0009] Among them, the recognition model includes: a wavelet threshold estimation module, which is used to perform threshold estimation on the I / Q signal component of the received signal based on a neural network to obtain a predicted wavelet denoising threshold; a wavelet reconstruction module, which is used to perform denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal; a multi-scale feature extraction module, which is used to perform feature extraction on the I-path signal component, Q-path signal component and IQ parallel signal component of the reconstructed signal, respectively, to obtain I-path features, Q-path features and IQ-path features; based on a channel attention network, different channel features of the I-path features, Q-path features and IQ-path features are fused to obtain fused features, and features with time series relationships of the fused features are obtained through a long short-term memory network; a prediction classification module, which is used to obtain the modulation category of the received signal based on the features with time series relationships of the fused features.

[0010] In one embodiment of the present invention, the expression of the received signal is as follows:

[0011]

[0012] Among them, S r (n) represents the received signal, Indicates that from M modulation modes The modulated signal generated in the transmission process, h(n) represents the impulse response of the wireless transmission channel, * represents the convolution operation, w(n) represents the signal noise, Δf represents the additional carrier frequency offset during the transmission process, Represents the random phase shift caused by phase jitter during transmission, n=0,1,…,N-1,N, N represents the total length of the received signal.

[0013] In one embodiment of the present invention, the expression of the I / Q signal component of the received signal is as follows:

[0014] I signal component I(n) = real(Sr (n));

[0015] Q-path signal component Q(n)=imag(S r (n));

[0016] Among them, real(S r (n)) represents the received signal S r The real part of (n), imag(S r (n)) represents the received signal S r The imaginary part of (n).

[0017] In one embodiment of the present invention, the training process of the recognition model based on signal reconstruction and neural network includes: after several iterative trainings, after reaching a preset maximum number of iterations, a trained recognition model based on signal reconstruction and neural network is obtained; wherein, the process of one iterative training includes:

[0018] Taking all received signals as a data set; dividing the data set into a training set, a test set and a validation set;

[0019] For each received signal in the training set, the wavelet threshold estimation module performs threshold estimation on the I / Q signal component of the received signal to obtain a predicted wavelet denoising threshold;

[0020] The wavelet reconstruction module performs denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal;

[0021] Using the multi-scale feature extraction module, feature extraction is performed on the I-path signal component, Q-path signal component and IQ parallel signal component of the reconstructed signal respectively to obtain I-path features, Q-path features and IQ-path features; based on the channel attention network, different channel features of the I-path features, Q-path features and IQ-path features are fused to obtain fused features; based on the long short-term memory network, the fused features are processed to obtain features with a temporal relationship of the fused features;

[0022] Utilizing the prediction classification module, based on the features with a time series relationship, a predicted label of the received signal is obtained; the loss function value of the current iterative training is obtained according to the true label and the predicted label; the parameters of the recognition model are updated using the loss function value to obtain an updated recognition model; the updated recognition model is used for the next iterative training.

[0023] In one embodiment of the present invention, the wavelet threshold estimation module performs threshold estimation on the I / Q signal component of the received signal to obtain a predicted wavelet denoising threshold, including:

[0024] The I / Q signal components of the received signal are processed in sequence using the first fully connected layer and the tanh activation layer in the wavelet threshold estimation module to obtain the feature F t ;

[0025] The feature F t The expression is as follows:

[0026] Ft=tanh(fc1(reshape(Sr(n))));

[0027] Among them, tanh() represents the tanh activation function, S r (n) represents the received signal, fc1 represents the first fully connected layer processing, and reshape() represents tensor operation;

[0028] After passing the reshaped signal through a second fully connected layer, the predicted wavelet denoising threshold is obtained;

[0029] The expression of the predicted wavelet denoising threshold is as follows:

[0030]

[0031] Among them, sigmoid() represents the activation function, and fc2() represents the second fully connected layer processing;

[0032] The network parameters of the wavelet threshold estimation module are updated through back propagation.

[0033] In one embodiment of the present invention, the wavelet reconstruction module performs denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal, including:

[0034] The wavelet reconstruction module decomposes the received signal using a first formula based on a preset wavelet basis and a decomposition layer number to obtain a low-frequency component and a high-frequency component;

[0035] The first formula is as follows:

[0036]

[0037] Among them, S r (n) represents the received signal, Indicates S r (n) The low-frequency part at scale k obtained after scale space projection, Indicates S r (n) The high-frequency part at scale l obtained after projection into wavelet space;

[0038] De-noising the high-frequency component based on the predicted wavelet de-noising threshold to obtain a de-noised high-frequency component;

[0039] A reconstructed signal is obtained based on the low-frequency component and the denoised high-frequency component.

[0040] In one embodiment of the present invention, the long short-term memory network includes: 2 serially connected sub-long short-term memory networks with 128 units.

[0041] In one embodiment of the present invention, using the prediction classification module to obtain a prediction label of the received signal based on the features with a temporal relationship includes: the prediction classification module processes the features with a temporal relationship to obtain a prediction probability;

[0042] Processing the predicted probability using a prediction formula to obtain a predicted label of the received signal;

[0043] The prediction formula is as follows:

[0044]

[0045] Among them, p(yx) represents the predicted probability, y represents the true label, x represents the received signal, and argmax() represents the argmax function.

[0046] In one embodiment of the present invention, the loss function is expressed as follows:

[0047]

[0048] Among them, Loss represents the loss value, K represents the total number of modulation categories, and y i represents the true label of the i-th received signal, represents the predicted probability of the i-th received signal.

[0049] Beneficial effects of the present invention:

[0050] In the recognition model based on signal reconstruction and neural network of the present invention, the wavelet threshold estimation module and the wavelet reconstruction module constitute the signal reconstruction path, and the multi-scale feature extraction module and the prediction classification module constitute the recognition path, and the two paths enhance each other; through the end-to-end framework, signal processing and deep learning are integrated to achieve the combination of the advantages of digital signal processing technology and neural network, thereby generating the modulation category of the received signal. The wavelet threshold estimation module is introduced to predict the wavelet denoising threshold through the neural network, and its own network parameters are updated according to the back propagation to realize the adaptive wavelet transform; the signal reconstruction path is combined with the recognition path to enhance the interpretability of the network, and the adaptive wavelet transform is used to extract multi-scale modulation features from the reconstructed signal, avoiding the dependence on prior information, and significantly improving the accuracy of modulation recognition in low signal-to-noise ratio scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the steps of a low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention provided by an embodiment of the present invention;

[0052] Figure 2 A schematic diagram of the steps of an iterative training of a recognition model based on signal reconstruction and a neural network provided in an embodiment of the present invention;

[0053] Figure 3 A schematic diagram of the structure of a wavelet threshold estimation module, a wavelet reconstruction module and a multi-scale feature extraction module in a recognition model based on signal reconstruction and neural network provided in an embodiment of the present invention

[0054] Figure 4 A comparison chart of the accuracy of the low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention provided by an embodiment of the present invention on an open source data set with the existing methods;

[0055] Figure 5 A curve diagram showing the influence of the feature selectivity of the reconstructed signal on the algorithm performance of the multi-scale feature extraction module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0057] In order to solve the problem that the accuracy of existing modulation recognition methods drops sharply in low signal-to-noise ratio scenarios, the embodiment of the present invention provides a low signal-to-noise ratio modulation recognition method based on wavelet transform and channel attention (Wavelet-Based Channel Attention Network, WCAN), such as Figure 1 As shown, it may include:

[0058] S1, obtain the received signal.

[0059] For S1, the expression of the received signal is as follows:

[0060]

[0061] Among them, S r (n) represents the received signal, Indicates that from M modulation modes The modulated signal generated in the transmission process, h(n) represents the impulse response of the wireless transmission channel, * represents the convolution operation, w(n) represents the signal noise, Δf represents the additional carrier frequency offset during the transmission process, Represents the random phase shift caused by phase jitter during transmission, n=0,1,…,N-1,N, N represents the total length of the received signal.

[0062] S2, inputting the received signal into a pre-trained recognition model based on signal reconstruction and neural network, and outputting the modulation category corresponding to the received signal;

[0063] Among them, the recognition model, such as Figure 3 As shown, it can include: a wavelet threshold estimation module, which is used to perform threshold estimation on the I / Q signal component of the received signal based on a neural network to obtain a predicted wavelet denoising threshold; a wavelet reconstruction module, which is used to perform denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal; a multi-scale feature extraction module, which is used to perform feature extraction on the I-channel signal component, Q-channel signal and component IQ parallel signal component of the reconstructed signal, respectively, to obtain I-channel features, Q-channel features and IQ-channel features; based on a channel attention network, different channel features of the I-channel features, Q-channel features and IQ-channel features are fused to obtain fused features, and features with time series relationships of the fused features are obtained through a long short-term memory network; a prediction classification module, which is used to obtain the modulation category of the received signal based on the features with time series relationships of the fused features.

[0064] The expressions of the I / Q signal components of the received signal are as follows:

[0065] I signal component I(n) = real(S r (n));

[0066] Q-path signal component Q(n)=imag(S r (n));

[0067] Among them, real(S r (n)) represents the received signal S r The real part of (n), imag(S r (n)) represents the received signal S r The imaginary part of (n).

[0068] It can be understood that the I / Q components can also be referred to as in-phase / quadrature components, which are stacked in parallel.

[0069] For S2, the training process of the recognition model based on signal reconstruction and neural network may include: after several iterative trainings, after reaching a preset maximum number of iterations, a trained recognition model based on signal reconstruction and neural network is obtained; wherein, the process of one iterative training, such as Figure 2 As shown, it may include:

[0070] S21, taking all received signals as a data set; dividing the data set into a training set, a test set and a validation set;

[0071] It can be understood that the training set in the data set can be used in the corresponding training process, and the test set and the validation set can be used to test and validate the trained recognition model.

[0072] S22, for each received signal in the training set, the wavelet threshold estimation module performs threshold estimation on the I / Q signal component of the received signal to obtain a predicted wavelet denoising threshold;

[0073] For S22, this may include:

[0074] S221, using the first fully connected layer and the tanh activation layer in the wavelet threshold estimation module to process the I / Q signal components of the received signal in sequence, and obtain the feature F t ;

[0075] Feature F t The expression is as follows:

[0076] Ft=tanh(fc1(reshape(Sr(n))));

[0077] Among them, tanh() represents the tanh activation function, S r (n) represents the received signal, fc1 represents the first fully connected layer processing, and reshape() represents tensor operation;

[0078] S222, after passing the reshaped signal through the second fully connected layer, a predicted wavelet denoising threshold is obtained;

[0079] The expression of the predicted wavelet denoising threshold is as follows:

[0080]

[0081] Among them, sigmoid() represents the activation function, and fc2() represents the second fully connected layer processing;

[0082] S223, updating the network parameters of the wavelet threshold estimation module through back propagation.

[0083] It can be understood that the neural network back propagation is used to update the network parameters in the wavelet threshold estimation module, and these parameters are dynamically adjusted to achieve the optimal effect of denoising and reconstruction.

[0084] S23, a wavelet reconstruction module, denoising and reconstructing the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal, may include:

[0085] S231, a wavelet reconstruction module decomposes the received signal using a first formula based on a preset wavelet basis and a decomposition layer number to obtain a low-frequency component and a high-frequency component;

[0086] The first formula is as follows:

[0087]

[0088] Among them, Sr(n) represents the received signal, S CAk (n) represents the low-frequency part of Sr(n) at scale k after projection in scale space, S CDl (n) represents the high-frequency part of Sr(n) at scale l obtained after projection in wavelet space;

[0089] represents the scale coefficient, k represents the scale, t represents the time, φ k,t (n) represents the scaling function, represents the wavelet coefficient, ψ l,t (n) represents the wavelet function.

[0090] The received signal is decomposed into wavelet coefficients of different scales by selecting appropriate wavelet basis and decomposition layers using wavelet transform. In each layer of decomposition, the received signal is decomposed into two parts: low-frequency component and high-frequency component. Therefore, the received signal can be represented by the sum of a low-frequency signal and a finite number of high-frequency signals.

[0091] S232, denoising the high-frequency component based on the predicted wavelet denoising threshold to obtain a denoised high-frequency component;

[0092] can contain the main characteristics of the received signal, while It can include details of the received signal, such as sharp changes in noise. It can reduce the noise of the signal by discarding some high-frequency wavelet coefficients.

[0093] S233, obtaining a reconstructed signal based on the low-frequency component and the denoised high-frequency component.

[0094] S24, using a multi-scale feature extraction module, respectively extracting features from the I-path signal component, the Q-path signal component and the IQ parallel signal component of the reconstructed signal to obtain I-path features, Q-path features and IQ-path features; fusing different channel features of the I-path features, the Q-path features and the IQ-path features based on a channel attention network to obtain fused features; processing the fused features based on a long short-term memory network to obtain features with a temporal relationship of the fused features;

[0095] Optionally, the long short-term memory network may include: 2 serially connected sub-long short-term memory networks with 128 units.

[0096] The fused features are processed based on a long short-term memory network to obtain features of the fused features with a temporal relationship.

[0097] It can be understood that the features with time series relationship of the fused features can represent time series features.

[0098] Specifically, the independent information obtained in the I channel, Q channel and IQ channel is fused in the channel dimension, and an attention network (Squeeze and Excitation, SE) suitable for the characteristics of different channels of the signal is used to obtain the features with channel attention, which are then input into the two-dimensional convolution to explore spatial correlation. Considering the temporal characteristics of the modulated signal, the obtained fused features are passed through two 128-unit long short-term memory (LSTM) layers in series to effectively process the sequential data to extract temporal correlation. Then, two fully connected layers with 128 neurons and a scaling index are added to obtain the fused features with temporal relationships.

[0099] S25, using the prediction classification module, based on the features with time series relationship, obtain the predicted label of the received signal; obtain the loss function value of the current iteration training according to the true label and the predicted label; use the loss function value to update the parameters of the recognition model to obtain an updated recognition model; the updated recognition model is used for the next iteration training.

[0100] In S25, using the prediction classification module to obtain the prediction label of the received signal based on the time series feature may include:

[0101] The prediction and classification module processes the features with temporal relationships to obtain the prediction probability;

[0102] The prediction probability is processed using the prediction formula to obtain the predicted label of the received signal;

[0103] The prediction formula is as follows:

[0104]

[0105] Among them, p(yx) represents the predicted probability, y represents the true label, x represents the received signal, and argmax() represents the argmax function.

[0106] For S25, the expression of the loss function is as follows:

[0107]

[0108] Among them, Loss represents the loss value, K represents the total number of modulation categories, and y i represents the true label of the i-th received signal, represents the predicted probability of the i-th received signal.

[0109] It can be understood that the predicted label is the classification recognition result; argmax() represents the function of the predicted label, which obtains the final predicted label according to the predicted probability; the maximum number of iterations Epoch and the number of categories N can be set to determine the number of iterative processing required for training the recognition model.

[0110] The data set may include: training set, test set and validation set. The recognition model is trained epoch times using the training set in the data set. The trained recognition model can be used to perform modulation recognition on all samples in the test set.

[0111] Specifically, all samples in the test set are taken out in turn, input into the trained recognition model for classification prediction, and the predicted label corresponding to the sample is output. The predicted label of the sample is compared with the true label of the sample to obtain the classification recognition result. The recognition results of all samples are counted to obtain the classification accuracy of the model.

[0112] The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention provided by the embodiment of the present invention, in the recognition model based on signal reconstruction and neural network, the wavelet threshold estimation module and the wavelet reconstruction module constitute the signal reconstruction path, the multi-scale feature extraction module and the prediction classification module constitute the recognition path, and the two paths enhance each other; through the end-to-end framework, signal processing and deep learning are integrated to achieve the combination of the advantages of digital signal processing technology and neural network, thereby generating the modulation category of the received signal. The wavelet threshold estimation module is introduced to predict the wavelet denoising threshold through the neural network, and its own network parameters are updated according to the back propagation to realize the adaptive wavelet transform; the signal reconstruction path is combined with the recognition path to enhance the interpretability of the network, and the adaptive wavelet transform is used to extract multi-scale modulation features from the reconstructed signal, avoiding the dependence on prior information, and significantly improving the accuracy of modulation recognition in low signal-to-noise ratio scenarios.

[0113] In order to demonstrate the beneficial effects of the low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention proposed in the embodiment of the present invention, the beneficial effects of the embodiment of the present invention are further illustrated through simulation experiments.

[0114] Open Source Dataset Experiments

[0115] The simulation experiment uses the standard open source data set RML2016.10a, which can include 220,000 modulated signals with signal-to-noise ratios (SNRs) ranging from -20dB to +18dB, and considers 11 commonly used modulations: WBFM, AM-DSB, AM-SSB, BPSK, CPFSK, GFSK, 4-PAM, 16-QAM, 64-QAM, QPSK, and 8PSK. Each signal in the data set has 128 time I / Q samples in complex form and is generated in a harsh simulated propagation environment that is corrupted by AWGN, multipath fading, sampling rate offset, and center frequency offset to simulate the actual environment.

[0116] The experiment divides the dataset into three parts in a ratio of 6:3:1: training set, test set, and validation set. Specifically, for each modulation type and signal-to-noise ratio, 1000 signals are randomly divided into 600 signals for training and 300 signals for testing. The Adam optimizer is used in model training, and the learning rate of each round is set to change with the number of training rounds. The learning rate λ = 1 / (0.1*Epoch+1), and the batch size of gradient updates during training can be 256.

[0117] In order to evaluate the effectiveness of the low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention under low signal-to-noise ratio conditions, the embodiment of the present invention selected 10 baseline methods for comparison. Specifically, there are CNN1, CNN2, ResNet, LSTM, GRU, CLDNN, DenseNet, CGDNet, PET-CGDNN and MAC models. The above data sets were used to conduct experiments on the low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention proposed in the embodiment of the present invention and the baseline method. For the accuracy comparison chart with the existing methods on the open source data set, please refer to Figure 4 .from Figure 4 It can be seen that this method achieves the best recognition effect, especially when the signal-to-noise ratio is lower than 0dB. The recognition accuracy of the low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention proposed in the embodiment of the present invention exceeds other existing methods. This result proves that the combination of DSP and deep neural network is effective. The WCAN proposed in the embodiment of the present invention can perform adaptive threshold denoising and reconstruction for signals with different signal-to-noise ratios. When the signal-to-noise ratio is equal to -4dB, the accuracy of the RML2016.10A dataset is also improved by 8.21% compared with the current method with the highest accuracy.

[0118] The test accuracy of different methods under all signal-to-noise ratios and the number of learning parameters of the model are compared, as shown in Table 1, which is a comparison table of the complexity and accuracy of the method proposed in the present invention and other models.

[0119] Table 1 Comparison of the complexity and accuracy of the method proposed in this invention and other models

[0120] Methods Learning parameters Accuracy △(↑％) CNN1 1.58M 53.14％ 9.38％ CNN2 0.86M 58.35％ 4.17％ ResNet 3.11M 59.11％ 3.41％ LSTM 0.23M 60.21％ 2.31％ GRU 0.15M 58.20％ 4.32％ PET-CGDNN 0.10M 59.95％ 2.57％ DenseNet 4.23M 60.28％ 2.24％ CLDNN 0.51M 58.02％ 4.50％ MAC 2.74M 60.15％ 2.37％ CGDNet 0.12M 58.95％ 3.57％ WCAN 0.72M 62.52％ -

[0121] In Table 1, Methods represents methods, Learning parameters represents learning parameters, Accuracy represents total accuracy, and △ represents the gap between the baseline method and the WCAN model. It can be seen from Table 1 that the WCAN proposed in the embodiment of the present invention still achieves satisfactory performance with a smaller number of model parameters, with a total accuracy of 62.52%, which reflects the effectiveness of the method WCAN proposed in the present invention in combining DSP with deep neural network.

[0122] Model performance analysis experiment

[0123] In order to verify the selectivity of the multi-scale feature extraction module for the reconstructed signal, the provided model is compared with the results after removing the attention network SE and the current mainstream attention modules, including spatial attention SA, convolutional block attention CBAM, triple attention TA and other attention network modules. For a graph of the influence of the selectivity of the multi-scale feature extraction module for the reconstructed signal on the algorithm performance, please refer to Figure 5 .from Figure 5 As can be seen from Figure 2, the attention network SE can selectively treat multi-scale features in the channel dimension and obtain the best recognition results. The selectable I channel, Q channel, and I / Q multi-channel maximize the use of complementary information after wavelet reconstruction.

[0124] Through simulation experiments, it is further proved that the wavelet threshold estimation module and wavelet reconstruction module in the recognition model based on signal reconstruction and neural network proposed in the embodiment of the present invention constitute the signal reconstruction path, and the multi-scale feature extraction module and the prediction classification module constitute the recognition path, and the two paths enhance each other; through the end-to-end framework, signal processing and deep learning are integrated to achieve the combination of the advantages of digital signal processing technology and neural networks, thereby generating the modulation category of the received signal. The wavelet threshold estimation module is introduced to predict the wavelet denoising threshold through the neural network, and its own network parameters are updated according to the back propagation to realize the adaptive wavelet transform; the signal reconstruction path is combined with the recognition path to enhance the interpretability of the network, and the adaptive wavelet transform is used to extract multi-scale modulation features from the reconstructed signal, avoiding the dependence on prior information, and significantly improving the accuracy of modulation recognition in low signal-to-noise ratio scenarios.

[0125] It should be noted that in the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A low signal-to-noise ratio conditional modulation recognition method based on wavelet transform and channel attention, characterized in that: include: Get the received signal; Inputting the received signal into a pre-trained recognition model based on signal reconstruction and neural network, and outputting a modulation category corresponding to the received signal; Among them, the recognition model includes: a wavelet threshold estimation module, which is used to perform threshold estimation on the I / Q signal component of the received signal based on a neural network to obtain a predicted wavelet denoising threshold; a wavelet reconstruction module, which is used to perform denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal; a multi-scale feature extraction module, which is used to perform feature extraction on the I-path signal component, Q-path signal component and IQ parallel signal component of the reconstructed signal, respectively, to obtain I-path features, Q-path features and IQ-path features; based on a channel attention network, different channel features of the I-path features, Q-path features and IQ-path features are fused to obtain fused features, and features with time series relationships of the fused features are obtained through a long short-term memory network; a prediction classification module, which is used to obtain the modulation category of the received signal based on the features with time series relationships of the fused features.

2. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 1 is characterized in that: The expression of the received signal is as follows: Among them, S r (n) represents the received signal, Indicates that from M modulation modes The modulated signal generated in the transmission process, h(n) represents the impulse response of the wireless transmission channel, * represents the convolution operation, w(n) represents the signal noise, Δf represents the additional carrier frequency offset during the transmission process, Represents the random phase shift caused by phase jitter during transmission, n=0,1,…,N-1,N, N represents the total length of the received signal.

3. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 1 is characterized in that: The expressions of the I / Q signal components of the received signal are as follows: I signal component I(n) = real(S r (n)); Q-path signal component Q(n)=imag(S r (n)); Among them, real(S r (n)) represents the received signal S r The real part of (n), imag(S r (n)) represents the received signal S r The imaginary part of (n).

4. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 3 is characterized in that: The training process of the recognition model based on signal reconstruction and neural network includes: after several iterative trainings, after reaching a preset maximum number of iterations, a trained recognition model based on signal reconstruction and neural network is obtained; wherein, the process of one iterative training includes: Taking all received signals as a data set; dividing the data set into a training set, a test set and a validation set; For each received signal in the training set, the wavelet threshold estimation module performs threshold estimation on the I / Q signal component of the received signal to obtain a predicted wavelet denoising threshold; The wavelet reconstruction module performs denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal; Using the multi-scale feature extraction module, feature extraction is performed on the I-path signal component, Q-path signal component and IQ parallel signal component of the reconstructed signal respectively to obtain I-path features, Q-path features and IQ-path features; based on the channel attention network, different channel features of the I-path features, Q-path features and IQ-path features are fused to obtain fused features; based on the long short-term memory network, the fused features are processed to obtain features with a temporal relationship of the fused features; Utilizing the prediction classification module, based on the features with a time series relationship, a predicted label of the received signal is obtained; the loss function value of the current iterative training is obtained according to the true label and the predicted label; the parameters of the recognition model are updated using the loss function value to obtain an updated recognition model; the updated recognition model is used for the next iterative training.

5. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 4 is characterized in that: The wavelet threshold estimation module performs threshold estimation on the I / Q signal component of the received signal to obtain a predicted wavelet denoising threshold, including: The I / Q signal components of the received signal are processed in sequence using the first fully connected layer and the tanh activation layer in the wavelet threshold estimation module to obtain the feature F t ; The feature F t The expression is as follows: F t =tanh(fc1(reshape(S r (n)))); Among them, tanh() represents the tanh activation function, S r (n) represents the received signal, fc1 represents the first fully connected layer processing, and reshape() represents tensor operation; After passing the reshaped signal through a second fully connected layer, the predicted wavelet denoising threshold is obtained; The expression of the predicted wavelet denoising threshold is as follows: Among them, sigmoid() represents the activation function, and fc2() represents the second fully connected layer processing; The network parameters of the wavelet threshold estimation module are updated through back propagation.

6. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 4 is characterized in that: The wavelet reconstruction module performs denoising and reconstruction on the received signal based on the predicted wavelet denoising threshold to obtain a reconstructed signal, including: The wavelet reconstruction module decomposes the received signal using a first formula based on a preset wavelet basis and a decomposition layer number to obtain a low-frequency component and a high-frequency component; The first formula is as follows: Among them, S r (n) represents the received signal, Indicates S r (n) The low-frequency part at scale k obtained after scale space projection, Indicates S r (n) The high-frequency part at scale l obtained after projection into wavelet space; De-noising the high-frequency component based on the predicted wavelet de-noising threshold to obtain a de-noised high-frequency component; A reconstructed signal is obtained based on the low-frequency component and the denoised high-frequency component.

7. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 1 is characterized in that: The long short-term memory network includes: 2 serially connected sub-long short-term memory networks with 128 units.

8. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 4 is characterized in that: Using the prediction classification module to obtain a prediction label of the received signal based on the features with a time series relationship, including: the prediction classification module processes the features with a time series relationship to obtain a prediction probability; Processing the predicted probability using a prediction formula to obtain a predicted label of the received signal; The prediction formula is as follows: Among them, p(yx) represents the predicted probability, y represents the true label, x represents the received signal, and argmax() represents the argmax function.

9. The low signal-to-noise ratio condition modulation recognition method based on wavelet transform and channel attention according to claim 4 is characterized in that: The expression of the loss function is as follows: Among them, Loss represents the loss value, K represents the total number of modulation categories, and y i represents the true label of the i-th received signal, represents the predicted probability of the i-th received signal.

Images