Signal modulation identification method, device, electronic device and storage medium

Abstract

The present disclosure provides a signal modulation recognition method, device, electronic device and storage medium, and relates to the field of signal processing technology. The method includes: obtaining an IQ signal; inputting the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, the amplitude-phase feature subnetwork includes a long short-term memory network LSTM and an attention model; inputting the IQ signal into a time-frequency feature subnetwork to obtain a time-frequency feature, the time-frequency feature subnetwork includes a capsule network; performing feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fusion feature; performing modulation recognition based on the fusion feature. The present disclosure realizes the complementary gain of multimodal information and improves the accuracy and robustness of signal modulation recognition.

Description

Background Art

[0002] With the continuous innovation and upgrading of communication technology, the modulation modes of communication signals are becoming more and more diverse. In non-cooperative communication, the receiving end of the communication system cannot know the specific modulation mode of the transmitted signal, so modulation recognition methods are needed to detect the modulation type of the signal. Modulation recognition methods play an important role in cognitive radio, electronic countermeasures, spectrum detection, etc.

[0003] In the related art, a modulation recognition method based on likelihood function and a pattern recognition method based on features are used to perform modulation recognition of communication signals. The modulation recognition performed by these two recognition methods has low accuracy.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present disclosure provides a signal modulation recognition method, device, electronic device and storage medium, which at least to a certain extent overcome the problem of low modulation recognition accuracy due to related technologies.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, a signal modulation recognition method is provided, the method comprising: obtaining an in-phase orthogonal IQ signal; inputting the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, wherein the amplitude-phase feature subnetwork comprises a long short-term memory network (LSTM) and an attention model; inputting the IQ signal into a time-frequency feature subnetwork to obtain a time-frequency feature, wherein the time-frequency feature subnetwork comprises a capsule network; performing feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fused feature; and performing modulation recognition based on the fused feature.

[0008] In one embodiment of the present disclosure, before inputting the IQ signal into the amplitude-phase feature subnetwork to obtain the amplitude-phase feature, the method further includes: preprocessing the IQ signal to obtain an amplitude sequence signal and a phase sequence signal; wherein, inputting the IQ signal into the amplitude-phase feature subnetwork to obtain the amplitude-phase feature includes: inputting the amplitude sequence signal and the phase sequence signal into the LSTM to obtain a hidden state vector; and inputting the hidden state vector into the attention model to obtain the amplitude-phase feature.

[0009] In one embodiment of the present disclosure, the inputting the hidden state vector into the attention model to obtain the amplitude-phase feature includes: encoding the hidden state vector by a first weight of the attention model to obtain a coding vector; performing an attention operation on the coding vector and an attention matrix based on a dot product alignment function to obtain a tracking coding vector; normalizing the tracking coding vector to obtain a second weight; and performing a weighted linear combination of the second weight and the hidden state vector to obtain the amplitude-phase feature.

[0010] In one embodiment of the present disclosure, before inputting the IQ signal into the time-frequency feature subnetwork to obtain the time-frequency feature, the method includes: converting the IQ signal into a time-frequency graph of the signal based on the time-frequency distribution CWD transformation; wherein, inputting the IQ signal into the time-frequency feature subnetwork to obtain the time-frequency feature includes: inputting the time-frequency graph into the capsule network to obtain the time-frequency feature.

[0011] In one embodiment of the present disclosure, the capsule network is a capsule network with multi-scale convolution and multi-path skip residual connection.

[0012] In one embodiment of the present disclosure, the feature fusion of the amplitude and phase features and the time-frequency features to obtain the fused features includes: inputting the amplitude and phase features and the time-frequency features into a feature fusion subnetwork to obtain the fused features, the feature fusion subnetwork includes a fully connected layer and an output layer, and the activation function of the output layer is a Softmax function.

[0013] In one embodiment of the present disclosure, the modulation identification based on the fusion feature includes: obtaining multiple signal fusion features, one signal fusion feature corresponding to one modulation type; selecting a target signal fusion feature with the highest matching degree with the fusion feature from the multiple signal fusion features; the modulation type corresponding to the target signal fusion feature is the modulation type of the IQ signal.

[0014] According to another aspect of the present disclosure, a signal modulation identification device is provided, the device comprising: an acquisition module, used to acquire an in-phase orthogonal IQ signal; an amplitude-phase feature generation module, used to input the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, wherein the amplitude-phase feature subnetwork comprises a long short-term memory network LSTM and an attention model; a time-frequency feature generation module, used to input the IQ signal into a time-frequency feature subnetwork to obtain a time-frequency feature, wherein the time-frequency feature subnetwork comprises a capsule network; a fusion module, used to perform feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fused feature; and an identification module, used to perform modulation identification based on the fused feature.

[0015] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any of the above-mentioned signal modulation identification methods by executing the executable instructions.

[0016] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any of the above-mentioned signal modulation identification methods is implemented.

[0017] According to another aspect of the present disclosure, a computer program product is provided, wherein the computer program product comprises a computer program or computer instructions, wherein the computer program or the computer instructions are loaded and executed by a processor so as to enable a computer to implement any of the above-mentioned signal modulation identification methods.

[0018] The signal modulation identification method, device, equipment and medium provided by the embodiments of the present disclosure obtain an IQ signal; input the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, the amplitude-phase feature subnetwork includes a long short-term memory network LSTM and an attention model; input the IQ signal into a time-frequency feature subnetwork to obtain a time-frequency feature, the time-frequency feature subnetwork includes a capsule network; perform feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fusion feature; perform modulation identification based on the fusion feature. The present disclosure performs feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fusion feature, and performs modulation identification based on the fusion feature, thereby achieving complementary gain of multimodal information and improving the accuracy and robustness of signal modulation identification.

[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0021] Figure 1 A schematic diagram showing a signal modulation identification system architecture in an embodiment of the present disclosure;

[0022] Figure 2 A flow chart of a signal modulation identification method in one embodiment of the present disclosure is shown;

[0023] Figure 3A structural block diagram of a signal modulation identification system in an embodiment of the present disclosure is shown;

[0024] Figure 4 A flow chart of a signal modulation identification method in another embodiment of the present disclosure is shown;

[0025] Figure 5 A schematic diagram showing an amplitude-phase characteristic subnetwork in an embodiment of the present disclosure;

[0026] Figure 6 A schematic diagram showing a time-frequency feature subnetwork in an embodiment of the present disclosure;

[0027] Figure 7 A schematic diagram showing a feature fusion subnetwork in an embodiment of the present disclosure is shown;

[0028] Figure 8 A schematic diagram showing signal fusion characteristics of 11 different modulation types in an embodiment of the present disclosure;

[0029] Fig. 9 A schematic diagram showing the accuracy of Random Forest in identifying 11 modulation types;

[0030] Fig.10 A schematic diagram showing the accuracy of CNN in identifying 11 modulation types;

[0031] Fig.11 A schematic diagram showing the accuracy of CLDNN in identifying 11 modulation types;

[0032] Fig.12 A schematic diagram showing the accuracy of LA-CapsNets in identifying 11 modulation types;

[0033] Fig.13 A schematic diagram of the confusion matrix of Random Forest when the signal-to-noise ratio is 8dB;

[0034] Fig.14 A schematic diagram of the confusion matrix of CNN when the signal-to-noise ratio is 8dB is shown;

[0035] Fig.15 A schematic diagram of the confusion matrix of CLDNN when the signal-to-noise ratio is 8dB is shown;

[0036] Fig.16 A schematic diagram showing the confusion matrix of LA-CapsNets when the signal-to-noise ratio is 8dB;

[0037] Fig.17 A schematic diagram of a signal modulation identification device in an embodiment of the present disclosure is shown;

[0038] Fig.18 A structural block diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0039] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0040] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0041] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0042] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0043] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0044] In the related art, modulation recognition methods include a modulation recognition method based on likelihood function and a pattern recognition method based on feature. The modulation recognition method based on likelihood function and the pattern recognition method based on feature are described below one by one.

[0045] For modulation recognition methods based on likelihood functions, the amount of computation is large and requires a lot of prior information. In non-cooperative communication, a large number of signal parameters are unknown. Maximizing the unknown information will lead to classification errors and is difficult to achieve in practical applications.

[0046] The feature-based pattern recognition method is easier to implement than the modulation recognition method based on likelihood function. It needs to extract features from the received signal and then determine the signal modulation type through a classifier. However, feature extraction depends on expert experience, and the quality of extracted features is related to the selection of classifiers. Improper selection will seriously affect the recognition results.

[0047] The above two modulation recognition methods both have the problem of low modulation recognition accuracy.

[0048] In order to solve the above problems, the present disclosure provides a signal modulation recognition method, which is applied to various scenarios of signal processing, such as modulation recognition scenarios of communication signals, cloud network operation scenarios, etc. The present disclosure fuses the amplitude phase features and the time-frequency features to obtain fused features, and performs modulation recognition based on the fused features, thereby achieving complementary gain of multi-modal information and improving the accuracy and robustness of signal modulation recognition.

[0049] The specific implementation of the embodiment of the present disclosure is described in detail below with reference to the accompanying drawings.

[0050] Figure 1 A schematic diagram showing a signal modulation identification system architecture in an embodiment of the present disclosure is shown. The system can apply the signal modulation identification method or signal modulation identification device in each embodiment of the present disclosure.

[0051] like Figure 1 As shown, the system architecture may include a signal sending device 101 and a signal receiving device 102. The signal sending device 101 is used to send a communication signal, and the signal receiving device 102 is used to receive the communication signal sent by the signal sending device 101 and perform modulation identification on the communication signal.

[0052] The signal sending device 101 and the signal receiving device 102 are connected to each other through a network, which may be a wired network or a wireless network.

[0053] Optionally, the wireless network or wired network described above uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. are used to represent data exchanged through the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above data communication technologies.

[0054] The signal sending device 101 may be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, wearable devices, augmented reality devices, virtual reality devices, etc. The signal sending device 101 may also be a network device, including but not limited to a base station, a router, and a transmitter.

[0055] Optionally, the client of the application installed in different signal sending devices 101 is the same, or is a client of the same type of application based on different operating systems. Based on different terminal platforms, the specific form of the client of the application can also be different, for example, the application client can be a mobile client, a PC client, etc.

[0056] The signal receiving device 102 may be a server that provides various services, such as a background management server that provides support for the device operated by the signal sending device 101. The background management server may analyze and process the received request and other data, and feed back the processing results to the terminal device.

[0057] Optionally, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The terminal can be a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0058] Those skilled in the art will know that Figure 1 The number of the signal sending devices 101 and the signal receiving devices 102 in the embodiment is only for illustration, and any number of the signal sending devices 101 and the signal receiving devices 102 may be provided according to actual needs. The embodiments of the present disclosure are not limited to this.

[0059] The present exemplary implementation is described in detail below with reference to the accompanying drawings and embodiments.

[0060] First, a signal modulation identification method is provided in an embodiment of the present disclosure, and the method can be executed by any electronic device with computing and processing capabilities.

[0061] Figure 2 A flow chart of a signal modulation identification method in an embodiment of the present disclosure is shown as follows: Figure 2 As shown, the signal modulation identification method provided in the embodiment of the present disclosure may include the following S201 to S205.

[0062] S201, obtaining an in-phase and quadrature IQ signal.

[0063] The IQ (In-phase Quadrate) signal is a signal generated by a signal receiving device after receiving the RF signal by performing signal processing on the RF signal, that is, generating an in-phase component I signal and a quadrature component Q signal. The signal receiving device may be a receiver.

[0064] S202, input the IQ signal into the amplitude-phase feature sub-network to obtain the amplitude-phase feature, and the amplitude-phase feature sub-network includes a long short-term memory network LSTM and an attention model.

[0065] The attention model can be an attention mechanism, which is a special structure embedded in a machine learning model to automatically learn and calculate the contribution of input data to output data. LSTM (Long Short Term Memory) learns the temporal dependency of IQ signals and extracts the signal features of IQ signals. The attention mechanism is connected in series after LSTM, and the attention mechanism assigns weights to the signal features of the IQ signal, thereby obtaining amplitude and phase features.

[0066] S203, inputting the IQ signal into the time-frequency feature sub-network to obtain time-frequency features, where the time-frequency feature sub-network includes a capsule network.

[0067] Among them, Capsule Networks are good at processing image data and can make full use of the feature information of different time and space of the data. In the communication system, the signal is affected by noise, causing its time-frequency diagram to undergo a certain degree of deformation. The main capsule layer in the Capsule Network can learn the weight information in the image and then extract the time-frequency characteristics of the signal. The network architecture with multi-scale jump connections can realize the fusion of local feature information and global feature information, which is conducive to extracting features of different branches.

[0068] S204, performing feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fused feature.

[0069] By fusing the amplitude and phase features with the time-frequency features, the complementary gain of multimodal information can be achieved, and the recognition of the IQ signal to be identified can be improved, so that the modulation type of the IQ signal can be accurately modulated and identified. Among them, the multimodal information can include the time domain features and frequency domain features (amplitude and phase features) of the IQ signal.

[0070] It should be noted that the feature fusion of amplitude and phase features and time-frequency features can be achieved through a feature fusion sub-network. The specific feature fusion sub-network is not limited in the embodiments of the present disclosure. For example, the feature fusion sub-network can be a convolutional neural network.

[0071] S205, performing modulation recognition based on the fused features.

[0072] like Figure 3 As shown, the amplitude phase feature and the time-frequency feature are subjected to feature fusion 33, and the obtained fusion feature is input into the classification network 34. The classification network 34 can store signal fusion features of all types of modulation types, and match the target signal fusion feature with the highest similarity to the fusion feature among all modulation types. The modulation type corresponding to the target signal fusion feature is the modulation type of the IQ signal. Among them, the classification network 34 can be a convolutional neural network with a fully connected layer.

[0073] The present invention obtains fused features by fusing amplitude-phase features and time-frequency features, and performs modulation recognition based on the fused features, thereby achieving complementary gains of multimodal information and improving the accuracy and robustness of signal modulation recognition.

[0074] The signal modulation identification method is described above. Now, how to obtain the amplitude and phase characteristics is described through the following exemplary embodiments.

[0075] In an exemplary embodiment, before inputting the IQ signal into the amplitude-phase feature subnetwork to obtain the amplitude-phase feature, the signal modulation identification method may further include: preprocessing the IQ signal to obtain an amplitude sequence signal and a phase sequence signal; wherein, inputting the IQ signal into the amplitude-phase feature subnetwork to obtain the amplitude-phase feature includes: inputting the amplitude sequence signal and the phase sequence signal into the LSTM to obtain a hidden state vector; and inputting the hidden state vector into the attention model to obtain the amplitude-phase feature. It should be noted that the amplitude sequence signal and the phase sequence signal extract signal features through the LSTM network, and the data in the LSTM network is unfolded according to the timeline and sequentially passes through the forget gate, the input gate, and the output gate.

[0076] like Figure 3 As shown, the preprocessing module 36 preprocesses the IQ signal to obtain an amplitude sequence signal and a phase sequence signal, and inputs the amplitude sequence signal and the phase sequence signal into the LSTM 31. The LSTM 31 performs well in processing tasks such as text and time series data, and the communication signal is a time series. The amplitude sequence signal and the phase sequence signal obtained by sampling each time step are converted into amplitude and phase, and normalized to obtain polarity representation information, and then the LSTM 31 is used to effectively learn the temporal dependency of these sequences. The attention mechanism 32 is cascaded after the LSTM 31. The attention mechanism 32 can distribute attention weights (second weights) for the features learned by the LSTM 31, thereby allowing the amplitude and phase feature sub-network to focus on more important features.

[0077] In another exemplary embodiment, inputting the hidden state vector into the attention model to obtain the amplitude-phase feature may include the following S401 to S404.

[0078] S401, encode the hidden state vector using the first weight of the attention model to obtain a coding vector.

[0079] like Figure 5 As mentioned above, A1…A 128 Used to indicate amplitude, P1…P 128 Used to indicate phase, A1…A 128 With P1…P 128One-to-one correspondence, that is, A1 corresponds to P1, and the corresponding A1 and P1 have the same timestamp. The input dimension of LSTM is 128×2, the number of cells is 128, and the weights in the cells are L2 regularized (a weight penalty term), the L2 norm is 0.001, and the time length of LSTM input and output is set to remain unchanged, that is, the hidden state vector of all time steps is obtained, which is expressed as: [h1, h2, h3..., h 128 ].

[0080] The hidden state vector is input into the attention model, and the first weight of the attention model is W s and b s , the hidden state vector is encoded by the following formula 1.

[0081] m i =tanh(W s ·h i +b s ) (1)

[0082] Among them, m i is the encoding vector of the i-th hidden state vector, W s and b s is the first weight, h i is the i-th hidden state vector, i is a natural number. In this implementation, the value range of i is [1, 128].

[0083] S402, performing an attention operation on the coding vector and the attention matrix based on the dot product alignment function to obtain a tracking coding vector.

[0084] Among them, the attention matrix is ​​a trainable attention matrix U randomly initialized by the attention model. The attention matrix U is consistent with the encoding vector m i The dot product-based alignment function is used to perform the attention operation to obtain the tracking encoding vector, which can be expressed as: m i U.

[0085] S403, performing a normalization operation on the tracking code vector to obtain a second weight.

[0086] For m i U performs Softmax (normalization) operation to obtain the second weight a i , which is calculated as shown in the following formula 2:

[0087]

[0088] Among them, a i is the second weight of the ith tracking code vector, m i U is the tracking code vector, and i is a natural number.

[0089] S404, performing a weighted linear combination on the second weight and the hidden state vector to obtain an amplitude-phase feature.

[0090] The learned second weight a i The hidden state vector h i Make a weighted linear combination to assign a second weight to the hidden state vector extracted by LSTM. At this time, the amplitude phase feature of the network is v ij , the calculation formula is shown in Formula 3 below:

[0091]

[0092] Among them, v ij is the amplitude phase characteristic, α ij is the second weight of the i-th row corresponding to the j-th column, h i is the i-th hidden state vector, and j is a natural number.

[0093] It should be noted that Figure 5 In, v ij The dimension is 128×1, that is, the value of j is 1.

[0094] The above exemplary embodiments illustrate how to obtain amplitude and phase characteristics. The following exemplary embodiments illustrate how to obtain time-frequency characteristics.

[0095] In an exemplary embodiment, before inputting the IQ signal into the time-frequency feature subnetwork to obtain the time-frequency feature, the signal modulation identification method may include: converting the IQ signal into a time-frequency graph of the signal based on the time-frequency distribution CWD transform; wherein, inputting the IQ signal into the time-frequency feature subnetwork to obtain the time-frequency feature includes: inputting the time-frequency graph into the capsule network to obtain the time-frequency feature. The embodiments of the present disclosure do not limit the specifics of the capsule network, as long as it is a model that can achieve the above functions, for example, the capsule network can be a capsule network with multi-scale convolution and multi-path skip residual connection, such as Figure 6 As shown, the capsule network includes a feature extraction network, a main capsule layer and a digital capsule layer, wherein the feature extraction network may include a convolution layer and a BN (Batch Normalization), the main capsule layer may be a model formed by multi-scale convolution and multi-path skip residual connection, and the digital capsule layer may include three DigitCaps (upper capsule layer), and DigitCaps is used to output the output 16D vector of the layer, which contains the instantiation parameters required to reconstruct the object.

[0096] like Figure 3As shown, the IQ signal is input into CWD (Choi-Willmas Distribution) 37, and the time-frequency diagram is output. The obtained time-frequency diagram is used as the input of the capsule network 35. The capsule network 35 is good at processing image data and can make full use of the feature information of different time and space of the data. In the communication system, the IQ signal is affected by noise, which causes a certain degree of deformation in its time-frequency diagram. The main capsule layer in the capsule network 35 can learn the weight information in the image, and then extract the time-frequency characteristics of the IQ signal. The network architecture of multi-scale jump connection can realize the fusion of local feature information and global feature information, which is conducive to extracting features of different branches.

[0097] This paper adopts a multi-scale convolution multi-path residual capsule network to construct a time-frequency feature sub-network. First, the capsule network based on multi-scale convolution and multi-path skip residual connection has multi-scale learning ability, such as Figure 6 As shown in the figure, three branches with different convolution sizes (3×3, 5×5, 7×7) are used for multi-scale feature extraction, and the higher-dimensional features of the three branches are extracted by dense cross-connection to optimize the capsule quality, increase the receptive field of feature extraction, and enhance the adaptability of capsules to data, thereby realizing the extraction of multi-scale features of time-frequency data. Secondly, a main capsule network with multiple densely connected residual blocks of different scales is used to extract features of different spatial scales of the data, and then the network is iteratively trained alternately by fusing global residuals and local residuals to obtain time-frequency features.

[0098] The above describes how to obtain time-frequency features through exemplary embodiments. The following describes how to perform feature fusion.

[0099] In an exemplary embodiment, the amplitude phase feature and the time-frequency feature are fused to obtain a fused feature, including: inputting the amplitude phase feature and the time-frequency feature into a feature fusion subnetwork to obtain a fused feature, wherein the feature fusion subnetwork includes a fully connected layer and an output layer, and the activation function of the output layer is a Softmax function. The feature fusion subnetwork may be a convolutional neural network including a fully connected layer.

[0100] like Figure 7 As shown, the amplitude phase feature v output by the amplitude phase feature sub-network is i , and the time-frequency feature c output by the time-frequency feature subnetwork iThe purpose of the feature fusion sub-network is to fuse the two features (amplitude phase features and time-frequency features). In order to better fuse the two features, a spliced ​​convolutional neural network is used, and then a fully connected layer with 128 neurons is used to reduce the feature dimension. In the output layer of the feature fusion sub-network, the activation function is Softmax. The fused features obtained after feature fusion are sent to the classification network, and finally the recognition results are output.

[0101] The above describes feature fusion, and the following describes how to perform modulation recognition.

[0102] In an exemplary embodiment, multiple signal fusion features are obtained, and one signal fusion feature corresponds to one modulation type; a target signal fusion feature with the highest matching degree with the fusion feature is selected from the multiple signal fusion features; the modulation type corresponding to the target signal fusion feature is the modulation type of the IQ signal.

[0103] The present disclosure does not limit how to match signal fusion features and fusion features. For example, the closest target signal fusion feature is determined by calculating the cosine similarity or Euclidean distance between the fusion feature and the signal fusion feature. Exemplarily, the cosine similarity between the fusion feature and one of the signal fusion features is 1, which means that the fusion feature is the same as the signal fusion feature, the signal fusion feature is the target signal fusion feature, and the modulation type corresponding to the target signal fusion feature is the modulation type of the IQ signal.

[0104] Modulation recognition methods in related technologies usually use prior information of data for recognition. The present invention uses a deep learning multi-scale capsule network for multi-modal modulation recognition, relying only on end-to-end feature mapping between networks, reducing the high dependence on prior information. The present invention effectively avoids the drawbacks of classifier selection, directly uses a deep learning network to automatically extract deep features of the signal, mines the deep features behind the data, and extracts the invisible features of the signal (hidden state vector).

[0105] The effect of modulation recognition using the method disclosed in the present invention is described below.

[0106] The open source dataset RadioML2016.10a established by O'Shea et al. is used as a benchmark for this study. The specific parameters of the dataset are shown in Table 1.

[0107] Table 1

[0108]

[0109] like Figure 8As described, the IQ data feature diagram (time domain waveform diagram) of 11 different modulation types when the signal-to-noise ratio is 18dB. It can be seen from the figure that there are some differences in the time domain diagrams of each modulation type, but due to pulse shaping, distortion and other influences of the channel, the decision features are not obvious. Therefore, the present invention uses a deep learning method to start from the IQ signal and fully explore the deep features of the data, so as to effectively identify the modulation type of the signal.

[0110] Under the same experimental conditions, the LA-CapsNets (LSTM with Attention and Capsule Networks) model designed using the method disclosed in the present invention was compared with other modulation recognition methods, including Random Forest with 150 decision trees, Convolutional Neural Networks (CNN) and Convolutional Long short-term Deep Neural Network (CLDNN).

[0111] The average recognition rate and peak recognition rate of the four schemes between -20dB and 18dB are shown in Table 2.

[0112] Table 2 Overall recognition rate

[0113] Random Forest CNN CLDNN LA-CapsNets Average recognition rate (%) 40.70 51.77 55.67 60.81 Peak recognition rate (%) 71.30 76.44 82.67 91.82

[0114] It can be seen from Table 2 that in terms of average recognition rate, the LA-CapsNets model reaches more than 60%, which is 49.41%, 17.46% and 9.23% higher than the Random Forest, CNN and CLDNN models respectively. In the case of high signal-to-noise ratio, the advantages of the model are more obvious, and the peak recognition rates are increased by 28.78%, 20.12% and 11.06% respectively. This shows that the signal modulation recognition method proposed in the present disclosure effectively combines the advantages of the signal time domain and frequency domain. Compared with traditional single-modal input, LA-CapsNets can extract wider and deeper features of data from different dimensions, which is conducive to improving the accuracy of signal modulation recognition.

[0115] In order to deeply analyze the changes of different modulation types with signal-to-noise ratio in each model, Figure 9-12 The recognition accuracy of 11 modulation types in four models is shown. Figure 9-12In the figure, the horizontal axis represents the signal-to-noise ratio and the vertical axis represents the accuracy. It can be seen from the figure that the recognition accuracy of each modulation mode in each modulation recognition method approaches the ideal performance as the signal-to-noise ratio increases. Among them, Random Forest can only recognize relatively simple modulation methods due to its own performance limitations; in CNN, the recognition performance has been improved to a certain extent, but its recognition accuracy for 16QAM is low. This is because CNN uses the original IQ signal and the network cannot learn enough features to distinguish them; the CLDNN solution connects a layer of LSTM network in series after CNN to further mine the characteristics of the data, and borrows the idea of ​​residual in the convolution layer to avoid overfitting of the network, so compared with the CNN solution, CLDNN has a significant improvement; compared with the other three solutions, the linear growth area of ​​LA-CapsNets is obviously shifted to the low signal-to-noise ratio interval, and the signal-to-noise ratio length of the linear interval is also reduced.

[0116] Since the LA-CapsNets scheme combines two modal inputs, the recognition rate of various modulation types has been greatly improved compared to the input of a single modality. In particular, it has been further improved in identifying the MQAM modulation mode. This is because the input data used in LA-CapsNets is the amplitude and phase features after polar coordinate mapping, which corresponds to the distinguishing features of MQAM-type signals. In addition, its curve smoothness is better, which shows that the LA-CapsNets model has a stronger generalization ability for noise and a higher recognition rate for modulation types.

[0117] In order to more intuitively compare the main types of misjudgments in the modulation recognition model, a visual confusion matrix diagram is drawn to further analyze the reasons for limiting the recognition rate, such as Figure 13-Figure 16 Shown is the confusion matrix of the four schemes when the signal-to-noise ratio is 8dB. The gradient bar graph on the right is used to indicate the accuracy of modulation recognition. The darker the color, the higher the accuracy. The horizontal axis of the confusion matrix is ​​the predicted label, and the vertical axis is the actual label.

[0118] Depend on Figure 13-Figure 16 It can be observed that all four schemes have different degrees of misjudgment between 16QAM and 64QAM. Since 16QAM and 64QAM both belong to MQAM modulation and have the same modulation principle, they cause misjudgment between the two modes. The misjudgment degree of the two modes in LA-CapsNets is the lowest, and there is basically no confusion between 16QAM and 64QAM. This also shows the effectiveness of fusing multiple modal information of IQ signals. In addition, Figure 13-Figure 16The recognition rate of WBFM signal is not high, and there is still a lot of confusion between AM-DSB signal and WBFM signal in the confusion matrix, and the model cannot distinguish the two modulation types well. This is because WBFM and AM-DSB modulation are generated by real speech signals. Speech signals have a silent period, and only carrier tone exists in the signal, resulting in many similarities in the samples, so it is difficult for the model to distinguish the two modulation modes.

[0119] In summary, the LA-CapsNets disclosed in the present invention have a high recognition rate, strong robustness and good generalization performance.

[0120] Based on the same inventive concept, the present disclosure also provides a signal modulation identification device, as described in the following embodiments. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0121] Fig.17 A schematic diagram of a signal modulation identification device in an embodiment of the present disclosure is shown. Fig.17 As shown, the signal modulation identification device includes an acquisition module 171, an amplitude phase feature generation module 172, a time-frequency feature generation module 173, a fusion module 174 and an identification module 175. The acquisition module 171 is used to acquire an in-phase orthogonal IQ signal; the amplitude phase feature generation module 172 is used to input the IQ signal into the amplitude phase feature sub-network to obtain the amplitude phase feature, and the amplitude phase feature sub-network includes a long short-term memory network LSTM and an attention model; the time-frequency feature generation module 173 is used to input the IQ signal into the time-frequency feature sub-network to obtain the time-frequency feature, and the time-frequency feature sub-network includes a capsule network; the fusion module 174 is used to perform feature fusion on the amplitude phase feature and the time-frequency feature to obtain the fusion feature; the identification module 175 is used to perform modulation identification based on the fusion feature.

[0122] In one embodiment, the amplitude-phase feature generation module 172 is used to preprocess the IQ signal to obtain an amplitude sequence signal and a phase sequence signal before inputting the IQ signal into the amplitude-phase feature subnetwork to obtain the amplitude-phase feature; inputting the amplitude sequence signal and the phase sequence signal into the LSTM to obtain a hidden state vector; and inputting the hidden state vector into the attention model to obtain the amplitude-phase feature.

[0123] In one embodiment, the amplitude-phase feature generation module 172 is also used to encode the hidden state vector through the first weight of the attention model to obtain a coding vector; perform an attention operation on the coding vector and the attention matrix based on the dot product alignment function to obtain a tracking coding vector; perform a normalization operation on the tracking coding vector to obtain a second weight; and perform a weighted linear combination of the second weight and the hidden state vector to obtain an amplitude-phase feature.

[0124] In one embodiment, before inputting the IQ signal into the time-frequency feature subnetwork to obtain the time-frequency feature, the time-frequency feature generation module 173 is also used to convert the IQ signal into a time-frequency graph of the signal based on the time-frequency distribution CWD transformation; and input the time-frequency graph into the capsule network to obtain the time-frequency feature.

[0125] In one embodiment, the capsule network is a capsule network with multi-scale convolution and multi-path skip residual connection.

[0126] In one embodiment, the fusion module 174 is further used to input the amplitude phase feature and the time-frequency feature into a feature fusion sub-network to obtain a fused feature. The feature fusion sub-network includes a fully connected layer and an output layer, and the activation function of the output layer is a Softmax function.

[0127] In one embodiment, the identification module 175 is also used to obtain multiple signal fusion features, one signal fusion feature corresponds to one modulation type; select the target signal fusion feature with the highest matching degree with the fusion feature from the multiple signal fusion features; the modulation type corresponding to the target signal fusion feature is the modulation type of the IQ signal.

[0128] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0129] Refer to the following Fig.18 1800 according to this embodiment of the present disclosure is described. Fig.18 The electronic device 1800 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0130] like Fig.18 As shown, the electronic device 1800 is in the form of a general computing device. The components of the electronic device 1800 may include but are not limited to: at least one processing unit 1810, at least one storage unit 1820, and a bus 1830 connecting different system components (including the storage unit 1820 and the processing unit 1810).

[0131] Among them, the storage unit stores a program code, and the program code can be executed by the processing unit 1810, so that the processing unit 1810 performs the steps described in the above "Exemplary Method" section of this specification according to various exemplary embodiments of the present disclosure. For example, the processing unit 1810 can perform the following steps of the above method embodiment: obtaining an in-phase orthogonal IQ signal; inputting the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, the amplitude-phase feature subnetwork includes a long short-term memory network LSTM and an attention model; inputting the IQ signal into a time-frequency feature subnetwork to obtain a time-frequency feature, the time-frequency feature subnetwork includes a capsule network; performing feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fusion feature; performing modulation recognition based on the fusion feature.

[0132] The storage unit 1820 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 18201 and / or a cache storage unit 18202 , and may further include a read-only storage unit (ROM) 18203 .

[0133] The storage unit 1820 may also include a program / utility 18204 having a set (at least one) of program modules 18205, such program modules 18205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0134] Bus 1830 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0135] The electronic device 1800 may also communicate with one or more external devices 1840 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1800, and / or communicate with any device that enables the electronic device 1800 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1850. Furthermore, the electronic device 1800 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1860. As shown, the network adapter 1860 communicates with other modules of the electronic device 1800 via a bus 1830. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0136] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0137] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is further provided, which may be a readable signal medium or a readable storage medium, on which a program product capable of implementing the above method of the present disclosure is stored.

[0138] In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps of various exemplary implementations of the present disclosure described in the above "Specific Implementation Methods" section of this specification.

[0139] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0140] In the present disclosure, a computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0141] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0142] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0143] The embodiments of the present disclosure provide a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the signal modulation identification method provided in various optional ways in any embodiment of the present disclosure.

[0144] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0145] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0146] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0147] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The description and examples are to be regarded as exemplary only, and the true scope of the present disclosure is indicated by the appended claims.

Claims

1. A signal modulation identification method, characterized in that: include: Obtain in-phase and quadrature IQ signals; Inputting the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, wherein the amplitude-phase feature subnetwork includes a long short-term memory network LSTM and an attention model; Inputting the IQ signal into a time-frequency feature subnetwork to obtain time-frequency features, wherein the time-frequency feature subnetwork includes a capsule network; Performing feature fusion on the amplitude-phase feature and the time-frequency feature to obtain a fusion feature; performing modulation recognition based on the fused features; Among them, the step of inputting the IQ signal into the amplitude-phase feature subnetwork to obtain the amplitude-phase feature includes: inputting the amplitude sequence signal and the phase sequence signal into the LSTM to obtain a hidden state vector; encoding the hidden state vector through the first weight of the attention model to obtain a coding vector; performing an attention operation on the coding vector and the attention matrix based on a dot product alignment function to obtain a tracking coding vector; normalizing the tracking coding vector to obtain a second weight; and performing a weighted linear combination of the second weight and the hidden state vector to obtain the amplitude-phase feature.

2. The signal modulation identification method according to claim 1, characterized in that: Before inputting the IQ signal into the amplitude-phase characteristic subnetwork to obtain the amplitude-phase characteristic, the method further includes: The IQ signal is preprocessed to obtain an amplitude sequence signal and a phase sequence signal.

3. The signal modulation identification method according to claim 1, characterized in that: Before inputting the IQ signal into the time-frequency feature subnetwork to obtain the time-frequency feature, the method includes: Converting the IQ signal into a time-frequency diagram of the signal based on a time-frequency distribution CWD transformation; The step of inputting the IQ signal into a time-frequency feature subnetwork to obtain time-frequency features includes: The time-frequency graph is input into the capsule network to obtain the time-frequency feature.

4. The signal modulation identification method according to claim 3, characterized in that: The capsule network is a capsule network with multi-scale convolution and multi-path skip residual connection.

5. The signal modulation identification method according to claim 1, characterized in that: The step of fusing the amplitude and phase features with the time-frequency features to obtain fused features includes: The amplitude and phase features and the time-frequency features are input into a feature fusion subnetwork to obtain fusion features. The feature fusion subnetwork includes a fully connected layer and an output layer, and the activation function of the output layer is a Softmax function.

6. The signal modulation identification method according to claim 1, characterized in that: The performing modulation recognition based on the fusion feature includes: Acquire multiple signal fusion features, where one signal fusion feature corresponds to one modulation type; Selecting a target signal fusion feature with the highest matching degree with the fusion feature from the multiple signal fusion features; The modulation type corresponding to the target signal fusion feature is the modulation type of the IQ signal.

7. A signal modulation identification device, characterized in that: include: An acquisition module, used for acquiring in-phase and quadrature IQ signals; An amplitude-phase feature generation module, used for inputting the IQ signal into an amplitude-phase feature subnetwork to obtain an amplitude-phase feature, wherein the amplitude-phase feature subnetwork includes a long short-term memory network LSTM and an attention model; A time-frequency feature generation module, used for inputting the IQ signal into a time-frequency feature sub-network to obtain time-frequency features, wherein the time-frequency feature sub-network includes a capsule network; A fusion module, used for fusing the amplitude and phase features with the time-frequency features to obtain fusion features; A recognition module, used for performing modulation recognition based on the fusion feature; The amplitude-phase feature generation module is further used to input the amplitude sequence signal and the phase sequence signal into the LSTM to obtain a hidden state vector; encode the hidden state vector by the first weight of the attention model to obtain a coding vector; perform an attention operation on the coding vector and the attention matrix based on the dot product alignment function to obtain a tracking coding vector; perform a normalization operation on the tracking coding vector to obtain a second weight; A weighted linear combination is performed on the second weight and the hidden state vector to obtain the amplitude-phase feature.

8. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the signal modulation identification method according to any one of claims 1 to 6 by executing the executable instructions.

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 signal modulation identification method according to any one of claims 1 to 6 is implemented.

Images