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

Abstract

The invention provides a signal modulation identification method and device, electronic equipment and a storage medium, and relates to the technical field of signal modulation identification, and the method comprises the steps: carrying out the blind equalization processing of a first synchronization signal obtained after blind matching and bit synchronization processing, and obtaining a second synchronization signal; converting the second synchronization signal into a corresponding constellation diagram, and performing feature extraction on the constellation diagram by adopting a first residual network to obtain a feature vector of the constellation diagram; and adopting a classification network to perform open set identification of modulation modes based on the feature vectors to obtain a target modulation mode, the target modulation mode comprising one of an unknown modulation mode and at least one known modulation mode. Therefore, the known modulation mode can be accurately identified, and the modulation mode of the to-be-identified signal can be accurately determined to be the unknown modulation mode when the to-be-identified signal is not modulated by adopting the known modulation mode, so that a complex and changeable electromagnetic environment and diversified modulation signals can be effectively coped with.

Description

Technical Field

[0001] This disclosure relates to the field of signal modulation recognition technology, and in particular to a signal modulation recognition method, apparatus, electronic device and storage medium. Background Technology

[0002] With the continuous development and widespread application of wireless communication technology, signal modulation types are becoming increasingly numerous and complex to meet the growing demands for communication efficiency and reliability. Accurately identifying signal modulation types and improving the recognition rate is a problem that urgently needs to be solved. Summary of the Invention

[0003] This disclosure aims to at least partially address one of the technical problems in the related art.

[0004] Therefore, the purpose of this disclosure is to provide a signal modulation identification method, apparatus, electronic device and storage medium to accurately identify the modulation type of a signal and improve the identification rate of the signal modulation type.

[0005] The first aspect of this disclosure proposes a signal modulation identification method, comprising: performing blind matching and bit synchronization processing on a signal to be identified to obtain a first synchronization signal; performing blind equalization processing on the first synchronization signal to obtain a second synchronization signal; converting the second synchronization signal into a corresponding constellation diagram, and using a first residual network to extract features from the constellation diagram to obtain a feature vector of the constellation diagram; and using a classification network to perform open set identification of the modulation mode based on the feature vector to obtain a target modulation mode of the signal to be identified, wherein the target modulation mode includes one of an unknown modulation mode and at least one known modulation mode.

[0006] A second aspect of this disclosure provides a signal modulation identification device, comprising: a matching synchronization module for performing blind matching and bit synchronization processing on a signal to be identified to obtain a first synchronization signal; a blind equalization module for performing blind equalization processing on the first synchronization signal to obtain a second synchronization signal; a feature extraction module for converting the second synchronization signal into a corresponding constellation diagram and using a first residual network to extract features from the constellation diagram to obtain a feature vector of the constellation diagram; and an identification module for using a classification network to perform open set identification of the modulation mode based on the feature vector to obtain a target modulation mode of the signal to be identified, wherein the target modulation mode includes one of an unknown modulation mode and at least one known modulation mode.

[0007] A third aspect of this disclosure provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the signal modulation recognition method as described in the first aspect of this disclosure.

[0008] A fourth aspect of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the signal modulation recognition method as described in the first aspect of this disclosure.

[0009] A fifth aspect of this disclosure provides a computer program product including a computer program that, when executed by a processor, implements the signal modulation recognition method as described in the first aspect of this disclosure.

[0010] The signal modulation identification method, apparatus, electronic device, computer-readable storage medium, and computer program product proposed in this disclosure obtain a first synchronization signal by performing blind matching and bit synchronization processing on the signal to be identified, then performing blind equalization processing on the first synchronization signal to obtain a second synchronization signal, converting the second synchronization signal into a corresponding constellation diagram, and using a first residual network to extract features from the constellation diagram to obtain a feature vector of the constellation diagram. A classification network is then used to perform open-set identification of the modulation mode based on the feature vector to obtain the target modulation mode of the signal to be identified. The target modulation mode includes one of an unknown modulation mode and at least one known modulation mode. This achieves the identification of the modulation type of the signal to be identified based on constellation diagram features. Since it does not rely on complex baseband symbol sequences, it avoids the need for a large amount of prior knowledge and cumbersome signal preprocessing, and can accurately identify the modulation type of the signal, thus improving the identification rate of the signal modulation type. In addition, by combining the first residual network and the classification network, the modulation type of the signal to be identified is identified based on the open set identification method. This not only accurately identifies known modulation methods, but also accurately determines that the modulation method of the signal to be identified is unknown when the signal to be identified is not modulated by a known modulation method. This effectively copes with complex and ever-changing electromagnetic environments and diverse modulation signals.

[0011] Additional aspects and advantages of this disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description

[0012] The above and / or additional aspects and advantages of this disclosure will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, in which:

[0013] Figure 1This is a schematic diagram of the architecture of a communication system provided in an embodiment of the present disclosure;

[0014] Figure 2 This is a schematic flowchart of a signal modulation recognition method provided in an embodiment of this disclosure;

[0015] Figure 3 This is a schematic flowchart of another signal modulation recognition method provided in an embodiment of the present disclosure;

[0016] Figure 4 This is a schematic diagram of the structure of the first residual network and the classification network provided in the embodiments of this disclosure;

[0017] Figure 5 This is a schematic flowchart of another signal modulation recognition method provided in an embodiment of the present disclosure;

[0018] Figure 6 This is a schematic diagram of the structure of the second residual network provided in an embodiment of this disclosure;

[0019] Figure 7 This is a schematic diagram of the structure of a signal modulation recognition device provided in an embodiment of the present disclosure;

[0020] Figure 8 This is a block diagram illustrating an electronic device for signal modulation identification according to an exemplary embodiment. Detailed Implementation

[0021] Embodiments of this disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0022] To better understand the signal modulation identification method, apparatus, electronic device, and storage medium disclosed in the embodiments of this disclosure, the communication system to which the signal modulation identification method disclosed in the embodiments of this disclosure is applicable is described below.

[0023] Please see Figure 1 , Figure 1 This is a schematic diagram of the architecture of a communication system provided in an embodiment of this disclosure. Figure 1 The following diagram illustrates a satellite communication system as an example. This satellite communication system may include, but is not limited to, a low-Earth orbit satellite 101 and a user terminal 102. Figure 1 The number and form of the devices shown are for illustrative purposes only and do not constitute a limitation on the embodiments of this disclosure. In actual applications, it may include two or more low-orbit satellites and two or more user terminals.

[0024] The low-Earth orbit satellite 101 in this disclosure is an entity used for transmitting or receiving signals. This disclosure does not limit the specific technologies or equipment used in the low-Earth orbit satellite.

[0025] In this embodiment of the disclosure, the user terminal 102 refers to a processing device used for communication with low-Earth orbit satellites within the coverage beam range of the low-Earth orbit satellite. For example, the user terminal can be a car, smart car, ship, aircraft, ground station, mobile phone, wearable device, tablet computer, etc., equipped with satellite communication capabilities. The embodiments of this disclosure do not limit the specific technology or device form used in the user terminal.

[0026] It is understood that the satellite communication system described in this disclosure is for the purpose of more clearly illustrating the technical solutions of this disclosure, and does not constitute a limitation on the technical solutions provided in this disclosure. As those skilled in the art will know, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided in this disclosure are also applicable to similar technical problems.

[0027] In the aforementioned satellite communication system, the low-orbit satellite 101 and the user terminal 102 need to have a certain signal modulation type recognition capability to ensure that when receiving the modulation signal sent by the signal transmitter during the communication process, they can correctly perform demodulation and decoding operations to ensure smooth and reliable communication between the two parties.

[0028] Signal modulation identification methods in related technologies mainly include traditional modulation identification methods and machine learning-based modulation identification methods. Traditional modulation identification methods can be further divided into two categories: likelihood-based modulation identification methods and feature-based modulation identification methods. Likelihood-based modulation identification methods are essentially treated as multiple hypothesis testing problems. Depending on the selected model, common likelihood-based modulation identification methods include the Average Likelihood Ratio Test (ALRT), the Generalized Likelihood Ratio Test (GLRT), and the Mixed Likelihood Ratio Test (HLRT). The key to feature-based modulation identification methods lies in the selection of feature parameters and decision criteria. Commonly used feature parameters include instantaneous signal characteristics, signal statistics, wavelet transform amplitude, and peak amplitude extrema. For decision criteria, probability density-based, Euclidean distance-based, and binary decision tree-based criteria are commonly used. In contrast to traditional modulation identification methods, machine learning-based modulation identification methods mainly use neural networks to cluster important modulation parameters. The modulation identification method based on machine learning separates the received signal according to key parameters, thereby identifying the modulation mode. This provides a new approach to improve the accuracy and robustness of modulation identification.

[0029] Furthermore, constellation diagram features play a crucial role in signal modulation identification methods, particularly suitable for identifying modulation types of amplitude-modulated (AM) and phase-modulated (PM) signals. The constellation diagrams corresponding to different modulation types exhibit unique characteristics, serving as an effective basis for modulation type identification. Through constellation diagram feature matching algorithms, accurate identification of different signal modulation types can be achieved with strong noise resistance, improving the robustness of modulation type identification. However, related technologies based on constellation diagram features still have some problems, mainly stemming from their dependence on complex baseband symbol sequences. This leads to the need for a large amount of prior knowledge and cumbersome signal preprocessing, thus affecting the identification effect of modulation types. In addition, related technologies based on constellation diagram features are all closed-set identification methods, unable to effectively cope with complex and changing electromagnetic environments and diverse modulation signals. Therefore, related technologies based on constellation diagram features require further optimization and improvement.

[0030] This disclosure provides a signal modulation identification method, apparatus, electronic device, and storage medium. By performing blind matching and bit synchronization processing on the signal to be identified, a first synchronization signal is obtained. Blind equalization processing is then performed on the first synchronization signal to obtain a second synchronization signal. The second synchronization signal is converted into a corresponding constellation diagram. A first residual network is used to extract features from the constellation diagram to obtain a feature vector. A classification network is used to identify the open set of modulation modes based on the feature vector, thereby obtaining the target modulation mode of the signal to be identified. The target modulation mode includes one of an unknown modulation mode and at least one known modulation mode. This method achieves modulation type identification of the signal to be identified based on constellation diagram features. Since it does not rely on complex baseband symbol sequences, it avoids the need for a large amount of prior knowledge and cumbersome signal preprocessing, accurately identifying the modulation type of the signal and improving the signal modulation type identification rate. In addition, by combining the first residual network and the classification network, the modulation type of the signal to be identified is identified based on the open set identification method. This not only accurately identifies known modulation methods, but also accurately determines that the modulation method of the signal to be identified is unknown when the signal to be identified is not modulated by a known modulation method. This effectively copes with complex and ever-changing electromagnetic environments and diverse modulation signals.

[0031] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0032] Figure 2 This is a schematic flowchart of a signal modulation recognition method provided in an embodiment of this disclosure.

[0033] It should be noted that this method is executed by a signal modulation identification device. This signal modulation identification device can be an electronic device or can be configured within an electronic device; this disclosure does not impose any limitations on it. For example, the electronic device can be a modulation identification receiver at the signal receiving end.

[0034] like Figure 2 As shown, the method may include, but is not limited to, the following steps 201-204.

[0035] Step 201: Perform blind matching and bit synchronization processing on the signal to be identified to obtain the first synchronization signal.

[0036] The signal to be identified is the modulation signal for which modulation type identification is to be performed. This signal may be, for example, a baseband time-domain IQ (In-phase / Quadrature) signal obtained by the signal receiver after preprocessing and frequency offset elimination of the received modulation signal, or other modulation signals. This disclosure does not limit this.

[0037] Taking the baseband time-domain IQ signal after frequency offset elimination as an example, by performing blind matching and ideal symbol synchronization on the signal to be identified, the symbol-synchronized I / Q complex baseband sequence, i.e. the first synchronization signal, can be obtained.

[0038] In some embodiments, since typical wireless communication signals are waveform shaped using root-raised-cosine filters, a root-raised-cosine filter can be used for blind matching of the signal to be identified. The roll-off factor of the root-raised-cosine filter can be set as needed, for example, to 0.35.

[0039] In some embodiments, bit synchronization can be achieved using an external synchronization method. In this method, the transmitting end sends a synchronization clock signal before sending the modulation signal used for communication. The receiving end uses this synchronization clock signal to lock its own clock pulse frequency, thereby achieving bit synchronization between the transmitting and receiving ends. In some embodiments, bit synchronization can also be achieved using a self-synchronization method. In this method, the receiving end uses a special code, such as Manchester encoding, containing a first synchronization signal to extract the first synchronization signal from the received modulation signal itself to lock its own clock pulse frequency, thus achieving bit synchronization.

[0040] Step 202: Perform blind equalization on the first synchronization signal to obtain the second synchronization signal.

[0041] The second synchronization signal is the synchronization signal obtained by blindly equalizing the first synchronization signal.

[0042] It is understandable that after the signal transmitter sends the modulated signal, the ground multipath fading channel environment will affect the phase-amplitude characteristics of the modulated signal. In order to accurately identify the modulation type in the future, equalization technology is needed to solve the impact of multipath fading on the modulated signal.

[0043] Blind equalization refers to adaptive equalization of a signal using only the first synchronization signal itself, without relying on other signal sequences.

[0044] In some embodiments, the first synchronization signal is input to an equalizer to perform blind equalization on the first synchronization signal to obtain the second synchronization signal.

[0045] Step 203: Convert the second synchronization signal into the corresponding constellation diagram, and use the first residual network to extract features from the constellation diagram to obtain the feature vector of the constellation diagram.

[0046] It is understandable that since the constellation diagrams corresponding to signals with different modulation methods will be quite different, the modulation type of the signal to be identified can be determined by image recognition based on the different constellation diagrams.

[0047] In some embodiments, the constellation diagram corresponding to the second synchronization signal can be input into the first residual network, and the first residual network can be used to extract features from the constellation diagram. The output of the first residual network is the feature vector of the constellation diagram.

[0048] The constellation chart can be a grayscale constellation chart.

[0049] The first residual network is a trained residual neural network used to extract features from the constellation diagram. The structure of the first residual network can be configured as needed.

[0050] Step 204: Using a classification network, open set identification of modulation mode is performed based on feature vectors to obtain the target modulation mode of the signal to be identified. The target modulation mode includes one of an unknown modulation mode and at least one known modulation mode.

[0051] The classification network is a trained neural network used to identify signal modulation patterns. It can be any neural network with this function, and this disclosure does not impose any restrictions on it. The structure of the classification network can be set as needed.

[0052] In some embodiments, the feature vector of the constellation diagram is input into a classification network, which can then perform open-set identification of the modulation scheme based on the feature vector to obtain the target modulation scheme of the signal to be identified.

[0053] The known modulation scheme refers to the modulation scheme known to the classification network, that is, the modulation scheme that has appeared in the training samples used when training the classification network. It can include one or more of the following modulation schemes: amplitude phase modulation, such as 32Amplitude Phase Shift Keying (32APSK) and 64Amplitude Phase Shift Keying (64APSK); multiple quadrature amplitude modulation (MQAM), such as 4QAM, 16QAM, 32QAM, and 64QAM; frequency-shift keying (FSK); and pulse amplitude modulation (PAM).

[0054] Among them, the unknown modulation method is the modulation method unknown to the classification network, that is, the modulation method used when training the classification network that has never appeared in the training samples.

[0055] When a classification network identifies that a signal to be identified is not modulated using any known modulation method, it can mark the modulation method of the signal as "unknown" or "does not belong to any known modulation method." This allows the signal modulation identification device to determine that the target modulation method of the signal is an unknown modulation method. By using an open-set classification network for identification, the possibility of unknown modulation methods can be considered. When encountering a signal to be identified modulated using a modulation method unknown to the classification network, the target modulation method is not simply determined to be a known modulation method, thus avoiding incorrect modulation type identification results. This ensures accurate modulation type identification even in complex and variable electromagnetic environments and with diverse modulation signals.

[0056] In summary, the signal modulation identification method provided in this disclosure performs blind matching and bit synchronization processing on the signal to be identified to obtain a first synchronization signal; performs blind equalization processing on the first synchronization signal to obtain a second synchronization signal; converts the second synchronization signal into a corresponding constellation diagram, and uses a first residual network to extract features from the constellation diagram to obtain a feature vector of the constellation diagram; and uses a classification network to perform open-set identification of the modulation mode based on the feature vector to obtain the target modulation mode of the signal to be identified, wherein the target modulation mode includes one of an unknown modulation mode and at least one known modulation mode. Thus, it realizes the identification of the modulation type of the signal to be identified based on constellation diagram features. Since it does not rely on complex baseband symbol sequences, it avoids the need for a large amount of prior knowledge and cumbersome signal preprocessing, and can accurately identify the modulation type of the signal, improving the identification rate of the signal modulation type. In addition, by combining the first residual network and the classification network, the modulation type of the signal to be identified is identified based on the open set identification method. This not only accurately identifies known modulation methods, but also accurately determines that the modulation method of the signal to be identified is unknown when the signal to be identified is not modulated by a known modulation method. This effectively copes with complex and ever-changing electromagnetic environments and diverse modulation signals.

[0057] To clearly illustrate how the above embodiments use the first residual network to extract features from the constellation diagram and how a classification network is used for open-set recognition of modulation schemes, the following will combine... Figure 3 The signal modulation recognition method provided in the embodiments of this disclosure will be further described.

[0058] Figure 3 This is a schematic flowchart of another signal modulation recognition method provided in an embodiment of the present disclosure.

[0059] like Figure 3 As shown, the signal modulation identification method may include the following steps 301-306.

[0060] Step 301: Perform blind matching and bit synchronization processing on the signal to be identified to obtain the first synchronization signal.

[0061] Step 302: Perform blind equalization on the first synchronization signal to obtain the second synchronization signal.

[0062] The specific implementation process and principle of steps 301-302 can be found in the descriptions of other embodiments, and will not be repeated here.

[0063] Step 303: Convert the second synchronization signal into the corresponding constellation diagram, and use the first residual network to extract features from the constellation diagram to obtain the feature vector of the constellation diagram.

[0064] The feature vector represents the features of the constellation diagram.

[0065] In some embodiments, the first residual network may include multiple convolutional layers, normalization layers, activation layers, fully connected layers, and an output layer. The first residual network adds a cross-layer connection after each convolutional layer, such that the output of the input constellation diagram on the first residual network is not only a transformation of the input, but also a transformation of the sum of the input and the residual, that is, the output equals the input plus the residual.

[0066] Step 304: Input the feature vector into the first classification layer of the classification network to perform closed-set identification of modulation mode and obtain the first probability distribution vector, wherein the first probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode.

[0067] The first probability distribution vector includes multiple elements, each corresponding to a known modulation scheme, representing the probability that the target modulation scheme of the signal to be identified belongs to that known modulation scheme. The sum of all elements in the first probability distribution vector is 1.

[0068] In some embodiments, the first classification layer may include an embedding module, a classification module, and a softmax module (normalization module). The embedding module maps the feature vectors of the constellation diagram to a low-dimensional vector space, obtaining a low-dimensional feature vector of the constellation diagram. The classification module, based on the low-dimensional feature vector, obtains the confidence level of the target modulation scheme of the signal to be identified belonging to each known modulation scheme. The softmax module normalizes the output of the classification module, making the sum of the confidence levels corresponding to all known modulation schemes equal to 1, thereby converting the output of the classification module into a probability distribution, obtaining a first probability distribution vector.

[0069] Step 305: Input the feature vector into the second classification layer of the classification network to perform open set identification of the modulation mode and obtain the second probability distribution vector, where the second probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode and the probability that it belongs to an unknown modulation mode.

[0070] The second probability distribution vector includes multiple elements, each corresponding to either a known modulation scheme or an unknown modulation scheme. The element corresponding to a known modulation scheme represents the probability that the target modulation scheme of the signal to be identified belongs to that known modulation scheme, and the element corresponding to an unknown modulation scheme represents the probability that the target modulation scheme of the signal to be identified belongs to an unknown modulation scheme. Since the first probability distribution vector includes the probability corresponding to the unknown modulation scheme, the sum of all elements is not equal to 1.

[0071] In some embodiments, the second classification layer can be called the OpenMax layer, which can use the feature vector of the constellation diagram to fit the Weiber distribution to obtain the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode and the probability that it belongs to an unknown modulation mode, thereby obtaining the second probability distribution vector.

[0072] Step 306: Determine the target modulation method based on the first probability distribution vector and the second probability distribution vector.

[0073] The target modulation scheme includes one of the following: an unknown modulation scheme and at least one known modulation scheme.

[0074] In some embodiments, a target weighting coefficient can be used to weight and sum the first probability distribution vector and the second probability distribution vector to obtain a third probability distribution vector. This third probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to any known modulation mode and the probability that it belongs to an unknown modulation mode. The target modulation mode is then determined based on this third probability distribution vector. The target weighting coefficient is a weighting coefficient obtained through ensemble learning.

[0075] The third probability distribution vector includes multiple elements, each corresponding to either a known modulation scheme or an unknown modulation scheme. The element corresponding to a known modulation scheme represents the probability that the target modulation scheme of the signal to be identified belongs to that known modulation scheme, while the element corresponding to an unknown modulation scheme represents the probability that the target modulation scheme of the signal to be identified belongs to an unknown modulation scheme. Because the third probability distribution vector includes probabilities corresponding to unknown modulation schemes, the sum of all elements is not equal to 1.

[0076] Among them, the modulation scheme corresponding to the largest element in the third probability distribution vector can be determined as the target modulation scheme.

[0077] In some embodiments, the first residual network and the classification network can be obtained through joint training, and the target weight coefficients can be learned during the training of the first residual network and the classification network. Correspondingly, the signal modulation recognition method may further include:

[0078] Multiple training samples are obtained, wherein each training sample includes a sample constellation diagram corresponding to the sample modulation signal, wherein each sample constellation diagram is labeled with the sample modulation method of the corresponding sample modulation signal, and the sample modulation method includes at least one of the known modulation methods;

[0079] Obtain the initial first residual network, the initial classification network, and the initial weight coefficients;

[0080] The initial first residual network and the initial classification network are trained based on multiple training samples. During the training process, the model parameters of the initial first residual network, the model parameters of the initial classification network, and the initial weight coefficients are adjusted to obtain the trained first residual network, classification network, and target weight coefficients.

[0081] Here, the initial first residual network is an untrained first residual network. The initial classification network is also an untrained classification network. The initial weight coefficients are the initial values ​​of the target weight coefficients.

[0082] Among them, the sample modulation signal is the modulation signal used as the training sample, and its modulation method can be called the sample modulation method, which is one of the known modulation methods.

[0083] In some embodiments, the process of training the initial first residual network and the initial classification network based on multiple training samples can be as follows: Inputting the sample constellation diagram corresponding to the sample modulation signal into the initial first residual network to extract features from the sample constellation diagram using the initial first residual network, obtaining a feature vector of the sample constellation diagram; inputting the feature vector of the sample constellation diagram into the initial classification network to perform closed-set identification of the modulation mode using the first classification layer of the initial classification network, obtaining a probability distribution vector (which can be called the first sample probability distribution vector) where the sum of the elements corresponding to the sample modulation signal is 1, where the probability distribution vector represents the probability that the modulation mode of the sample modulation signal belongs to each known modulation mode; and, the second classification layer of the initial classification network can be used to perform open-set identification of the modulation mode, obtaining a probability distribution vector (which can be called the second sample probability distribution vector) where the sum of the elements corresponding to the sample modulation signal is not 1. The probability distribution vector represents the probability that the modulation mode of the sample modulation signal belongs to each known modulation mode and the probability that it belongs to an unknown modulation mode. Then, using initial weight coefficients, the two probability distribution vectors are weighted and summed to obtain a probability distribution vector whose sum is not 1 (which can be called the third sample probability distribution vector). This probability distribution vector represents the probability that the modulation mode of the sample modulation signal belongs to each known modulation mode and the probability that it belongs to an unknown modulation mode. Based on this probability distribution vector (i.e., the third sample probability distribution vector), the modulation mode of the sample modulation signal can be predicted. Based on the difference between the predicted modulation mode and the sample modulation mode, the initial first residual network, the initial classification network, and the initial weight coefficients are adjusted. By iteratively training with multiple training samples, the trained first residual network, the trained classification network, and the target weight coefficients can be obtained.

[0084] The loss function used in the training process of the first residual network and the classification network can be the mean squared error function or other loss functions, and this disclosure does not impose any restrictions on it.

[0085] The optimizer used during the training of the first residual network and the classification network can be the Adam optimizer, with parameters set to an initial learner coefficient of 0.01 and a learning rate decay of 0.99 every 10 epochs (cycles or rounds). The batch size for the first stage of each neural network is 1024, the batch size for the second stage is 64, and the maximum number of iterations is set to 200.

[0086] The following is for reference. Figure 4 Taking the sample constellation diagram corresponding to the sample modulation signal as an example, the first residual network and classification network in this embodiment of the present disclosure are illustrated. It should be noted that... Figure 4 For the first residual network, only a portion of its convolutional layers are shown.

[0087] refer to Figure 4 The first residual network can include 16 convolutional layers. The input during training is a grayscale constellation image with dimensions (128, 1, 1024, 1024). Here, 128 represents the number of training samples; 1 represents the number of channels, which are the grayscale values ​​of the constellation image; the first 1024 represents the width of the input grayscale constellation image; and the second 1024 represents the length of the input grayscale constellation image. The first convolutional layer has a kernel set to (3, 3), 1 input channel, 64 output channels, and a stride of 1. The second and third convolutional layers also have kernels set to (3, 3), 64 input and output channels, allow padding, and have a stride of 1. The fourth convolutional layer incorporates the output of the first convolutional layer's input data. The fourth convolutional layer has a kernel set to (1, 1), and its input and output channels are the same as those of the first convolutional layer. In this process, BatchNorm2d (two-dimensional batch normalization) and ReLU layers (activation layers) are applied between adjacent convolutional layers for batch normalization, and the output vector is (128, 64, 1024, 1024). Figure 4(Not shown in the diagram), where 128 represents the number of training samples; 64 represents the number of channels; the first 1024 represents the width of the output data of this layer; the second 1024 represents the length of the output data of this layer. Next, the input channels of the fifth layer are the 64-channel output of the third layer plus the 64-channel output of the original input from a single convolutional layer, and the number of output channels of the fifth convolutional layer is set to twice the number of input channels. The settings for the sixth and seventh layers are the same as the second and third layers. The input of the fifth convolutional layer is added to the eighth layer, and the number of output channels in the ninth convolutional layer is doubled, increasing the feature dimension of the channel mapping, so that the output of the ninth layer is twice the number of output channels of the eighth layer. This continues until the output dimension of the 16th layer is 256. The convolutional stride of layers 6, 10, and 14 is 2, and the output is a feature vector with dimensions (128, 256, 128, 128), where 128 represents the number of training samples, 256 represents the number of channels, the first 128 represents the width of the output feature vector, and the second 128 represents the length of the output feature vector.

[0088] After performing average pooling and linear mapping operations on the output of the convolutional layer, the feature vectors of the constellation diagrams corresponding to each sample modulation signal modulated by each known modulation method can be obtained.

[0089] After the feature vectors of the constellation diagrams corresponding to each sample modulation signal pass through the first and second classification layers of the classification network, we can obtain the probability distribution vectors (i.e., the first sample probability distribution vectors) with a sum of 1 for each sample modulation signal and the probability distribution vectors (i.e., the second sample probability distribution vectors) with a sum of non-1 for each sample modulation signal. Then, through the integration layer, we can predict the modulation mode of each sample modulation signal.

[0090] By adjusting the initial weight coefficients in the first residual network, classification network, and ensemble layer based on the difference between the predicted modulation scheme and the corresponding sample modulation scheme, and by iteratively training with multiple training samples, the trained first residual network, trained classification network, and target weight coefficients can be obtained. These trained first residual network, trained classification network, and target weight coefficients can then be used to identify the modulation type of the signal to be identified.

[0091] In summary, the signal modulation identification method provided in this disclosure identifies the modulation type of the signal to be identified based on constellation diagram features. Since it does not rely on complex baseband symbol sequences, it avoids the need for extensive prior knowledge and cumbersome signal preprocessing, accurately identifying the modulation type of the signal and improving the identification rate. Furthermore, by combining a first residual network and a classification network, and using an open-set identification method to identify the modulation type of the signal to be identified, it can not only accurately identify known modulation methods but also accurately determine the modulation method of the signal to be identified as unknown when the signal is not modulated by a known modulation method. This effectively addresses complex and variable electromagnetic environments and diverse modulation signals.

[0092] To clearly illustrate how the above embodiments acquire the first synchronization signal and how blind equalization processing is performed on the first synchronization signal, the following will combine... Figure 5 The signal modulation recognition method provided in the embodiments of this disclosure will be further described.

[0093] Figure 5 This is a schematic flowchart of another signal modulation recognition method provided in an embodiment of the present disclosure.

[0094] like Figure 5 As shown, the signal modulation identification method may include the following steps 501-507.

[0095] Step 501: Preprocess the acquired input signal to obtain the preprocessed signal.

[0096] In some embodiments, the signal transmitter can be a satellite, and the signal transmitted by the satellite can be a single-carrier signal with different modulation methods. The signal receiver can use an analog-to-digital converter (ADC) module to perform operations such as acquisition and quantization on the single-carrier signal to obtain the input signal. The input signal can be a radio frequency digital signal.

[0097] In some embodiments, the preprocessed signal is a baseband time-domain IQ (In-phase / Quadrature) signal, and step 501 can be implemented in the following way:

[0098] The input signal is down-converted to baseband to obtain the baseband signal.

[0099] Perform a Fourier transform on the baseband signal to obtain the frequency domain data of the input signal;

[0100] The center frequency and bandwidth of the input signal are estimated based on the frequency domain data. The center frequency and bandwidth of the input signal are obtained. Based on the obtained center frequency and bandwidth, the input signal is processed by downconversion, filtering and downsampling, and finally the baseband time domain IQ signal of the input signal is obtained.

[0101] Among them, the baseband time-domain IQ signal r(nT) s It can be expressed in the form shown in formula (1) below:

[0102]

[0103] Among them, a k It is the k-th symbol sent; E is the energy of the sent symbol; g T This is the impulse response of the raised cosine roll-off filter; T s It is the sampling period; T b ε is the symbol period; ε is the normalized symbol timing error, which can range from -0.5 to ε to 0.5. It is the initial phase of the carrier; w(nT) s ) is Gaussian white noise with a mean of 0 and a variance of N0; N represents the number of sampling points, and n represents the nth sampling point.

[0104] Step 502: Perform frequency offset estimation on the preprocessed signal to obtain the frequency offset estimate.

[0105] In some embodiments, the frequency offset of the preprocessed signal can be estimated based on the Welch power spectrum. The process of estimating the frequency offset of the preprocessed signal based on the Welch power spectrum can be as follows:

[0106] Assume the preprocessed signal is {x} n The length of the signal, i.e., the number of sampling points, is N. This preprocessed signal is divided into K subsequences of length L, and a window is applied to each subsequence. The window function is denoted as d(n). Adjacent subsequences may have overlapping data.

[0107] The frequency offset of the preprocessed signal can be estimated using the following formula (2):

[0108]

[0109] Where df represents the frequency offset estimate; The signal represents the center frequency of the preprocessed signal; n represents the nth sampling point. Since the spectrum is generally symmetrical, only half the number of sampling points, i.e., N / 2, needs to be used.

[0110] Taking d(n) as a rectangular window as an example, the Welch power spectrum of the preprocessed signal can be obtained using the following formula (3).

[0111] in Represents the i-th subsequence; e -jwn Let be a complex exponential function, where e is the base of the natural logarithm, j is the imaginary unit, and w is the angular frequency.

[0112] The calculated Welch power spectrum is used as the spectral data |x in formula (2). n By substituting | into formula (2), the frequency offset estimate can be obtained.

[0113] Step 503: Perform frequency offset elimination processing on the preprocessed signal based on the frequency offset estimate to obtain the signal to be identified.

[0114] Step 504: Perform blind matching and bit synchronization processing on the signal to be identified to obtain the first synchronization signal.

[0115] In some embodiments, step 504 can be implemented as follows: performing blind matching on the signal to be identified to obtain a matched signal; estimating the timing error of the matched signal, and performing interpolation filtering on the non-optimally sampled data points in the matched signal based on the obtained timing error, so that the non-optimally sampled data points approximate the optimal sampling points, thereby obtaining a first synchronization signal.

[0116] In some embodiments, since typical wireless communication signals are waveform shaped using root-raised cosine filters, a root-raised cosine filter can be used to perform blind matching on the frequency offset-canceled signal to be identified, resulting in a matched signal. The roll-off factor of the root-raised cosine filter can be set as needed, for example, to 0.35.

[0117] In some embodiments, a cubic interpolation filter can be used to interpolate and filter data points that are not optimally sampled. The form of the cubic interpolation filter can be as shown in the following formula (4):

[0118] y(t)=b -2 (u)x[(m+2)T s ]+b -1 (u)x[(m+1)T s ]+b0(u)x[mT s ]+b1(u)x[(m-1)T s (4)

[0119] The coefficients of the cubic interpolation filter are:

[0120]

[0121] Where y(t) is the first synchronization signal; u is the fractional interval parameter in the interpolation filtering algorithm, which is calculated from the estimated timing error; x[mT s [The signal after matching is in mT] s The sampled value at time; x[(m+2)T s [The signal after matching is in mT] s The previous two moments, i.e. (m+2)T s The sampled value at time; x[(m+1)T s [The signal after matching is in mT] s The previous time point is (m+1)T s The sampled value at time; x[(m-1)T s [The signal after matching is in mT] s The sampled value at the next time step, i.e., time (m-1)T; b -2 For mT s The cubic interpolation filter coefficients at the previous two time points, b -1 For mT s The cubic interpolation filter coefficients from the previous time step; b0 represents the current time step mT. s The cubic interpolation filter coefficients; b1 is mT s The cubic interpolation filter coefficients at the next time step. Using the cubic interpolation filter described above, the optimal sampling point for a symbol can be estimated based on the four sampling points. The result obtained from the cubic interpolation filter is then compensated into the matched signal via a loop to obtain a first synchronization signal that approximates the optimal sampling point.

[0122] Step 505: Input the first synchronization signal into the second residual network to perform blind equalization on the first synchronization signal using the second residual network to obtain the second synchronization signal.

[0123] The second residual network is a trained residual neural network used for blind equalization. The structure of the second residual network can be configured as needed.

[0124] In some embodiments, the second residual network may include convolutional layers, normalization layers, activation layers, fully connected layers, and an output layer, and the second residual network adds a cross-layer connection after each convolutional layer, such that the output of the input first synchronization signal on the second residual network is not only a transformation of the input, but also a transformation of the sum of the input and the residual, that is, the output is equal to the input plus the residual.

[0125] It is understandable that, when the preprocessed signal is a baseband time-domain IQ signal, the first synchronization signal is a baseband time-domain IQ signal. This signal is usually a complex signal, so it can be divided into two dimensions: real part and imaginary part. These two parts are then input into the second residual network and pass through each network layer of the second residual network in sequence to obtain the second synchronization signal.

[0126] In some embodiments, before performing blind equalization processing on the first synchronization signal using the second residual network, the second residual network may also be trained. Accordingly, before step 505, the following may be included:

[0127] Multiple training sample pairs are obtained, wherein each training sample pair includes a reference signal before the multipath fading channel and the corresponding multipath fading signal after the multipath fading channel.

[0128] The multipath fading signal is input into the initial second residual network to perform blind equalization on the multipath fading signal, and the predicted signal after blind equalization is obtained.

[0129] Based on the difference between the predicted signal and the corresponding reference signal, the model parameters of the initial second residual network are adjusted to obtain the trained second residual network.

[0130] The initial second residual network is an untrained second residual network. The signal before the multipath fading channel can be called the reference signal; the reference signal after passing through the multipath fading channel can be called the multipath fading signal.

[0131] The loss function used in the training of the second residual network can be the mean squared error function or other loss functions, and this disclosure does not impose any restrictions on this. The optimizer used in the training of the second residual network can be the Adam optimizer.

[0132] The following is for reference. Figure 6 Taking the signal in the training sample pair after passing through a multipath fading channel as an example, which is the multipath fading IQ signal, the second residual network in this embodiment of the present disclosure will be illustrated. It should be noted that... Figure 6 For the second residual network, only a portion of its network layers are shown.

[0133] refer to Figure 6The second residual network can include at least four convolutional layers, two normalization layers, two activation layers, and one fully connected layer. During training, the input to the second residual network is a multipath fading IQ signal with dimensions (128, 1, 2, 1024), where 128 represents the number of training sample pairs, 1 represents the number of channels, 2 represents the real and imaginary parts of the IQ data (i.e., the multipath fading IQ signal), and 1024 represents the length of the input data. After processing by the convolutional, normalization, and activation layers, the output vector has dimensions (128, 4, 2, 1024), where 128 represents the number of training sample pairs, 4 represents the number of channels, 2 represents the real and imaginary parts of the IQ data, and 1024 represents the length of the output data. After the output vector passes through the last convolutional layer and the fully connected linear mapping layer, the output is the equalized IQ signal with dimensions (128, 1, 2, 1024). This IQ signal is the predicted signal after blind equalization. Here, 128 represents the number of training sample pairs, 1 represents the number of channels, 2 represents the real and imaginary dimensions of the output IQ data, and 1024 represents the length of the output data.

[0134] By adjusting the model parameters of the initial second residual network based on the difference between the predicted signal and the corresponding reference signal, and iteratively training it using multiple training sample pairs, a trained second residual network can be obtained. This trained second residual network can then be used to perform blind equalization on the first synchronization signal corresponding to the signal to be identified.

[0135] Step 506: Convert the second synchronization signal into the corresponding constellation diagram, and use the first residual network to extract features from the constellation diagram to obtain the feature vector of the constellation diagram.

[0136] Step 507: Using a classification network, open set identification of modulation mode is performed based on feature vectors to obtain the target modulation mode of the signal to be identified. The target modulation mode includes one of an unknown modulation mode and at least one known modulation mode.

[0137] The signal modulation identification method of this disclosure identifies the modulation type of the signal to be identified based on constellation diagram features. Since it does not rely on complex baseband symbol sequences, it avoids the need for extensive prior knowledge and cumbersome signal preprocessing, accurately identifying the modulation type and improving the identification rate. Furthermore, by combining a first residual network and a classification network, and using an open-set identification method to identify the modulation type of the signal to be identified, it can not only accurately identify known modulation schemes but also accurately determine the modulation scheme of the signal to be identified as unknown when the signal is not modulated by a known scheme. This effectively addresses complex and variable electromagnetic environments and diverse modulation signals. Moreover, by using a second residual network to perform blind equalization on the first synchronization signal, adaptive equalization can be achieved using only the first synchronization signal itself without relying on other signal sequences, mitigating the impact of ground multipath fading channel environments on the modulated signal and further improving the accuracy of signal modulation type identification.

[0138] To achieve the above embodiments, this disclosure also proposes a signal modulation recognition device.

[0139] Figure 7 This is a schematic diagram of a signal modulation identification device provided in an embodiment of the present disclosure. The signal modulation identification device can execute the signal modulation identification method described in the foregoing embodiments. The signal modulation identification device can be an electronic device or can be configured within an electronic device; this disclosure does not impose any limitations on this.

[0140] like Figure 7 As shown, the signal modulation recognition device 700 includes: a matching synchronization module 710, a blind equalization module 720, a feature extraction module 730, and a recognition module 740.

[0141] Among them, the matching synchronization module 710 is used to perform blind matching and bit synchronization processing on the signal to be identified to obtain the first synchronization signal;

[0142] Blind equalization module 720 is used to perform blind equalization processing on the first synchronization signal to obtain the second synchronization signal;

[0143] The feature extraction module 730 is used to convert the second synchronization signal into a corresponding constellation diagram and use the first residual network to extract features from the constellation diagram to obtain the feature vector of the constellation diagram.

[0144] The identification module 740 is used to perform open-set identification of modulation mode based on feature vectors using a classification network to obtain the target modulation mode of the signal to be identified, wherein the target modulation mode includes one of an unknown modulation mode and at least one known modulation mode.

[0145] As one possible implementation of this disclosure, the identification module 740 includes:

[0146] The first identification unit is used to input the feature vector into the first classification layer of the classification network to perform closed-set identification of the modulation mode and obtain the first probability distribution vector, wherein the first probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode;

[0147] The second identification unit is used to input the feature vector into the second classification layer of the classification network to perform open set identification of the modulation mode and obtain the second probability distribution vector, wherein the second probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode and the probability that it belongs to an unknown modulation mode.

[0148] The determining unit is used to determine the target modulation mode based on the first probability distribution vector and the second probability distribution vector.

[0149] As one possible implementation of this disclosure, the determining unit is configured to:

[0150] Using target weighting coefficients, the first probability distribution vector and the second probability distribution vector are weighted and summed to obtain the third probability distribution vector, where the third probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode and the probability that it belongs to an unknown modulation mode.

[0151] The target modulation scheme is determined based on the third probability distribution vector.

[0152] As one possible implementation of this disclosure, the signal modulation identification device 700 further includes:

[0153] The first acquisition module is used to acquire multiple training samples, wherein each training sample includes a sample constellation diagram corresponding to the sample modulation signal, wherein each sample constellation diagram is labeled with the sample modulation method of the corresponding sample modulation signal, and the sample modulation method includes at least one of the known modulation methods;

[0154] The second acquisition module is used to acquire the initial first residual network, the initial classification network, and the initial weight coefficients;

[0155] The first training module is used to train the initial first residual network and the initial classification network based on multiple training samples. During the training process, the model parameters of the initial first residual network, the model parameters of the initial classification network, and the initial weight coefficients are adjusted to obtain the trained first residual network, the classification network, and the target weight coefficients.

[0156] As one possible implementation of this disclosure, the blind equalization module 720 is used for:

[0157] The first synchronization signal is input into the second residual network, and the first synchronization signal is blindly equalized using the second residual network to obtain the second synchronization signal.

[0158] As one possible implementation of this disclosure, the signal modulation identification device 700 further includes:

[0159] The third acquisition module is used to acquire multiple training sample pairs, wherein each training sample pair includes a reference signal before the multipath fading channel and the corresponding multipath fading signal after the multipath fading channel.

[0160] The second training module is used to input the multipath fading signal into the initial second residual network, so as to perform blind equalization processing on the multipath fading signal using the initial second residual network to obtain the blindly equalized prediction signal; and to adjust the model parameters of the initial second residual network according to the difference between the prediction signal and the corresponding reference signal to obtain the trained second residual network.

[0161] As one possible implementation of this disclosure, the matching synchronization module 710 is used for:

[0162] Blind matching is performed on the signal to be identified to obtain the matched signal;

[0163] The timing error of the matched signal is estimated, and the non-optimally sampled data points in the matched signal are interpolated and filtered based on the obtained timing error so that the non-optimally sampled data points are close to the optimal sampling points, thus obtaining the first synchronization signal.

[0164] As one possible implementation of this disclosure, the signal modulation identification device 700 further includes:

[0165] The preprocessing module is used to preprocess the acquired input signal to obtain the preprocessed signal;

[0166] The frequency offset estimation module is used to estimate the frequency offset of the preprocessed signal to obtain the frequency offset estimate value;

[0167] The frequency offset cancellation module is used to perform frequency offset cancellation processing on the preprocessed signal based on the frequency offset estimate to obtain the signal to be identified.

[0168] The signal modulation identification device of this disclosure performs blind matching and bit synchronization processing on the signal to be identified to obtain a first synchronization signal; performs blind equalization processing on the first synchronization signal to obtain a second synchronization signal; converts the second synchronization signal into a corresponding constellation diagram, and uses a first residual network to extract features from the constellation diagram to obtain a feature vector of the constellation diagram; and uses a classification network to perform open-set identification of the modulation mode based on the feature vector to obtain the target modulation mode of the signal to be identified, wherein the target modulation mode includes one of an unknown modulation mode and at least one known modulation mode. Thus, it realizes the identification of the modulation type of the signal to be identified based on constellation diagram features. Since it does not rely on complex baseband symbol sequences, it avoids the need for a large amount of prior knowledge and cumbersome signal preprocessing, and can accurately identify the modulation type of the signal, improving the identification rate of the signal modulation type. Furthermore, by combining the first residual network and the classification network to identify the modulation type of the signal to be identified based on the open-set identification method, it can not only accurately identify known modulation modes, but also accurately determine that the modulation mode of the signal to be identified is an unknown modulation mode when the signal to be identified is not modulated by a known modulation mode, thereby effectively coping with complex and changing electromagnetic environments and diverse modulation signals.

[0169] It should be noted that the foregoing explanation of the signal modulation recognition method embodiment also applies to the signal modulation recognition device of this embodiment, and will not be repeated here.

[0170] To implement the above embodiments, this disclosure also proposes an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the signal modulation recognition method as proposed in the foregoing embodiments of this disclosure.

[0171] To implement the above embodiments, this disclosure also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the signal modulation recognition method as proposed in the foregoing embodiments of this disclosure.

[0172] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program that, when executed by a processor, implements the signal modulation recognition method as proposed in the foregoing embodiments of this disclosure.

[0173] Figure 8 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present disclosure is shown. Figure 8 The electronic device 12 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0174] like Figure 8 As shown, the electronic device 12 is represented in the form of a general-purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors 16, a memory 28, and a bus 18 connecting different system components (including the memory 28 and the processor 16). The processor 16 can be any component capable of data processing.

[0175] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0176] Electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0177] Memory 28 may include computer system readable media in the form of volatile memory, such as Random Access Memory (RAM) 30 and / or cache memory 32. Electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (… Figure 8 Not shown; usually referred to as a "hard drive".

[0178] although Figure 8Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disc drive for reading and writing to a removable non-volatile optical disc (e.g., a compact disc read-only memory (CD-ROM), a digital video disc read-only memory (DVD-ROM), or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this disclosure.

[0179] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of this disclosure.

[0180] Electronic device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with electronic device 12, and / or with any device that enables electronic device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 22. Furthermore, electronic device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. Figure 8 As shown, network adapter 20 communicates with other modules of electronic device 12 via bus 18. It should be understood that, although... Figure 8 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0181] The processor 16 executes various functional applications and data processing by running programs stored in the memory 28, such as implementing the signal modulation recognition method mentioned in the foregoing embodiments.

[0182] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise expressly and specifically defined.

[0183] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0184] Although embodiments of the present disclosure have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present disclosure.

Claims

1. A signal modulation recognition method, characterized in that, The method includes: Blind matching and bit synchronization processing are performed on the signal to be identified to obtain the first synchronization signal; Blind equalization is performed on the first synchronization signal to obtain the second synchronization signal; The second synchronization signal is converted into a corresponding constellation diagram, and the first residual network is used to extract features from the constellation diagram to obtain the feature vector of the constellation diagram. A classification network is used to identify the modulation scheme by open set based on the feature vector, thereby obtaining the target modulation scheme of the signal to be identified. The target modulation scheme includes one of an unknown modulation scheme and at least one known modulation scheme.

2. The method according to claim 1, characterized in that, The step of employing a classification network to perform open-set identification of the modulation scheme based on the feature vector to obtain the target modulation scheme of the signal to be identified includes: The feature vector is input into the first classification layer of the classification network to perform closed-set identification of modulation mode and obtain a first probability distribution vector, wherein the first probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode; The feature vector is input into the second classification layer of the classification network to perform open set identification of the modulation mode, and a second probability distribution vector is obtained, wherein the second probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode and the probability that it belongs to the unknown modulation mode. The target modulation scheme is determined based on the first probability distribution vector and the second probability distribution vector.

3. The method according to claim 2, characterized in that, The step of obtaining the target modulation scheme based on the first probability distribution vector and the second probability distribution vector includes: Using target weighting coefficients, the first probability distribution vector and the second probability distribution vector are weighted and summed to obtain a third probability distribution vector, wherein the third probability distribution vector represents the probability that the target modulation mode of the signal to be identified belongs to each known modulation mode and the probability that it belongs to the unknown modulation mode; The target modulation scheme is determined based on the third probability distribution vector.

4. The method according to claim 3, characterized in that, The method further includes: Multiple training samples are obtained, wherein each training sample includes a sample constellation diagram corresponding to the sample modulation signal, wherein each sample constellation diagram is labeled with the sample modulation method of the corresponding sample modulation signal, and the sample modulation method includes one of the at least one known modulation method; Obtain the initial first residual network, the initial classification network, and the initial weight coefficients; The initial first residual network and the initial classification network are trained based on the multiple training samples. During the training process, the model parameters of the initial first residual network, the model parameters of the initial classification network, and the initial weight coefficients are adjusted to obtain the trained first residual network, the classification network, and the target weight coefficients.

5. The method according to claim 1, characterized in that, The step of performing blind equalization on the first synchronization signal to obtain the second synchronization signal includes: The first synchronization signal is input into the second residual network to perform blind equalization on the first synchronization signal, thereby obtaining the second synchronization signal.

6. The method according to claim 5, characterized in that, Before inputting the first synchronization signal into the second residual network to perform blind equalization on the first synchronization signal using the second residual network to obtain the second synchronization signal, the method further includes: Multiple training sample pairs are obtained, wherein each training sample pair includes a reference signal before the multipath fading channel and the corresponding multipath fading signal after the multipath fading channel. The multipath fading signal is input into an initial second residual network to perform blind equalization on the multipath fading signal, thereby obtaining a predicted signal after blind equalization. Based on the difference between the predicted signal and the corresponding reference signal, the model parameters of the initial second residual network are adjusted to obtain the trained second residual network.

7. The method according to any one of claims 1-6, characterized in that, The process of blind matching and bit synchronization of the signal to be identified to obtain a first synchronization signal includes: Blind matching is performed on the signal to be identified to obtain the matched signal; The timing error is estimated for the matched signal, and the non-optimally sampled data points in the matched signal are interpolated and filtered based on the obtained timing error to make the non-optimally sampled data points approximate the optimal sampling points, thereby obtaining the first synchronization signal.

8. The method according to any one of claims 1-6, characterized in that, Before performing blind matching and bit synchronization processing on the signal to be identified to obtain the first synchronization signal, the process may further include: The acquired input signal is preprocessed to obtain the preprocessed signal; Frequency offset estimation is performed on the preprocessed signal to obtain the frequency offset estimate value; The preprocessed signal is subjected to frequency offset elimination processing based on the frequency offset estimate to obtain the signal to be identified.

9. A signal modulation identification device, characterized in that, The device includes: The matching and synchronization module is used to perform blind matching and bit synchronization processing on the signal to be identified to obtain the first synchronization signal. A blind equalization module is used to perform blind equalization processing on the first synchronization signal to obtain a second synchronization signal; The feature extraction module is used to convert the second synchronization signal into a corresponding constellation diagram and use a first residual network to extract features from the constellation diagram to obtain the feature vector of the constellation diagram. The identification module is used to perform open-set identification of the modulation mode based on the feature vector using a classification network to obtain the target modulation mode of the signal to be identified, wherein the target modulation mode includes one of an unknown modulation mode and at least one known modulation mode.

10. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; characterized in that, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1-8.

Images