A signal modulation format identification method, system, device and medium

Abstract

The application discloses a signal modulation format identification method, system, device and medium, and belongs to the field of optical fiber communication. The multi-feature input hybrid neural network constructed by the method comprises a multi-scale convolutional neural network and a deep neural network. After constellation diagram features and direction gradient histogram features of a signal are extracted, the two input features of the multi-feature input hybrid neural network are input into the multi-scale convolutional neural network and the deep neural network respectively, so that the modulation format of the signal is identified. By utilizing different types of features and different levels of representation, the application improves the traditional single modulation format identification method based on constellation diagram feature extraction, and realizes accurate identification of the modulation formats of probability shaping signals and uniform shaping signals in an optical fiber communication system.

Description

Technical Field

[0001] This invention relates to the field of optical fiber communication, and in particular to a method, system, device and medium for identifying signal modulation formats. Background Technology

[0002] In fiber optic communication systems, probabilistic shaping (PS) technology has emerged to meet ever-increasing data demands and improve transmission performance, as it can significantly enhance the performance of fiber optic communication systems. PS technology optimizes data transmission by dynamically adjusting the probability distribution of signal constellation points, especially under high-order modulation formats, which is crucial for improving system transmission capacity and spectral efficiency. However, due to the changing probability distribution of constellation points, the receiver needs to accurately identify the modulation format of the transmitted signal for effective demodulation. Signal modulation format identification is essential to ensuring the correct transmission of data in fiber optic communication systems. Different modulation formats require different demodulation methods; therefore, an intelligent mechanism is needed to determine the specific modulation format used by the signal.

[0003] Traditional signal modulation format identification methods are primarily designed for standard modulation formats. However, their accuracy may be affected by probabilistic constellation shaping techniques. This is because the signal modulation format after probabilistic shaping may resemble other uniformly shaped (US) signal modulation formats in certain situations, especially when channel quality varies. Therefore, a more efficient signal modulation format identification method is needed. Summary of the Invention

[0004] The purpose of this invention is to provide a signal modulation format identification method, system, device and medium that can accurately identify the modulation format of probabilistic shaped signals and uniformly shaped signals in optical fiber communication systems.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A signal modulation format identification method, comprising:

[0007] Construct a multi-feature input hybrid neural network; the multi-feature input hybrid neural network includes a multi-scale convolutional neural network, a deep neural network, a fusion layer and a fully connected layer, the output of the multi-scale convolutional neural network and the output of the deep neural network are both connected to the input of the fusion layer, and the output of the fusion layer is connected to the input of the fully connected layer;

[0008] Different types of signals in an optical fiber communication system are acquired, and the constellation diagram features and directional gradient histogram features of each signal are extracted; among them, the different types of signals include probabilistically shaped signals and uniformly shaped signals;

[0009] The constellation diagram features of each signal are input into the multi-scale convolutional neural network, and the orientation gradient histogram features of each signal are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain a modulation format recognition model.

[0010] After extracting constellation diagram features and directional gradient histogram features from the signal to be identified, they are input into the modulation format recognition model, and the modulation format of the signal to be identified is output.

[0011] Optionally, constellation diagram features and histogram features of directional gradients are extracted for each signal, specifically including:

[0012] The collected signals of different types are normalized and then converted into constellation diagrams;

[0013] The constellation diagram of each signal is converted to grayscale to obtain the grayscale constellation diagram of each signal;

[0014] All grayscale constellation images are combined into a constellation image set, and the constellation image set is divided into a training set and a test set in a 4:1 ratio.

[0015] Binarization and downsampling are performed on each grayscale constellation image to obtain the constellation image features of each signal;

[0016] Gradient calculation is performed on each grayscale constellation image to obtain the gradient information of each grayscale constellation image; wherein, the gradient information includes the gradient magnitude and gradient direction;

[0017] The grayscale constellation image is divided into multiple feature blocks, each feature block is divided into multiple feature units of equal size, and the gradient direction is divided into several intervals.

[0018] The gradient magnitude of each feature unit is accumulated according to the interval corresponding to the gradient direction to form the directional gradient histogram feature of each feature unit.

[0019] The directional gradient histogram features of all feature units in each feature block are concatenated and normalized to obtain the directional gradient histogram features of each feature block.

[0020] The directional gradient histogram features of all feature blocks are concatenated to form the directional gradient histogram features of each signal.

[0021] Optionally, gradient calculation is performed on each grayscale constellation image to obtain the gradient information of each grayscale constellation image, specifically including:

[0022] According to the formula Gradient calculation is performed on the grayscale constellation image to obtain the gradient in the horizontal direction and the gradient in the vertical direction; where U i(i,j) represents the gradient of pixel (i,j) in the horizontal direction, U j (i,j) represents the gradient of pixel (i,j) in the vertical direction, H(i+1,j) represents the pixel value of pixel (i+1,j), H(i-1,j) represents the pixel value of pixel (i-1,j), H(i,j+1) represents the pixel value of pixel (i,j+1), and H(i,j-1) represents the pixel value of pixel (i,j-1).

[0023] Based on the gradients of the grayscale constellation image in the horizontal and vertical directions, using the formula... Calculate the gradient magnitude and gradient direction of each pixel in the grayscale constellation image; where G(i,j) is the gradient magnitude of pixel (i,j) and θ(i,j) is the gradient direction of pixel (i,j).

[0024] Optionally, the constellation diagram features of each signal are input into the multi-scale convolutional neural network, and the histogram features of the directional gradients of each signal are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain a modulation format recognition model, specifically including:

[0025] The constellation graph features in the training set are input into the multi-scale convolutional neural network, and the directional gradient histogram features in the training set are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain the trained multi-feature input hybrid neural network.

[0026] Based on the constellation diagram features and directional gradient histogram features in the test set, the performance of the trained multi-feature input hybrid neural network is tested, and the trained multi-feature input hybrid neural network that passes the test is determined as the modulation format recognition model.

[0027] Optionally, the multi-scale convolutional neural network includes: three parallel convolutional neural networks and an average pooling layer;

[0028] Each convolutional neural network includes a convolutional layer, a linear correction unit activation function, and a one-dimensional max pooling layer connected in sequence. The convolutional layer is used to perform convolution operations on the input constellation map features and output a feature map. The linear correction unit activation function is used to perform a nonlinear transformation on the feature map output by the convolutional layer to obtain a nonlinearly transformed feature image. The one-dimensional max pooling layer is used to perform max pooling on the nonlinearly transformed feature image to form a feature vector.

[0029] The one-dimensional max pooling layers of the three convolutional neural networks are all connected to the average pooling layer; the average pooling layer is used to perform average pooling on the feature vectors obtained from the three convolutional neural networks and output constellation map feature vectors.

[0030] The convolutional layers in the three parallel convolutional neural networks use convolutional kernels of different sizes.

[0031] Optionally, the deep neural network includes an input layer and a hidden layer connected in sequence; the input layer is used to receive histogram of oriented gradients (HOR) features and transmit them to the hidden layer; the hidden layer is used to generate a histogram of oriented gradients (HOR) feature vector based on the HOR features.

[0032] Both the average pooling layer and the hidden layer are connected to the fusion layer;

[0033] The fusion layer is used to merge constellation graph feature vectors and histogram of oriented gradients feature vectors;

[0034] The fully connected layer is used to transform the merged constellation graph feature vector and the directional gradient histogram feature vector into a one-dimensional vector, and then output the predicted modulation format classification result through a soft maximum function.

[0035] A signal modulation format recognition system, comprising:

[0036] A hybrid neural network construction module is used to construct a multi-feature input hybrid neural network; the multi-feature input hybrid neural network includes a multi-scale convolutional neural network, a deep neural network, a fusion layer and a fully connected layer, the output of the multi-scale convolutional neural network and the output of the deep neural network are both connected to the input of the fusion layer, and the output of the fusion layer is connected to the input of the fully connected layer.

[0037] The feature extraction module is used to collect different types of signals in the optical fiber communication system and extract the constellation diagram features and directional gradient histogram features of each signal; among them, the different types of signals include probabilistically shaped signals and uniformly shaped signals;

[0038] The training module is used to input the constellation diagram features of each signal into the multi-scale convolutional neural network and the orientation gradient histogram features of each signal into the deep neural network to train the multi-feature input hybrid neural network and obtain the modulation format recognition model.

[0039] The application module is used to extract constellation diagram features and directional gradient histogram features from the signal to be identified, input them into the modulation format recognition model, and output the modulation format of the signal to be identified.

[0040] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the signal modulation format recognition method as described above.

[0041] A computer-readable storage medium having a computer program stored thereon, which, when executed, implements the signal modulation format recognition method as described above.

[0042] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0043] This invention discloses a signal modulation format recognition method, system, device, and medium. The constructed multi-feature input hybrid neural network includes a multi-scale convolutional neural network and a deep neural network. After extracting the constellation diagram features and histogram of directional gradients (HGPs) features of the signal, these are used as dual input features to the multi-scale convolutional neural network and the deep neural network, respectively, to identify the signal modulation format. This invention improves upon traditional single-constellation diagram feature extraction-based modulation format recognition methods by utilizing different types of features and different levels of representation, achieving accurate identification of the modulation formats of probabilistically shaped signals and uniformly shaped signals in optical fiber communication systems. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart of a signal modulation format recognition method provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the signal modulation format recognition method provided in an embodiment of the present invention;

[0047] Figure 3 A more specific flowchart of the signal modulation format recognition method provided in the embodiments of the present invention;

[0048] Figure 4 This is a schematic diagram of a constellation chart sample provided for an embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of the constellation diagram feature acquisition process provided in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of the directional gradient histogram feature acquisition process provided in an embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram illustrating the recognition accuracy provided in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] To address the shortcomings of existing technologies, this invention proposes a signal modulation format recognition method based on a multi-feature input hybrid neural network. This method requires fewer feature parameters, has simple steps, low complexity, and high recognition rate. It can also overcome the limitations of single features and single neural networks and is suitable for probabilistically shaped signals and uniformly shaped signals in optical fiber communication systems.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] like Figure 1 As shown, this embodiment of the invention provides a signal modulation format identification method, including:

[0056] Step 1: Construct a multi-feature input hybrid neural network; the multi-feature input hybrid neural network includes a multi-scale convolutional neural network, a deep neural network, a fusion layer and a fully connected layer. The output of the multi-scale convolutional neural network and the output of the deep neural network are both connected to the input of the fusion layer, and the output of the fusion layer is connected to the input of the fully connected layer.

[0057] The multi-scale convolutional neural network employs three parallel convolutional neural networks, each consisting of a convolutional layer, a linear rectifier unit (LRC) activation function, and a pooling layer. The purpose is to extract different scales and local features from the input data. The formula for the LRC activation function is shown below:

[0058] f(ε)=max(0,ε) (1)

[0059] Where ε represents the output after the convolution operation.

[0060] The convolutional layers in the three parallel convolutional neural networks employ convolutional kernels of different sizes. Each convolutional layer performs a convolution operation on the input constellation map features and outputs a feature map. This strategy allows the model to extract features at multiple scales, helping to better capture information from different scales of the input data and thus improving the model's ability to represent the input data.

[0061] A linear correction unit activation function is employed to perform a non-linear transformation on the feature maps output by the convolutional layers. This step enables multi-scale convolutional neural networks to learn more complex features and patterns.

[0062] A one-dimensional max-pooling layer was created in each of the three parallel convolutional neural networks to perform max pooling on the nonlinearly transformed feature images. Within each pooling window, the maximum value of multiple pixels was converted into a single pixel, extracting the most prominent features and forming a feature vector. This helps reduce the dimensionality of the data, decrease computational cost, and preserve the main features.

[0063] The feature vectors obtained from three parallel convolutional neural networks are subjected to average pooling and then output. This step helps to integrate information from different scales and features to obtain the final constellation map feature vectors.

[0064] A deep neural network consists of an input layer and a hidden layer. The input layer receives histogram of oriented gradients (HARQs) features as input, and the hidden layer processes these HARQ features to obtain a feature vector. These components work together to process the input data and generate the output of the neural network, resulting in the final HARQ feature vector.

[0065] In the fusion layer, the output constellation map feature vector and the directional gradient histogram feature vector are merged and input into the fully connected layer to become a one-dimensional vector. After passing through the soft maximum function, as shown in Equation (2), the probability distribution of the input sample in different signals is obtained, and the predicted classification result is output.

[0066]

[0067] Among them, z k =∑w nk x k +b k , z k ξ represents the output of the k-th neuron in the network. k Indicate z k The probability of x in all output results k w represents the input to the k-th neuron. nk b represents the weight. k This is the offset item.

[0068] Step 2: Collect different types of signals in the optical fiber communication system and extract the constellation diagram features and directional gradient histogram features of each signal.

[0069] I. Pretreatment

[0070] The collected signals of different types are normalized and converted into constellation diagrams, such as PNG format constellation diagrams with a pixel size of 512*512. Some samples are as follows: Figure 4 As shown, Figure 4In this context, "Modulation Format" refers to the modulation format. The constellation image is converted to grayscale to obtain a grayscale image. The constellation image set is then divided into a training set and a test set in a 4:1 ratio, as shown below. Figure 3 As shown in the figure. The different types of signals include probabilistically shaped signals and uniformly shaped signals.

[0071] II. Extracting constellation diagram features and directional gradient histogram features of the signal

[0072] Binarization and downsampling operations are performed on the grayscale constellation image to obtain constellation image features, such as... Figure 5 As shown. Binarization and downsampling help simplify the process and speed up the process.

[0073] Gradient calculations are performed on the grayscale constellation image to obtain its gradient information in the horizontal and vertical directions, such as... Figure 6 As shown.

[0074]

[0075] In formula (3): at pixel (i,j), U i (i,j) represents the gradient in the horizontal direction; U j (i,j) represents the gradient in the vertical direction; H(i,j) represents the pixel value. The gradient magnitude and gradient direction of pixel (i,j) can be calculated by formula (4):

[0076]

[0077] Where G(i,j) is the gradient magnitude of pixel (i,j), and θ(i,j) is the gradient direction of pixel (i,j).

[0078] The signal constellation diagram is divided into feature units of equal size.

[0079] The gradient direction is divided into several intervals, meaning the histogram of the signal constellation diagram is divided into several ranges. The magnitude of each feature unit is accumulated according to the region corresponding to the gradient direction. Then, the range of each feature unit is calculated, and finally, the histogram of that feature unit is calculated, forming the directional gradient histogram feature of each feature unit. That is, the horizontal axis of the histogram is the gradient direction, and the vertical axis is the gradient magnitude. Each interval is grouped into a rectangular diagram.

[0080] Several feature units are combined into a feature block, and the directional gradient histogram features of all feature units in a feature block are concatenated and normalized to obtain the directional gradient histogram features of that feature block.

[0081] The directional gradient histogram features of all feature blocks are concatenated to form the directional gradient histogram features of the entire image. The total directional gradient histogram features are obtained by concatenating the directional gradient histogram features of all images, which is the final directional gradient histogram features.

[0082] Step 3: Input the constellation diagram features of each signal into the multi-scale convolutional neural network, and input the directional gradient histogram features of each signal into the deep neural network to train the multi-feature input hybrid neural network and obtain the modulation format recognition model.

[0083] The constellation graph features in the training set are input into the multi-scale convolutional neural network, and the directional gradient histogram features in the training set are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain the trained multi-feature input hybrid neural network.

[0084] Based on the constellation diagram features and directional gradient histogram features in the test set, the performance of the trained multi-feature input hybrid neural network is tested, and the trained multi-feature input hybrid neural network that passes the test is determined as the modulation format recognition model.

[0085] The constellation diagrams from the test set are input into a modulation format recognition model constructed using a multi-feature hybrid neural network to test the performance of the constructed model (e.g., Figure 2 (Performance analysis in the model). The recognition rate of the classifier is used as the evaluation metric to measure the recognition performance of the model.

[0086] Step 4: After extracting constellation diagram features and directional gradient histogram features from the signal to be identified, input them into the modulation format recognition model and output the modulation format of the signal to be identified.

[0087] This method identifies three types of PS-quadrature amplitude modulation (QAM) signals (3-bit / symbol PS-16QAM, 4-bit / symbol PS-64QAM, and 5-bit / symbol PS-64QAM) and two types of US-QAM signals (4-bit / symbol US-16QAM and 6-bit / symbol US-64QAM) in a polarization-multiplexed (PDM) high-speed optical communication system. The identification accuracy of these five signal types increases with the optical signal-to-noise ratio as follows: Figure 7 As shown, when the optical signal-to-noise ratio is greater than or equal to 17.5dB, the recognition accuracy of these five signal modulation formats remains stable at 100%.

[0088] This invention proposes a signal modulation format recognition method based on a multi-feature input hybrid neural network. First, the constellation diagram features and directional gradient histogram features of the signal are extracted and used as dual input features of the multi-feature input hybrid neural network. The network is then trained in a multi-scale convolutional neural network and a deep neural network, respectively, and a modulation format recognition model is constructed. Finally, by utilizing different types of features and different levels of representation, the accurate recognition of probabilistically shaped and uniformly shaped signals in optical fiber communication systems is achieved.

[0089] This invention improves upon traditional single-method modulation format recognition based on constellation diagram feature extraction. It is simple in principle, has low computational complexity, and can be easily upgraded from existing systems by integrating with neural networks. This method exhibits good recognition performance for probabilistically shaped and uniformly shaped signals in optical fiber communication systems.

[0090] In order to execute the methods corresponding to the above embodiments and achieve the corresponding functions and technical effects, a signal modulation format recognition system is provided below, including:

[0091] A hybrid neural network construction module is used to construct a multi-feature input hybrid neural network. The multi-feature input hybrid neural network includes a multi-scale convolutional neural network, a deep neural network, a fusion layer, and a fully connected layer. The outputs of the multi-scale convolutional neural network and the deep neural network are both connected to the input of the fusion layer, and the output of the fusion layer is connected to the input of the fully connected layer.

[0092] The feature extraction module is used to acquire different types of signals in the optical fiber communication system and extract the constellation diagram features and directional gradient histogram features of each signal. The different types of signals include probabilistically shaped signals and uniformly shaped signals.

[0093] The training module is used to input the constellation diagram features of each signal into the multi-scale convolutional neural network and the directional gradient histogram features of each signal into the deep neural network, thereby training the multi-feature input hybrid neural network to obtain a modulation format recognition model.

[0094] The application module is used to extract constellation diagram features and directional gradient histogram features from the signal to be identified, input them into the modulation format recognition model, and output the modulation format of the signal to be identified.

[0095] The signal modulation format recognition system provided in this embodiment of the invention has a similar working principle and beneficial effects to the signal modulation format recognition method described in the above embodiments, so it will not be described in detail here. For details, please refer to the introduction of the above method embodiments.

[0096] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the signal modulation format recognition method as described above.

[0097] Furthermore, when the computer program in the aforementioned memory is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0098] Furthermore, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the signal modulation format recognition method as described above.

[0099] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0100] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying signal modulation formats, characterized in that, include: Construct a multi-feature input hybrid neural network; the multi-feature input hybrid neural network includes a multi-scale convolutional neural network, a deep neural network, a fusion layer and a fully connected layer, the output of the multi-scale convolutional neural network and the output of the deep neural network are both connected to the input of the fusion layer, and the output of the fusion layer is connected to the input of the fully connected layer; Different types of signals in an optical fiber communication system are acquired, and the constellation diagram features and directional gradient histogram features of each signal are extracted; among them, the different types of signals include probabilistically shaped signals and uniformly shaped signals; The constellation diagram features of each signal are input into the multi-scale convolutional neural network, and the orientation gradient histogram features of each signal are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain a modulation format recognition model. After extracting constellation diagram features and directional gradient histogram features from the signal to be identified, the modulation format identification model is input and the modulation format of the signal to be identified is output. The multi-scale convolutional neural network includes: three parallel convolutional neural networks and an average pooling layer; Each convolutional neural network includes a convolutional layer, a linear correction unit activation function, and a one-dimensional max pooling layer connected in sequence. The convolutional layer is used to perform convolution operations on the input constellation map features and output a feature map. The linear correction unit activation function is used to perform a nonlinear transformation on the feature map output by the convolutional layer to obtain a nonlinearly transformed feature image. The one-dimensional max pooling layer is used to perform max pooling on the nonlinearly transformed feature image to form a feature vector. The one-dimensional max pooling layers of the three convolutional neural networks are all connected to the average pooling layer; the average pooling layer is used to perform average pooling on the feature vectors obtained from the three convolutional neural networks and output constellation map feature vectors. The convolutional layers in the three parallel convolutional neural networks use convolutional kernels of different sizes; The deep neural network includes an input layer and a hidden layer connected in sequence; the input layer is used to receive histogram of oriented gradients (HOR) features and transmit them to the hidden layer; the hidden layer is used to generate a histogram of oriented gradients (HOR) feature vector based on the HOR features. Both the average pooling layer and the hidden layer are connected to the fusion layer; The fusion layer is used to merge constellation graph feature vectors and histogram of oriented gradients feature vectors; The fully connected layer is used to transform the merged constellation graph feature vector and the directional gradient histogram feature vector into a one-dimensional vector, and then output the predicted modulation format classification result through a soft maximum function.

2. The signal modulation format recognition method according to claim 1, characterized in that, Extract constellation diagram features and histogram features of directional gradients for each signal, specifically including: The collected signals of different types are normalized and then converted into constellation diagrams; The constellation diagram of each signal is converted to grayscale to obtain the grayscale constellation diagram of each signal; All grayscale constellation images are combined into a constellation image set, and the constellation image set is divided into a training set and a test set in a 4:1 ratio. Binarization and downsampling are performed on each grayscale constellation image to obtain the constellation image features of each signal; Gradient calculation is performed on each grayscale constellation image to obtain the gradient information of each grayscale constellation image; wherein, the gradient information includes the gradient magnitude and gradient direction; The grayscale constellation image is divided into multiple feature blocks, each feature block is divided into multiple feature units of equal size, and the gradient direction is divided into several intervals. The gradient magnitude of each feature unit is accumulated according to the interval corresponding to the gradient direction to form the directional gradient histogram feature of each feature unit. The directional gradient histogram features of all feature units in each feature block are concatenated and normalized to obtain the directional gradient histogram features of each feature block. The directional gradient histogram features of all feature blocks are concatenated to form the directional gradient histogram features of each signal.

3. The signal modulation format recognition method according to claim 2, characterized in that, Gradient calculation is performed on each grayscale constellation image to obtain its gradient information, specifically including: According to the formula The gradient of the grayscale constellation image is calculated to obtain the gradient in the horizontal direction and the gradient in the vertical direction; where, Represents pixels The gradient in the horizontal direction, Represents pixels The gradient in the vertical direction, Represents pixels pixel values, Represents pixels pixel values, Represents pixels pixel values, Represents pixels Pixel values; Based on the gradients of the grayscale constellation image in the horizontal and vertical directions, using the formula... Calculate the gradient magnitude and gradient direction of each pixel in the grayscale constellation image; where, For pixels gradient magnitude, For pixels The gradient direction.

4. The signal modulation format recognition method according to claim 2, characterized in that, The constellation diagram features of each signal are input into the multi-scale convolutional neural network, and the histogram features of the directional gradients of each signal are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain a modulation format recognition model, specifically including: The constellation graph features in the training set are input into the multi-scale convolutional neural network, and the directional gradient histogram features in the training set are input into the deep neural network. The multi-feature input hybrid neural network is trained to obtain the trained multi-feature input hybrid neural network. Based on the constellation diagram features and directional gradient histogram features in the test set, the performance of the trained multi-feature input hybrid neural network is tested, and the trained multi-feature input hybrid neural network that passes the test is determined as the modulation format recognition model.

5. A signal modulation format recognition system, characterized in that, The system is used to perform the signal modulation format recognition method according to any one of claims 1 to 4, and the system comprises: A hybrid neural network construction module is used to construct a multi-feature input hybrid neural network; the multi-feature input hybrid neural network includes a multi-scale convolutional neural network, a deep neural network, a fusion layer and a fully connected layer, the output of the multi-scale convolutional neural network and the output of the deep neural network are both connected to the input of the fusion layer, and the output of the fusion layer is connected to the input of the fully connected layer. The feature extraction module is used to collect different types of signals in the optical fiber communication system and extract the constellation diagram features and directional gradient histogram features of each signal; among them, the different types of signals include probabilistically shaped signals and uniformly shaped signals; The training module is used to input the constellation diagram features of each signal into the multi-scale convolutional neural network and the orientation gradient histogram features of each signal into the deep neural network to train the multi-feature input hybrid neural network and obtain the modulation format recognition model. The application module is used to extract constellation diagram features and directional gradient histogram features from the signal to be identified, input them into the modulation format recognition model, and output the modulation format of the signal to be identified.

6. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the signal modulation format recognition method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed, implements the signal modulation format recognition method as described in any one of claims 1 to 4.

Images