A modulation format recognition method based on constellation diagram multi-feature extraction algorithm

Abstract

The present invention discloses a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm. First, the signal constellation diagram is divided using the fractal dimension of the weighted linear least squares algorithm, and the fractal dimension in each area is calculated, and this result is used as one of the image features; then, the gray-level co-occurrence matrix is used to extract the eigenvalues of the image in different directions, and their mean and variance are calculated, and the result is used as another feature; finally, the two features are input into a modulation format classifier constructed by a support vector machine to realize the recognition of the modulation format in the optical communication system. The method of the present invention improves the traditional single modulation recognition method based on feature extraction, has a simple principle, low computational complexity, high recognition accuracy for commonly used modulation formats, and is integrated with machine learning algorithms, making it easy to upgrade from existing systems. The average recognition rate of modulation formats of this method exceeds 98%.

Description

Technical Field

[0001] The present invention relates to the field of optical communication technology, and in particular to a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm. Background Art

[0002] In recent years, with the rapid development of services such as 5G, cloud computing, high-definition video, and cloud conferencing, the demand for bandwidth and spectrum utilization has continued to increase, and the channel capacity in optical communication systems has grown exponentially. Future fiber-optic communication systems are expected to be dynamic and heterogeneous, accommodating a variety of signals with different modulation formats to meet the diverse needs of users. Furthermore, optical signals are susceptible to various transmission impairments that can change dynamically over time. This places new demands on optical receivers.

[0003] To demodulate a transmitted signal in a digital receiver, the modulation format must be known. Therefore, correctly identifying the modulation format is crucial for high-quality communication. Existing algorithms for modulation format recognition can be broadly divided into two categories based on their identification principles: likelihood-based hypothesis testing methods and feature-based recognition methods.

[0004] Likelihood-based hypothesis testing methods formulate the modulation recognition problem using probabilities and hypothesis testing parameters. This approach requires formulating correct hypotheses and carefully selecting appropriate thresholds. Likelihood-based hypothesis testing minimizes the probability of misidentification, thus providing an optimal solution in Bayesian semantics. However, this approach carries a higher computational complexity and is difficult to implement.

[0005] In feature-based recognition methods, significant features need to be extracted from the received signal, and then these features are used to identify the signal modulation format. Although feature-based recognition methods are simpler and easier to use than hypothesis testing methods based on likelihood functions, they do not provide optimal recognition performance and are therefore suboptimal methods. Therefore, how to make feature-based recognition provide near-optimal recognition performance through reasonable design is a major problem that needs to be solved in this field. Based on this, the present invention proposes a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm. First, the signal constellation diagram is divided using the fractal dimension of the weighted linear least squares algorithm, and the fractal dimension in each area is calculated, and this result is used as one of the image features. Then, the gray-level co-occurrence matrix is used to extract the eigenvalues of the image in different directions, and their mean and variance are calculated, and the result is used as another feature. Finally, the two features are converted into multiple feature vectors and input into a modulation format classifier constructed by a support vector machine to identify the modulation formats of different signals. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention proposes a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm. This method requires few feature parameters, has simple steps, low complexity, high recognition rate, and can make up for the limitations of a single feature. It is suitable for the recognition of various types of modulation signals.

[0007] In order to achieve the above object, the technical solution of the present invention is as follows:

[0008] The present invention proposes a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm. First, the signal constellation diagram is divided into sections using the fractal dimension of the weighted linear least squares algorithm, and the fractal dimension in each area is calculated, and this result is used as one of the image features; then, the gray-level co-occurrence matrix is used to extract the eigenvalues of the image in different directions, and their mean and variance are calculated, and the result is used as another feature; finally, the two features are converted into multiple feature vectors and input into a modulation format classifier constructed by a support vector machine to identify the modulation formats of different signals.

[0009] Furthermore, the modulation format recognition method based on the constellation diagram multi-feature extraction algorithm includes the following steps:

[0010] S1. Signal preprocessing: saving the generated constellation diagram as an image, performing binarization processing and analysis on the image to obtain a binary image;

[0011] S2. First, the signal constellation is partitioned using the fractal dimension of the weighted linear least squares algorithm. The fractal dimension of each region is calculated and used as one of the image features. The gray-level co-occurrence matrix is then used to extract the eigenvalues of the image in different directions. Their mean and variance are calculated and used as another feature. Finally, the two features are converted into a multi-eigenvector.

[0012] S3, constructing a modulation format classifier in combination with a support vector machine;

[0013] S4. Input the multiple feature vectors in step S2 into the modulation format classifier established in step S3 to identify the modulation format of the signal to be tested.

[0014] Furthermore, the step S1 saves the constellation diagram as an image in PNG format with a pixel size of 512*512.

[0015] Furthermore, the fractal dimension algorithm of the weighted linear least squares algorithm in step S2 is:

[0016] Represent the two-dimensional constellation map as a three-dimensional surface, where (x, y) represents the pixel position on the image plane and the third coordinate z is represented by the pixel grayscale value;

[0017] Divide the constellation map with pixel size of M*M into small blocks of size s*s, where s is an integer, the scale of the grid is r, and it is given by Each small block has a column of s*s*h boxes, where G is the total number of gray levels of the grayscale image, which is calculated by formula (1):

[0018]

[0019] The maximum and minimum gray levels of the image at position (i, j) fall into the gth and lth boxes respectively, n r is the number of boxes of size s*s*h, calculated by formula (2):

[0020] n r (i,j)=g-l+1 (2)

[0021] The total number of boxes N at grid scale r r Calculated by formula (3):

[0022] N r =∑ i,j n r (i,j) (3)

[0023] Fitting the straight line to each point (x d ,y d ) corresponds to a different grid size s, where y d =logN r , using the trapezoidal membership function to assign weight W(s) to each grid size to represent the corresponding point (x d ,y d ), as shown in formula (4):

[0024]

[0025] Among them, a t 、b t 、c t and d t The value of is estimated using formula (5):

[0026]

[0027] At the point (x d ,y d ), the weighted least squares algorithm is used to fit these points to the regression line y = ax + b, and finally the fractal dimension of the image is calculated using formula (6):

[0028]

[0029] Furthermore, the gray level co-occurrence matrix algorithm in step S2 is:

[0030] The gray-level co-occurrence matrix is an n*n square matrix, where n is the number of gray-level categories in the image. The (i, j) element in the matrix represents the number of pixel pairs. The gray-level co-occurrence matrices in the four directions of 0°, 45°, 90°, and 135° are calculated, and p(i, j) is normalized according to formula (7):

[0031]

[0032] Where R is the normalization constant;

[0033] The image blocks are retrieved using the four features of angular second moment m1, entropy m2, contrast m3 and correlation m4, which are calculated using formula (8):

[0034]

[0035] Where: μ x =∑ i ∑ j i*P(i,j), μ y =∑ i ∑ j j*P(i,j),δ x =∑ i ∑ j (i-μ x ) 2 *P(i,j),δ y =∑ i ∑ j (j-μ y ) 2 *P(i,j);

[0036] The mean and variance of m1, m2, m3 and m4 are used as texture features.

[0037] Furthermore, the step S3 performs binary classification processing on the feature data using a support vector machine, and separates the two types of data by constructing a hyperplane.

[0038] Furthermore, the training process of constructing the modulation format classifier using the support vector machine in step S3 is as follows:

[0039] Given a training data set D = {(x1, y1), (x2, y2), ..., (x M ,y M )}. The feature vector and label of the sample are x i and y i , where the eigenvectors satisfy There are m features in total, and the labels satisfy y i ∈{-1,+1}, to solve the following optimization problem to train a maximum margin model with parameters w∈R and b∈R:

[0040]

[0041] The classification accuracy of the modulation format classifier is expressed by formula (10):

[0042]

[0043] Where Ⅱ represents the indicator function, which is 1 when the condition in the brackets is met, otherwise it is 0, i is a vector of nonnegative Lagrange multipliers.

[0044] Furthermore, the step S3 converts the feature data into one-dimensional data, and randomly extracts feature sets multiple times in a ratio of 7:3 to obtain training feature data and test feature data.

[0045] Furthermore, the signal to be tested in step S4 includes multiple modulation formats in optical communication.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention proposes a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm. First, the signal constellation diagram is divided using the fractal dimension of the weighted linear least squares algorithm, and the fractal dimension in each area is calculated, and the result is used as one of the image features. Then, the gray-level co-occurrence matrix is used to extract the eigenvalues of the image in different directions, and their mean and variance are calculated, and the result is used as another feature. Finally, the two features are input into the modulation format classifier constructed by the support vector machine to realize the modulation format recognition in the optical communication system. The average modulation format recognition rate of this method exceeds 98%. This method improves the traditional single modulation recognition method based on feature extraction. It has a simple principle, low computational complexity, high recognition accuracy for commonly used modulation formats, and is integrated with the machine learning algorithm, making it easy to upgrade from the existing system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention, and those skilled in the art can also obtain other drawings based on these drawings.

[0049] Figure 1 A schematic flow chart of a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm provided in an embodiment of the present invention.

[0050] Figure 2 A schematic diagram of a constellation diagram sample required by an embodiment of the present invention.

[0051] Figure 3 A schematic diagram of fractal dimension algorithm processing of the weighted linear least squares algorithm provided in an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of the gray-level co-occurrence matrix algorithm processing provided by an embodiment of the present invention;

[0053] Figure 5 A schematic diagram of a modulation format classifier constructed using a support vector machine according to an embodiment of the present invention;

[0054] Figure 6 The optical communication system of the prior art provided by the embodiment of the present invention;

[0055] Figure 7 A schematic diagram of the average recognition rate provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0058] In order to better understand the present technical solution, the method of the present invention is described in detail below with reference to the accompanying drawings.

[0059] refer to Figure 1 As shown, the present invention provides a modulation format recognition method based on a constellation diagram multi-feature extraction algorithm, comprising:

[0060] 1. Preprocessing

[0061] The constellation map is saved as a PNG image with a pixel size of 512*512. A total of 10,000 images are included. Some samples are as follows: Figure 2As shown in the figure, by binarizing and analyzing the image, a binary image is obtained. This is beneficial for further processing of the image. The collective properties of the image are only related to the positions of points with pixel values of 0 or 255, and no longer involve multi-level values of pixels, making processing simpler and reducing the amount of data processing and compression.

[0062] 2. Multi-feature extraction algorithm based on constellation diagram

[0063] First, the signal constellation is partitioned using the fractal dimension of the weighted linear least squares algorithm. The fractal dimension of each region is calculated and used as one of the image features. Next, the gray-level co-occurrence matrix is used to extract eigenvalues of the image in different directions. Their mean and variance are calculated and used as another feature. Finally, these two features are converted into a multi-feature vector. This method abandons the single detection feature used by previous methods and adopts a multi-dimensional feature detection mechanism, improving detection accuracy.

[0064] The fractal dimension algorithm of the weighted linear least squares algorithm is as follows Figure 3 As shown in , when the constellation map is used as the input image to calculate its fractal dimension, since the grayscale value of the vector point concentrated area is generally higher than that of other blank areas, the fractal dimension calculated thereby will inevitably have a large difference, such as Figure 3 As shown, the constellation diagram is divided into 16×16 regions, and the size of each region is subsequently determined and the fractal box dimension is calculated. When constructing the fractal feature matrix, the fractal box dimension of each small region is calculated using the weighted least squares algorithm and sorted according to its position in the region, resulting in a 16×16 feature matrix.

[0065] The calculation process is to imagine the two-dimensional constellation map as a three-dimensional surface, where (x, y) represents the pixel position on the image plane and the third coordinate z is represented by the pixel grayscale value. We divide the constellation map with a pixel size of M*M into small blocks of size s*s, where s is an integer. The scale of the grid is r, which is determined by Each small block has a column of s*s*h boxes. Where G is the total number of gray levels of the 8-bit grayscale image, that is, 256, which can be calculated using formula (1).

[0066]

[0067] The maximum and minimum gray levels of the image at position (i, j) fall into the gth and lth boxes respectively, n r is the number of boxes of size s*s*h, which can be calculated using formula (2).

[0068] n r (i,j)=g-l+1 (2)

[0069] The total number of boxes N at grid scale r r It can be calculated using formula (3).

[0070] N r =∑ i,j n r (i,j) (3)

[0071] The fractal dimension of the weighted linear least squares algorithm uses the rule of trapezoidal membership function to assign weights to each grid size. d ,y d ) corresponds to a different grid size s, and the grid size s is assigned a weight W(s). y d =logN r , in other words, W(s) is the point (x d ,y d ) weights, W(s) and w s can be used interchangeably to indicate the corresponding point (x d ,y d The trapezoidal membership function uses formula (4) to assign a weight to each data point.

[0072]

[0073] According to formula (4), the trapezoidal membership function first divides all data points into five groups, whose points must be located in the interval (-∞, a t )、(a t ,b t ),[b t ,c t ]、(c t ,d t ) and [d t ,+∞). The weights of all points in the first and fifth groups are 0, while the weights of all points in the third group are 1. The weights of the points in the second and fourth groups are assigned in the range (0,1) according to their proximity to the boundary points of the corresponding groups. In this study, a t 、b t 、c t and d t The value of is estimated using formula (5).

[0074]

[0075] Since the size of all images we use is 512×512 pixels, we can calculate a by formula (5): t =1, b t =8, c t =27 and dt = 257. At the point of acquisition (x d ,y d ), the weighted linear least squares algorithm is used to fit these points to the regression line y = ax + b, and finally the fractal dimension of the image is calculated using formula (6).

[0076]

[0077] like Figure 4 The following shows an example of a gray-level co-occurrence matrix algorithm for image conversion. This matrix is an n*n square matrix, where n is the number of grayscale levels in the image. The (i, j) element in the matrix represents the number of pixel pairs. The gray-level co-occurrence matrix is calculated for four directions (0°, 45°, 90°, and 135°). p(i, j) is then normalized:

[0078]

[0079] Where R is the normalization constant.

[0080] The image blocks are retrieved using the four features of angular second moment m1, entropy m2, contrast m3 and correlation m4, which are calculated using formula (8):

[0081]

[0082] Where: μ x =∑ i ∑ j i*P(i,j), μ y =∑ i ∑ j j*P(i,j),δ x =∑ i ∑ j (i-μ x ) 2 *P(i,j),δ t =∑ i ∑ j (j-μ y ) 2 *P(i,j). For the convenience of calculation, the mean and variance of m1, m2, m3 and m4 are used as texture features.

[0083] Finally, the fractal dimension of the weighted linear least squares algorithm and the features obtained by the gray-level co-occurrence matrix are fused to generate multiple eigenvectors of the constellation diagram.

[0084] 3. Support vector machine constructs modulation format classifier for classification and recognition

[0085] In this method, support vector machines are used to perform binary classification on feature data. A hyperplane is constructed to separate the two types of data so as to maximize the boundary. Given a training data set D containing M samples, D = {(x1, y1), (x2, y2), ..., (x M ,y M )}. The feature vector and label of the sample are x i and y i , where the eigenvectors satisfy There are m features in total, and the labels satisfy y i ∈{-1,+1}. It solves the following optimization problem to train a maximum margin model with parameters w∈R and b∈R.

[0086]

[0087] Measuring the performance of the classifier requires comparing the prediction results with y i For comparison, the classification accuracy can be expressed by formula (10) as

[0088]

[0089] Where Ⅱ represents the indicator function, which is 1 when the condition in the brackets is met, otherwise it is 0, i is a non-negative Lagrange multiplier vector. According to formula (10), the accuracy of the classifier can be obtained by adding the number of predicted samples whose labels are the same as the original labels and dividing it by the total number of samples M.

[0090] like Figure 5 As shown in the figure, in order to facilitate the learning of the support vector machine learning algorithm, the feature data needs to be converted into one-dimensional data. The feature set is randomly extracted multiple times in a ratio of 7:3 to obtain training feature data and test feature data. The support vector machine is used to construct a classifier, and the support vector machine learning algorithm is used to learn the training data to construct a modulation format recognition classifier to achieve modulation format recognition.

[0091] The confusion matrix drawn by the average recognition accuracy of the seven types of signals used in this method is as follows Figure 7 As shown, the recognition rates for QPSK, 8QAM, 16QAM, 32QAM, and 256QAM signals are all high, exceeding 98%. QPSK and 8QAM signals, in particular, achieve a 100% recognition rate. However, there is a possibility of misidentification between 64QAM and 128QAM signals. However, the recognition algorithm used in this paper achieves an overall recognition rate of over 95% for 64QAM and over 97% for 128QAM.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. A modulation format recognition method based on a constellation diagram multi-feature extraction algorithm, characterized in that: First, the signal constellation is partitioned using the fractal dimension of the weighted linear least squares algorithm. The fractal dimension of each region is calculated and used as one of the image features. Then, the gray-level co-occurrence matrix is used to extract the eigenvalues of the image in different directions. Their mean and variance are calculated and used as another feature. Finally, the two features are converted into multi-feature vectors and input into the modulation format classifier constructed by the support vector machine to identify the modulation formats of different signals. The modulation format identification method includes the following steps: S1. Signal preprocessing: saving the generated constellation diagram as an image, performing binarization processing and analysis on the image to obtain a binary image; S2. First, the signal constellation is divided using the fractal dimension of the weighted linear least squares algorithm. The fractal dimension of each region is calculated and used as one of the image features. Then, the gray-level co-occurrence matrix is used to extract the eigenvalues of the image in different directions. Their mean and variance are calculated and used as another feature. Finally, the two features are converted into a multi-eigenvector. S3, constructing a modulation format classifier in combination with a support vector machine; S4, inputting the multiple feature vectors in step S2 into the modulation format classifier established in step S3 to identify the modulation format of the signal to be tested; The fractal dimension algorithm of the weighted linear least squares algorithm in step S2 is: Represent the two-dimensional constellation map as a three-dimensional surface, where (x, y) represents the pixel position on the image plane and the third coordinate z is represented by the pixel grayscale value; Divide the constellation map with pixel size of M*M into small blocks of size s*s, where s is an integer, the scale of the grid is r, and it is given by Each small block has a column of s*s*h boxes, where G is the total number of gray levels of the grayscale image, which is calculated by formula (1): The maximum and minimum gray levels of the image at position (i, j) fall into the gth and lth boxes respectively, n r is the number of boxes of size s*s*h, calculated by formula (2): n r (i,j)=g-l+1 (2) The total number of boxes N at grid scale r r Calculated by formula (3): N r =∑ i,j n r (i,j) (3) Fitting the straight line to each point (x d ,y d ) corresponds to a different grid size s, where y d =logN r , using the trapezoidal membership function to assign weight W(s) to each grid size to represent the corresponding point (x d ,y d ), as shown in formula (4): Among them, a t 、b t 、c t and d t The value of is estimated using formula (5): At the point (x d ,y d ), the weighted least squares algorithm is used to fit these points to the regression line y = ax + b, and finally the fractal dimension of the image is calculated using formula (6):

2. The modulation format recognition method based on the constellation diagram multi-feature extraction algorithm according to claim 1 is characterized in that: The gray level co-occurrence matrix algorithm in step S2 is: The gray-level co-occurrence matrix is an n*n square matrix, where n is the number of gray-level categories in the image. The (i, j) element in the matrix represents the number of pixel pairs. The gray-level co-occurrence matrices in the four directions of 0°, 45°, 90°, and 135° are calculated, and p(i, j) is normalized according to formula (7): Where R is the normalization constant; The image blocks are retrieved using the four features of angular second moment m1, entropy m2, contrast m3 and correlation m4, which are calculated using formula (8): among them: m x =∑ i ∑ j i*P(i,j),μ y =∑ i ∑ j j*P(i,j),δ x =∑ i ∑ j (i-m x ) 2 *P(i,j),δ y =∑ i ∑ j (j-μ y ) 2 *P(u,j); The mean and variance of m1, m2, m3 and m4 are used as texture features.

3. The modulation format recognition method based on the constellation diagram multi-feature extraction algorithm according to claim 1 is characterized in that: The step S3 uses a support vector machine to perform binary classification processing on the feature data, and separates the two types of data by constructing a hyperplane.

4. The modulation format recognition method based on the constellation diagram multi-feature extraction algorithm according to claim 1 is characterized in that: The training process of constructing the modulation format classifier using the support vector machine in step S3 is as follows: Given a training data set D = {(x1, y1), (x2, y2), ..., (x M ,y M )}, the feature vector and label of the sample are x i and y i , where the eigenvectors satisfy There are m features in total, and the labels satisfy y i ∈{-1,+1}, to solve the following optimization problem to train a maximum margin model with parameters w∈R and b∈R: The classification accuracy of the modulation format classifier is expressed by formula (10): Where Ⅱ represents the indicator function, which is 1 when the condition in the brackets is met, otherwise it is 0, i is a vector of nonnegative Lagrange multipliers.

5. The modulation format recognition method based on the constellation diagram multi-feature extraction algorithm according to claim 4 is characterized in that: The step S3 converts the feature data into one-dimensional data, and randomly extracts feature sets multiple times in a ratio of 7:3 to obtain training feature data and test feature data.

6. The modulation format recognition method based on the constellation diagram multi-feature extraction algorithm according to claim 1 is characterized in that: The signal to be tested in step S4 includes multiple modulation formats in optical communication.

Images