A communication signal modulation identification method and system based on Angora rabbit search mechanism

Abstract

The present invention discloses a communication signal modulation recognition method and system based on an Angora rabbit search mechanism, and relates to the field of communication signal processing technology. The technical points of the present invention include: constructing a digital modulation signal data set; preprocessing the data in the digital modulation signal data set to obtain a training set and a test set; based on the training set and the test set, using the Angora rabbit search mechanism to optimize and solve the hyperparameters of the support vector machine, the hyperparameters including a penalty factor and a Gaussian kernel; using the training set to train a support vector machine with the optimal hyperparameters; and using the trained support vector machine to perform modulation recognition on the communication signal to be tested. The present invention solves the overfitting problem of the support vector machine caused by improper selection of hyperparameters. It has the advantages of high recognition accuracy and fast convergence speed in the field of communication signal modulation recognition, and can also obtain good recognition effects under low signal-to-noise ratios. It has a wider range of application prospects and scenarios and can be applied to solving various types of engineering application problems.

Description

Technical Field

[0001] The present invention relates to the technical field of communication signal processing, and in particular to a communication signal modulation identification method and system based on an Angora rabbit search mechanism. Background Art

[0002] Modulation identification is a critical technology in wireless communications. Identifying the modulation scheme of wireless communication signals is fundamental to electronic countermeasures, electronic reconnaissance, non-cooperative communications, smart antennas, and wireless spectrum management. This technology has widespread application and significant value in both military and civilian fields. In recent years, with the rapid advancements in wireless communications, electronics, and signal processing, the modulation schemes used in wireless communication signals have become increasingly complex, and the types of modulation signals have expanded. With the proliferation of both military and civilian wireless communication equipment, noise interference in wireless communication channels has increased significantly, with Gaussian noise being the most typical noise model in communication channels.

[0003] Typically, modulation identification of communication signals utilizes pattern recognition methods. This involves first extracting features from the signal received at the receiving end, extracting the corresponding characteristic parameters, and then using a pattern recognition system to determine the signal's modulation type. Classifiers for determining signal modulation types generally fall into two categories: classic pattern recognition classifiers and neural network classifiers. Using classic classifiers such as the K-nearest neighbor classifier, support vector machine (SVM), and decision tree classifiers to identify signal modulation types is currently a mainstream research area in communication signal modulation identification.

[0004] After searching relevant literature, Li Yang et al. proposed an automatic modulation recognition technique in their paper "Research on Automatic Identification Algorithms for Digital Modulation Modes of Communication Signals" published in the Journal of Electronics and Information Technology (2005, No. 2, pp. 197–201). This algorithm has a simple structure, low computational complexity, and is suitable for real-time computation. It also achieves good recognition results even in low signal-to-noise ratio environments. However, this method was proposed relatively early, and compared with more recent algorithms, its modulation recognition accuracy still needs to be improved. Wang Haibin et al. proposed a method for modulation mode recognition based on transient characteristics of digital communication signals in their paper "Modulation Mode Recognition Method Based on Instantaneous Characteristics of Digital Communication Signals" published in Modern Electronic Technology (2019, Vol. 41, No. 19, pp. 22–25). This method achieves high recognition accuracy and speed even in low signal-to-noise ratio environments. However, this method uses a decision-theoretic discriminant tree as a classifier, requiring not only the selection of appropriate thresholds for each feature parameter but also the appropriate order of feature parameter discrimination in the classification algorithm. This method may not achieve high recognition rates when the environment changes. Chen Fatang et al. proposed a modulation recognition algorithm based on space-time convolutional network in the article "Communication Signal Modulation Recognition Based on Space-Time Convolutional Network" published in "Telecommunication Technology" (2024, pp. 1-7). Compared with the traditional neural network modulation signal recognition method, it has fewer parameters and higher recognition accuracy. However, its recognition accuracy is significantly reduced under low signal-to-noise ratio conditions, and the applicable signal-to-noise ratio range is small. In addition, the parameter setting process is complicated and time-sensitive, making it difficult to solve complex engineering application problems.

[0005] Existing literature indicates that communication signal modulation recognition using neural networks as classifiers often faces challenges such as complex parameter selection, significant influence of initial weights, thresholds, network structure, and training functions on network training speed, training results, and convergence, and low timeliness. Classical classifiers for modulation signal recognition are susceptible to environmental factors. Furthermore, both methods exhibit a significant decline in recognition accuracy in low signal-to-noise ratio environments. To achieve high-precision modulation recognition, a feasible and effective communication signal modulation recognition method is needed. Summary of the Invention

[0006] In view of the above problems, the present invention proposes a communication signal modulation identification method and system based on the Angora rabbit search mechanism.

[0007] According to one aspect of the present invention, a communication signal modulation identification method based on an Angora rabbit search mechanism is proposed, the method comprising:

[0008] Step 1: Construct a digital modulation signal dataset;

[0009] Step 2: Preprocess the data in the digital modulation signal dataset to obtain a training set and a test set;

[0010] Step 3: Based on the training set and the test set, the hyperparameters of the support vector machine are optimized using the Angora rabbit search mechanism, where the hyperparameters include a penalty factor and a Gaussian kernel;

[0011] Step 4: Use the training set to train the support vector machine with the optimal hyperparameters;

[0012] Step 5: Use the trained support vector machine to perform modulation recognition on the communication signal to be tested.

[0013] Furthermore, the digital modulation signal data set in step 1 is constructed as follows:

[0014] A unipolar sequence with M code elements is input, and the digital sequence is mapped into a pulse waveform using a unipolar non-return-to-zero code to obtain a digital baseband signal.

[0015] Constructing a shaping filter, wherein the shaping filter adopts a raised cosine roll-off function to perform shaping processing on the digital baseband signal;

[0016] Modulating the digital baseband signal after shaping and filtering using different modulation methods to obtain a modulated signal;

[0017] Gaussian noise with different signal-to-noise ratios is inserted into the modulated signal to obtain a digital modulation signal dataset of the communication signal under the Gaussian noise background.

[0018] Furthermore, the step 2 of preprocessing the data in the digital modulation signal dataset to obtain the training set and the test set includes:

[0019] The digital modulation signal dataset under Gaussian noise background is normalized and segmented, and the communication signal of each modulation mode is divided into multiple data segments of equal length. Each data segment is combined with the corresponding label to obtain a digital modulation signal preprocessing dataset.

[0020] Performing a Hilbert transform on a digital modulation signal preprocessing data set to extract the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the digital signal, and then extracting instantaneous characteristic parameters to obtain a characteristic parameter data set; the instantaneous characteristic parameters include the mean of the instantaneous amplitude envelope, the fourth-order moment compactness of the normalized center instantaneous frequency, the variance of the normalized instantaneous frequency, the standard deviation of the absolute value of the nonlinear component of the instantaneous phase of the zero-centered non-weak signal segment, and the standard deviation of the absolute value of the instantaneous phase;

[0021] Normalize the feature parameter data to its maximum and minimum values, and randomly shuffle the normalized feature data and its corresponding category labels to obtain the final digital modulation signal dataset;

[0022] The final digital modulation signal data set is divided into a training set and a test set. The feature data and its labels accounting for α1 are divided into the training set, and the remaining data is used as the test set.

[0023] Furthermore, the optimization of the support vector machine hyperparameters using the Angora rabbit search mechanism described in step 3 includes:

[0024] Step 3.1: Use the directed acyclic graph method to construct the support vector machine and set the objective function and constraints for hyperparameter optimization;

[0025] Step 32: Initialize the position of each rabbit in the Angora rabbit population, use the inverse of the support vector machine test set classification accuracy as the fitness function of the Angora rabbit search mechanism, calculate the fitness value, and obtain the initial optimal position of the Angora rabbit population; wherein the position of each Angora rabbit is a hyperparameter of the support vector machine;

[0026] Step 3. Energy contraction phase: Calculate the energy of each Angora rabbit and determine whether the rabbit will enter the detour foraging phase or the random hiding phase;

[0027] Step 3 and 4: Foraging phase: Angora rabbits obtain sufficient food by searching for grass in the territories of random individuals in the population.

[0028] Step 3.5: Random hiding stage: move closer to the rabbit with the best position to avoid being hunted by natural enemies;

[0029] Step 36: Introduce a random solution generation strategy to update the random position of Angora rabbit individuals;

[0030] Step 37: Calculate the fitness value of each Angora rabbit after the updated position, and update the position of the optimal solution;

[0031] Step 38: Determine whether the maximum number of iterations has been reached. If not, return to step 33 and continue iterating; otherwise, output the optimal position.

[0032] Furthermore, the objective function in step 31 is set as follows:

[0033]

[0034] In the formula, min means minimization; w ij is the weight vector of the hyperplane of class i and class j; C is the penalty factor; ζ ij represents the slack variable of the i-th and j-th support vector machines; L represents the total number of categories; Indicates the Slack variables for the i-th and j-th samples;

[0035] The constraints are set as follows:

[0036]

[0037] Where b ij is the bias of the hyperplane of class i and class j; A mapping function that maps data samples to high dimensions; Indicates the first samples, ω i and ω j Represents the category labels of category i and category j.

[0038] Furthermore, step 32 specifically includes: assuming that the number of rabbits in the Angora rabbit population is N, the maximum number of iterations is T, and the lower bound R = [R1, R2, ..., R D ], upper bound U=[U1,U2,...,U D ], D is the maximum dimension of the search; let the position of the nth rabbit at the tth iteration be Where t is the current iteration number, t∈{1,2,...,T}; let is the position of the nth Angora rabbit in the dth dimension at the tth iteration, then the fitness function is expressed as Where χ is the recognition accuracy of the modulation signal; the initialization of the d-dimensional position of the n-th rabbit in the first generation is (R d ,U d ), d=1,2,...,D,U d is the upper bound of the d-th dimension, R d is the lower bound of the d-th dimension;

[0039] After the population initialization is completed, the position of the first-generation Angora rabbit individuals is used as the hyperparameter of the support vector machine. The training set is input into the support vector machine model for training. After the training is completed, the test set is input into the support vector machine model to obtain the recognition accuracy, and then the initial fitness value of each Angora rabbit is obtained. The position of the Angora rabbit individual with the best fitness value in the initial population is set as the initial optimal position.

[0040] Furthermore, step 33 specifically includes: the energy of the rabbit is defined by the following expression:

[0041]

[0042] Where, E t is the energy of the prey at the tth iteration, is a random number between (0,1) selected at the tth iteration, e is the base of the natural logarithm; when |E t When |≥1, the rabbit enters the detour foraging stage, otherwise the rabbit enters the random hiding stage;

[0043] Steps 3 and 4 specifically include: The position update formula of the nth rabbit individual is:

[0044]

[0045] Where, is the walking length of the rabbit, where j=1,2,...,D, is the position of a random individual in the Angora rabbit population, where is a random integer between [1,N] and and is a random number between (0,1), is a random number that obeys the standard normal distribution, β1 and β2 are fixed exploration factors, exp() represents the exponential function with e as the base, The symbol for rounding up.

[0046] Furthermore, steps 3 and 5 specifically include: the position update formula of the nth Angora rabbit individual is:

[0047]

[0048] Where, represents the location of the burrow built by the nth rabbit at the tth iteration, is a random number between (0,1);

[0049] Step 36 specifically includes: the random solution generation formula for the nth Angora rabbit is:

[0050]

[0051] Where, is a random number between (0, 1), is the location of the random solution of the nth Angora rabbit;

[0052] According to the fitness function Calculate the fitness value of the nth Angora rabbit at a random position, and calculate the fitness value of the nth Angora rabbit after completing the position update like but

[0053] Furthermore, step 37 specifically includes: after completing the position update, the position of the nth Angora rabbit individual in the t+1th iteration is As the hyperparameter of the support vector machine model, the training set is input to train the support vector machine model. After the training is completed, the test set is input into the support vector machine model to obtain the recognition accuracy, and then the fitness value of each Angora rabbit individual in the t+1th iteration is obtained; the position of the individual with the best fitness value in the Angora rabbit population is set as the optimal position, and the optimal position of the Angora rabbit population in the t+1th iteration is recorded; if the fitness value of the optimal position of the Angora rabbit population in the t+1th iteration is better than the fitness value of the optimal position of the Angora rabbit population in the tth iteration, then the optimal position of the Angora rabbit population in the t+1th iteration is equal to the optimal position of the Angora rabbit population in the t+1th iteration; otherwise, the optimal position of the Angora rabbit population in the t+1th iteration is equal to the optimal position of the Angora rabbit population in the tth iteration.

[0054] According to another aspect of the present invention, a communication signal modulation identification system based on an Angora rabbit search mechanism is proposed, the system comprising:

[0055] a data acquisition module configured to construct a digital modulation signal data set;

[0056] A data preprocessing module is configured to preprocess the data in the digital modulation signal data set to obtain a training set and a test set;

[0057] A hyperparameter optimization solution module is configured to optimize and solve the hyperparameters of the support vector machine based on the training set and the test set using the Angora rabbit search mechanism, wherein the hyperparameters include a penalty factor and a Gaussian kernel;

[0058] a model training module configured to train a support vector machine with optimal hyperparameters using a training set;

[0059] The modulation identification module is configured to perform modulation identification on the communication signal to be tested using the trained support vector machine.

[0060] Compared with the prior art, the beneficial technical effects of the present invention are:

[0061] 1) The present invention designs an Angora rabbit search mechanism to select the hyperparameters of a support vector machine, which solves the problem that the hyperparameters of traditional support vector machines are difficult to select and avoids the overfitting problem of the support vector machine due to improper hyperparameter selection. This enables the present invention to have the advantages of high recognition accuracy and fast convergence speed in the field of communication signal modulation recognition, and can also achieve good recognition effects under low signal-to-noise ratios.

[0062] 2) Compared with existing modulation recognition methods, the designed Angora rabbit mechanism-evolved support vector machine has superior classification and generalization capabilities in complex modulation recognition environments. It can significantly improve the accuracy of communication signal modulation recognition under Gaussian noise at low signal-to-noise ratios, and is applicable to both high and low signal-to-noise ratio environments, and has broader application prospects and scenarios.

[0063] 3) The Angora rabbit search mechanism for support vector machine hyperparameter selection proposed in the present invention has better global convergence and convergence speed than other traditional intelligent optimization search mechanisms. It can achieve high convergence accuracy while converging quickly, and is not prone to falling into local optimal solutions and premature maturity. The evolved support vector machine designed for modulated signal recognition can achieve higher recognition accuracy under the same mixed signal-to-noise ratio and can be applied to solving various engineering application problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The present invention can be better understood by referring to the description given below in conjunction with the accompanying drawings, which together with the following detailed description are included in this specification and form a part of this specification, and are used to further illustrate the preferred embodiments of the present invention and explain the principles and advantages of the present invention.

[0065] Figure 1 This is a flowchart of a communication signal modulation identification method based on an Angora rabbit search mechanism according to an embodiment of the present invention.

[0066] Figure 2 This is another flow chart of a communication signal modulation identification method based on an Angora rabbit search mechanism according to an embodiment of the present invention.

[0067] Figure 3 4 is a convergence curve diagram of the Angora rabbit search mechanism and its comparison mechanism in an embodiment of the present invention.

[0068] Figure 4 It is a histogram of recognition accuracy of communication signal modulation recognition using the Angora rabbit search mechanism, comparison mechanism, and support vector machine automatic hyperparameter optimization in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] In order to enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below with reference to the accompanying drawings. Obviously, the described embodiments or examples are only some of the embodiments or examples of the present invention, and not all of them. Based on the embodiments or examples of the present invention, all other embodiments or examples obtained by those skilled in the art without creative work should fall within the scope of protection of the present invention.

[0070] The present invention addresses the shortcomings of existing communication signal modulation recognition methods in low recognition accuracy under Gaussian noise conditions and the difficulty in determining the optimal hyperparameters of a support vector machine (SVM) used as a modulation recognition classifier. By using a feature extraction method to obtain a dataset containing characteristic parameters, an Angora Rabbit search mechanism is designed to select SVM hyperparameters, obtain the optimal SVM hyperparameters, and use the SVM with the optimal hyperparameters as a classifier to efficiently identify communication signal modulation modes under Gaussian noise conditions. In the process of constructing a dataset of modulation signal characteristic parameters, the present invention provides a wide coverage of signal-to-noise ratios (SNRs), encompassing both low and high SNR conditions, with a low SNR interval, enabling better simulation of communication channel conditions in various engineering application scenarios. Regarding the search mechanism, the present invention designs a novel intelligent optimization search mechanism, the Angora Rabbit search mechanism, which combines a relatively fast convergence speed with high convergence accuracy and is less prone to falling into local optimal solutions. Compared to some traditional communication signal modulation recognition methods, the communication signal modulation recognition method designed in the present invention, based on the Angora Rabbit search mechanism, can obtain the optimal SVM hyperparameters under Gaussian noise conditions, thereby achieving a higher recognition rate.

[0071] The embodiment of the present invention proposes a communication signal modulation identification method based on the Angora rabbit search mechanism, such as Figure 1-2 As shown, the method includes:

[0072] Step 1: Construct a digital modulation signal dataset;

[0073] Step 2: Preprocess the data in the digital modulation signal dataset to obtain a training set and a test set;

[0074] Step 3: Based on the training set and the test set, the hyperparameters of the support vector machine are optimized using the Angora rabbit search mechanism, where the hyperparameters include a penalty factor and a Gaussian kernel;

[0075] Step 4: Use the training set to train the support vector machine with the optimal hyperparameters;

[0076] Step 5: Use the trained support vector machine to perform modulation recognition on the communication signal to be tested.

[0077] The method begins with step 1. In step 1, a digital modulation signal dataset is constructed.

[0078] According to an embodiment of the present invention, a unipolar sequence with M code elements is input, and the digital sequence is mapped into a pulse waveform through a unipolar non-return-to-zero code to obtain a digital baseband signal. A shaping filter is constructed, and the shaping filter uses a raised cosine roll-off function to shape the digital baseband signal, and the expression is: in is the sampling time, δ is the roll-off coefficient, The digital baseband signal after shaping filtering is modulated using different modulation methods to obtain modulated signals. Gaussian noise with different signal-to-noise ratios is inserted into the modulated signals to obtain a digital modulation signal dataset of the communication signal under the Gaussian noise background.

[0079] Then, step 2 is executed. In step 2, the data in the digital modulation signal dataset is preprocessed to obtain a training set and a test set.

[0080] According to an embodiment of the present invention, a digital modulation signal data set of a communication signal under a Gaussian noise background is normalized and segmented, and the communication signal of each modulation mode is divided into multiple data segments of equal length. Each data segment is set with a corresponding label to obtain a digital modulation signal preprocessing data set. Instantaneous feature parameters are extracted from the digital modulation signal preprocessing data set to obtain a feature parameter data set. First, a Hilbert transform is performed on the digital modulation signal preprocessing data set to extract the instantaneous amplitude A(i), instantaneous phase φ(i) and instantaneous frequency f(i) of the digital signal, where i is the data segment currently used for feature extraction. Based on the instantaneous information of the signal obtained under a Gaussian noise environment, multiple feature statistics of the instantaneous information of the digital modulation signal can be further extracted to obtain five feature parameters that can effectively resist Gaussian noise and improve the recognition rate when identifying the modulation signal.

[0081] Characteristic parameter 1: mean m of the instantaneous amplitude envelope a , Where A(i) is the instantaneous amplitude of the communication signal, is the number of sampling points.

[0082] Characteristic parameter 2: fourth-order moment compactness of normalized central instantaneous frequency Where E represents the mathematical expectation, f cn (i) is the normalized center instantaneous frequency,

[0083] Characteristic parameter 3: variance of normalized instantaneous frequency

[0084] Characteristic parameter 4: Standard deviation σ of the absolute value of the nonlinear component of the instantaneous phase of the zero-centered non-weak signal segment ap , Among them, C is The number of non-weak signal values in the sampled data. Non-weak signal means that the signal amplitude is greater than the amplitude judgment threshold level a t The signal A n (i) = A(i) / m a ,φ NL (i) is the nonlinear component of the instantaneous phase after zero-centering. When the carrier is completely synchronized, φNL (i) = φ(i) - φ0, where

[0085] Characteristic parameter 5: Absolute value standard deviation of instantaneous phase σ dp , in

[0086] By extracting the characteristic parameters, a dataset containing five characteristic parameters is obtained. The characteristic data in this dataset is normalized to their maximum and minimum values. The characteristic data and their corresponding category labels in the normalized dataset are randomly shuffled to obtain a digital modulation signal dataset of communication signals in a Gaussian noise background. After the shuffling is completed, the digital modulation signal dataset is divided into training and test sets. The characteristic data and their labels accounting for α1% are divided into the training set, and the remaining data is used as the test set. The training set and test set are denoted as I1′ and I2′, respectively.

[0087] Then, step three is performed. In step three, based on the training set and the test set, the hyperparameters of the support vector machine are optimized and solved using the Angora rabbit search mechanism. The hyperparameters include a penalty factor and a Gaussian kernel. Specifically, the following steps are performed:

[0088] Step 3.1: Use the directed acyclic graph method to construct the support vector machine and set the objective function and constraints for hyperparameter optimization;

[0089] Step 32: Initialize the position of each rabbit in the Angora rabbit population, use the inverse of the support vector machine test set classification accuracy as the fitness function of the Angora rabbit search mechanism, calculate the fitness value, and obtain the initial optimal position of the Angora rabbit population; wherein the position of each Angora rabbit is a hyperparameter of the support vector machine;

[0090] Step 3. Energy contraction phase: Calculate the energy of each Angora rabbit and determine whether the rabbit will enter the detour foraging phase or the random hiding phase;

[0091] Step 3 and 4: Foraging phase: Angora rabbits obtain sufficient food by searching for grass in the territories of random individuals in the population.

[0092] Step 3.5: Random hiding stage: move closer to the rabbit with the best position to avoid being hunted by natural enemies;

[0093] Step 36: Introduce a random solution generation strategy to update the random position of Angora rabbit individuals;

[0094] Step 37: Calculate the fitness value of each Angora rabbit after the updated position, and update the position of the optimal solution;

[0095] Step 38: Determine whether the maximum number of iterations has been reached. If not, return to step 33 and continue iterating; otherwise, output the optimal position.

[0096] According to an embodiment of the present invention, first, a model is constructed for training, testing, and classifying a data set using a support vector machine. Since the data set formed after feature extraction of the modulated signal is a multi-class data set, a directed acyclic graph method is used to construct a support vector machine model that can be used for multi-classification to classify the data set input to the support vector machine. Assuming that the number of features contained in the data set input to the support vector machine is k, the model is constructed. A nonlinear separable support vector machine transforms a k-class problem into The second-class problem is solved, and then the data set containing k categories is accurately classified.

[0097] During the classification process, the support vector machine is first trained. The training set I1′ and the two hyperparameters of the support vector machine, namely the penalty factor and the Gaussian kernel parameter, are input into the support vector machine for training. The hyperparameters of the support vector machine are the position of each Angora rabbit. The training process is equivalent to solving a quadratic programming problem. By introducing Lagrange multipliers, a constrained optimization equation is constructed. The constrained optimization equations for the i-th and j-th categories are The constraints are and in is the Lagrange multiplier of the i-th and j-th categories, C is the penalty factor, is the kernel function of the support vector machine, and K represents the kernel function symbol. Due to the lack of prior knowledge, the kernel function is selected to be mapped to the infinite-dimensional Gaussian kernel function, and its expression is in is the Euclidean distance between two data samples, exp() is the exponential function with the natural constant e as the base, and γ is the Gaussian kernel parameter. The training process is solved by The optimal solution To construct the optimal classification hyperplane of class i and class j, the solution process uses the sequence minimum optimization method to search the records And the slack variables ζ of the i-th and j-th support vector machines are ij Perform grid optimization within the range [r1, r2] and use the c-fold cross validation method for cross validation. Calculate the optimal weight vectors for class i and class j and optimal bias Then construct the decision function of the i-th and j-th categories Where sgn() is a symbolic function that satisfies any variable u.

[0098] By permuting and combining k categories, we can get There are three combinations. Each combination is trained according to the above process to obtain the support vector machine. The decision functions that adapt to different data category combinations are formed to complete the model construction of the directed acyclic graph support vector machine.

[0099] After the training is completed, the trained model is saved and the test set I2′ is input into the support vector machine for testing. Use discrete discrimination rules to determine the category to which the data sample belongs. If If the class is i, then it is class i; otherwise, it is class j. After the test set is classified, the class labels obtained by the support vector machine are compared with the correct class labels. If the two are the same, the classification is correct; otherwise, the classification is incorrect. The classification accuracy is the percentage of correctly classified data samples out of the total number of test set data samples.

[0100] For a data set with k data categories, the category to which the data sample belongs is ω=[ω1,ω2,...,ω k ], the objective function of the constructed support vector machine is:

[0101]

[0102] In the formula, min means minimization; w ij is the weight vector of the hyperplane of class i and class j; C is the penalty factor; ζ ij represents the slack variable of the i-th and j-th support vector machines; L represents the total number of categories; Indicates the Slack variables for the i-th and j-th samples;

[0103] And satisfy the following constraints:

[0104]

[0105] where w ij is the weight vector of the hyperplane of class i and class j, b ij is the bias of the hyperplane of class i and class j, A mapping function that maps data samples to high dimensions; Indicates the first samples, ω i and ω j Represents the category labels of category i and category j. The objective function is used to find the optimal classification hyperplane to achieve accurate classification of complex data sets.

[0106] Then, the position of each rabbit in the Angora rabbit population is initialized, and the inverse of the classification accuracy of the support vector machine test set is used as the fitness function of the Angora rabbit search mechanism. The fitness value is calculated to obtain the initial optimal position of the Angora rabbit population.

[0107] Specifically, let the number of rabbits in the Angora rabbit population be N, the maximum number of iterations be T, and the lower bound R = [R1, R2, ..., R D ], upper bound U=[U1,U2,...,U D ], D is the maximum dimension of the search. Let the position of the nth rabbit at the tth iteration be Where t is the current iteration number, t∈{1,2,...,T}. Since the recognition accuracy is used as the measure of the modulation signal recognition accuracy, the inverse of the recognition accuracy is selected as the fitness function of the Angora rabbit search mechanism. Let is the position of the nth Angora rabbit in the dth dimension at the tth iteration, then the fitness function can be expressed as Where χ is the recognition accuracy of the modulation signal.

[0108] Initialize the d-dimensional position of the n-th rabbit in the first generation to (R d ,U d ), d=1,2,...,D,U d is the upper bound of the d-th dimension, R d is the lower bound of the dth dimension. After population initialization, the positions of the first-generation angora rabbit individuals are used as hyperparameters of the support vector machine model described in step 2. The training set is input into the support vector machine model for training. After training, the test set is input into the support vector machine model to obtain the recognition accuracy, thereby obtaining the initial fitness value of each angora rabbit. The smaller the fitness value, the better the individual's position. The position of the angora rabbit with the best fitness value in the initial population is set as the initial optimal position.

[0109] Then, in the energy contraction phase, the energy of each angora rabbit is calculated to determine whether the rabbit will enter the detour foraging phase or the random hiding phase.

[0110] Specifically, to survive, Angora rabbits engage in two activities: foraging and avoiding predators. These activities are described by two phases: foraging for food and hiding randomly. The energy of the rabbit determines which phase it is in. The energy of the rabbit can be defined by the following expression: Where E t is the energy of the prey at the tth iteration, is a random number between (0,1) selected at the tth iteration, and e is the base of the natural logarithm. t When |≥1, the rabbit enters the detour foraging stage, otherwise the rabbit enters the random hiding stage.

[0111] Then, in the detour foraging stage, the Angora rabbit population conducts an exploration phase of detour foraging, in which the Angora rabbits obtain enough food by searching for grass in the territories of random individuals in the population.

[0112] Specifically, in the detour foraging stage, the rabbit searches for the food it needs in the distance and chooses to forage in the territory of a random individual in the population. The position update formula of the nth rabbit individual is: in is the walking length of the rabbit, where j=1,2,...,D, is the position of a random individual in the Angora rabbit population, where is a random integer between [1,N] and and is a random number between (0,1), is a random number that obeys the standard normal distribution, β1 and β2 are fixed exploration factors, exp() represents the exponential function with e as the base, The symbol for rounding up.

[0113] Then, in the random hiding stage, the Angora rabbit population conducts a random hiding development phase, moving closer to the rabbits with the best position to avoid being hunted by natural enemies.

[0114] Specifically, during the random hiding phase, each rabbit will build a cave within its own territory to avoid being preyed upon by its natural enemies. The position update formula for the nth Angora rabbit is: in represents the location of the burrow built by the nth rabbit at the tth iteration, and is a random number between (0,1).

[0115] Then, a random solution generation strategy is introduced to update the random positions of Angora rabbit individuals to increase the diversity of solutions within the population.

[0116] Specifically, after all Angora rabbits in the population have completed their position updates, a random solution generation strategy is introduced to simultaneously update the positions of each individual in the population to increase the diversity of candidate solutions within the population. The random solution generation formula for the nth Angora rabbit is: in is a random number between (0, 1), is the location of the random solution of the nth Angora rabbit. According to the fitness function Calculate the fitness value of the nth Angora rabbit at a random position, and calculate the fitness value of the nth Angora rabbit after completing the position update The fitness value is used to evaluate whether to use a random position to replace the updated position. but

[0117] Then, the fitness value of each Angora rabbit after the updated position is calculated, and the location of the optimal solution is updated.

[0118] Specifically, after completing the position update, the position of the nth Angora rabbit individual in the t+1th iteration is Set t as the hyperparameter of the support vector machine model. Input the training set to train the support vector machine model. After training, input the test set to the support vector machine model to obtain the recognition accuracy, and then obtain the fitness value of each angora rabbit individual in the t+1th iteration. The position of the individual with the best fitness value in the angora rabbit population is set as the optimal position, and the optimal position of the angora rabbit population in the t+1th iteration is recorded. If the fitness value of the optimal position of the angora rabbit population in the t+1th iteration is greater than the fitness value of the optimal position of the angora rabbit population in the tth iteration, then the optimal position of the angora rabbit population in the t+1th iteration is equal to the optimal position of the angora rabbit population in the t+1th iteration. Otherwise, the optimal position of the angora rabbit population in the t+1th iteration is equal to the optimal position of the angora rabbit population in the tth iteration.

[0119] Then, determine whether the maximum number of iterations T is reached. If not, set t=t+1 and return to step 33 to continue iterating; otherwise, output the optimal position.

[0120] Then, step 4 is performed, in which the support vector machine with the optimal hyperparameters is trained using the training set.

[0121] According to an embodiment of the present invention, the optimal position of the Angora rabbit population in the last generation is used as a hyperparameter of a support vector machine model, a training set is input to train the support vector machine model, and after the training is completed, a test set is input into the support vector machine model to obtain a final recognition accuracy rate, and the result of modulation recognition is output.

[0122] Finally, step five is executed. In step five, the trained support vector machine is used to perform modulation recognition on the communication signal to be tested.

[0123] The technical effects of the present invention are further verified through experiments.

[0124] The Angora rabbit search mechanism proposed in this invention is abbreviated as ARBO. The search mechanisms used for comparison are the Crow search mechanism and the Energy Valley search mechanism, abbreviated as CSA and EVO, respectively. Specific parameters of some models in the simulation experiment are set as follows.

[0125] The digital modulation signal types used are 2ASK, 4ASK, 2PSK, 4PSK, 8PSK, 16QAM, 32QAM, 64QAM, 2FSK and 4FSK. The method used in the present invention is not limited to these modulation methods. The parameters of the digital modulation signal are set as follows: Carrier frequency The carrier frequencies of 2FSK are set to and For 4FSK, in addition to the two carrier frequencies that are the same as those of 2FSK, there are and Sampling rate Symbol rate The sampling time T = 20 seconds, the number of sampling points per symbol is 25; the roll-off coefficient of the shaping filter is δ = 0.4; Gaussian noise is added at 0.5 dB intervals within a signal-to-noise ratio range of -30 dB to 20 dB. The feature parameter data set is divided into 30% training set and 70% test set, but the method of the present invention is not limited to this division method.

[0126] In order to comprehensively compare the performance of the three methods, the three methods are initialized identically, and the parameters of the Angora rabbit search mechanism are set to N = 20, T = 40, U d =100, R d =0.01, α1=0.3. For the relevant parameters of CSA, please refer to “A novel metaheuristic method for solving constrained engineering optimization problems: Crow search algorithm” published by Alireza Askarzadeh et al. in Computers and Structures (2016, Vol.106, pp.1-12); for the relevant parameters of EVO, please refer to “Energy valley optimizer: a novel metaheuristic algorithm for global and engineering” published by Mahdi Azizi et al. in Scientific Reports (2023, Vol.13, No.01). The population size, maximum number of iterations, upper bound and lower bound of the two comparison mechanisms are the same as those of the Angora rabbit search mechanism, which are N=20, T=40, U d =100, R d =0.01. Each method was run independently 30 times, and the average fitness of the 30 runs was taken to draw the fitness curve. The average modulation recognition accuracy of the 30 independent runs was taken as the final communication signal modulation recognition accuracy.

[0127] The fitness curve simulation results of the Angora rabbit search mechanism and its comparison mechanism under the same initial value conditions are as follows: Figure 3 As shown, the average value of 30 independent runs is plotted. The average recognition accuracy obtained by the 30 independent runs of the Angora rabbit search mechanism and its comparison mechanism for hyperparameter optimization and support vector machine automatic hyperparameter optimization is shown in Figure 4 shown.

[0128] Depend on Figure 3 It can be seen that under the same initial value conditions, compared with EVO and CSA, ARBO has the advantages of fast convergence speed and high convergence accuracy, which fully demonstrates that ARBO has better convergence.

[0129] Depend on Figure 4 It can be seen that in the case of the same digital modulation signal dataset under the background of Gaussian noise, the average accuracy of communication signal modulation recognition obtained by the method of using the intelligent optimization search mechanism to optimize the hyperparameters of the support vector machine is significantly better than the average recognition accuracy obtained by the automatic hyperparameter optimization of the support vector machine, and the average accuracy of modulation recognition obtained by using the Angora rabbit search mechanism to optimize the hyperparameters of the support vector machine is the best among the four methods. Therefore, it can be concluded that the communication signal modulation recognition method combining the Angora rabbit search mechanism with the support vector machine is a reasonable and accurate communication signal modulation recognition method. This method can obtain high-accuracy recognition results for multiple types of digital signals in the field of communication signal modulation recognition under Gaussian noise. Since the Gaussian noise added when constructing the digital modulation signal dataset of the communication signal under the background of Gaussian noise covers both high signal-to-noise ratio and low signal-to-noise ratio situations, this modulation recognition method is suitable for Gaussian noise environments with different signal-to-noise ratios, and has broad application prospects and strong robustness.

[0130] Another embodiment of the present invention provides a communication signal modulation identification system based on an Angora rabbit search mechanism, the system comprising:

[0131] a data acquisition module configured to construct a digital modulation signal data set;

[0132] A data preprocessing module is configured to preprocess the data in the digital modulation signal data set to obtain a training set and a test set;

[0133] A hyperparameter optimization solution module is configured to optimize and solve the hyperparameters of the support vector machine based on the training set and the test set using the Angora rabbit search mechanism, wherein the hyperparameters include a penalty factor and a Gaussian kernel;

[0134] a model training module configured to train a support vector machine with optimal hyperparameters using a training set;

[0135] The modulation identification module is configured to perform modulation identification on the communication signal to be tested using the trained support vector machine.

[0136] The functions of the communication signal modulation identification system based on the Angora rabbit search mechanism described in an embodiment of the present invention can be described by the aforementioned communication signal modulation identification method based on the Angora rabbit search mechanism. Therefore, for the parts not described in detail in the system embodiment, please refer to the above method embodiment and will not be repeated here.

[0137] Although the present invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of the foregoing description, will appreciate that other embodiments are contemplated within the scope of the invention thus described. This disclosure is intended to be illustrative rather than restrictive of the scope of the invention, which is defined by the appended claims.

Claims

1. A communication signal modulation identification method based on Angora rabbit search mechanism, characterized in that: include: Step 1: Construct a digital modulation signal dataset; Step 2: Preprocess the data in the digital modulation signal dataset to obtain a training set and a test set; Step 3: Based on the training set and the test set, the hyperparameters of the support vector machine are optimized and solved using the Angora rabbit search mechanism, wherein the hyperparameters include a penalty factor and a Gaussian kernel; including: Step 3. Build a support vector machine using a directed acyclic graph method and set the objective function and constraints for hyperparameter optimization; the objective function is set as follows: In the formula, min means minimization; w ij is the weight vector of the hyperplane of class i and class j; C is the penalty factor; ζ ij represents the slack variable of the i-th and j-th support vector machines; L represents the total number of categories; Indicates the Slack variables for the i-th and j-th samples; The constraints are set as follows: Where b ij is the bias of the hyperplane of class i and class j; A mapping function that maps data samples to high dimensions; Indicates the first samples, ω i and ω j Represents the category labels of category i and category j; Step 32: Initialize the position of each rabbit in the Angora rabbit population, use the inverse of the support vector machine test set classification accuracy as the fitness function of the Angora rabbit search mechanism, calculate the fitness value, and obtain the initial optimal position of the Angora rabbit population; wherein the position of each Angora rabbit is a hyperparameter of the support vector machine; Step 3. Energy contraction phase: Calculate the energy of each Angora rabbit and determine whether the rabbit will enter the detour foraging phase or the random hiding phase; Step 3 and 4: Foraging phase: Angora rabbits obtain sufficient food by searching for grass in the territories of random individuals in the population. Step 3.5: Random hiding stage: move closer to the rabbit with the best position to avoid being hunted by natural enemies; Step 36: Introduce a random solution generation strategy to update the random position of Angora rabbit individuals; Step 37: Calculate the fitness value of each Angora rabbit after the updated position, and update the position of the optimal solution; Step 38: Determine whether the maximum number of iterations has been reached. If not, return to step 33 and continue iterating. Otherwise, output the optimal position. Step 4: Use the training set to train the support vector machine with the optimal hyperparameters; Step 5: Use the trained support vector machine to perform modulation recognition on the communication signal to be tested.

2. A communication signal modulation identification method based on Angora rabbit search mechanism according to claim 1, characterized in that, The digital modulation signal dataset described in step 1 is constructed as follows: A unipolar sequence with M code elements is input, and the digital sequence is mapped into a pulse waveform using a unipolar non-return-to-zero code to obtain a digital baseband signal. Constructing a shaping filter, wherein the shaping filter adopts a raised cosine roll-off function to perform shaping processing on the digital baseband signal; Modulating the digital baseband signal after shaping and filtering using different modulation methods to obtain a modulated signal; Gaussian noise with different signal-to-noise ratios is inserted into the modulated signal to obtain a digital modulation signal dataset of the communication signal under the Gaussian noise background.

3. A communication signal modulation identification method based on Angora rabbit search mechanism according to claim 2, characterized in that, The preprocessing of the digital modulation signal data set in step 2 to obtain the training set and test set includes: The digital modulation signal dataset under Gaussian noise background is normalized and segmented, and the communication signal of each modulation mode is divided into multiple data segments of equal length. Each data segment is combined with the corresponding label to obtain a digital modulation signal preprocessing dataset. Performing a Hilbert transform on a digital modulation signal preprocessing data set to extract the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the digital signal, and then extracting instantaneous characteristic parameters to obtain a characteristic parameter data set; the instantaneous characteristic parameters include the mean of the instantaneous amplitude envelope, the fourth-order moment compactness of the normalized center instantaneous frequency, the variance of the normalized instantaneous frequency, the standard deviation of the absolute value of the nonlinear component of the instantaneous phase of the zero-centered non-weak signal segment, and the standard deviation of the absolute value of the instantaneous phase; Normalize the feature parameter data to its maximum and minimum values, and randomly shuffle the normalized feature data and its corresponding category labels to obtain the final digital modulation signal dataset; The final digital modulation signal data set is divided into a training set and a test set. The feature data and its labels accounting for α1 are divided into the training set, and the remaining data is used as the test set.

4. A communication signal modulation identification method based on Angora rabbit search mechanism according to claim 3, characterized in that: Step 32 specifically includes: assuming that the number of rabbits in the Angora rabbit population is N, the maximum number of iterations is T, and the lower bound R = [R1, R2, ..., R D ], upper bound U=[U1,U2,...,U D ], D is the maximum dimension of the search; let the position of the nth rabbit at the tth iteration be Where t is the current iteration number, t∈{1,2,...,T}; let is the position of the nth Angora rabbit in the dth dimension at the tth iteration, then the fitness function is expressed as Where χ is the recognition accuracy of the modulation signal; the initialization of the d-dimensional position of the n-th rabbit in the first generation is (R d ,U d ), d=1,2,...,D,U d is the upper bound of the d-th dimension, R d is the lower bound of the d-th dimension; After the population initialization is completed, the position of the first-generation Angora rabbit individuals is used as the hyperparameter of the support vector machine. The training set is input into the support vector machine model for training. After the training is completed, the test set is input into the support vector machine model to obtain the recognition accuracy, and then the initial fitness value of each Angora rabbit is obtained. The position of the Angora rabbit individual with the best fitness value in the initial population is set as the initial optimal position.

5. A communication signal modulation identification method based on Angora rabbit search mechanism according to claim 4, characterized in that: Step 3 specifically includes: The energy of the rabbit is defined by the following expression: Where, E t is the energy of the prey at the tth iteration, is a random number between (0,1) selected at the tth iteration, e is the base of the natural logarithm; when |E t When |≥1, the rabbit enters the detour foraging stage, otherwise the rabbit enters the random hiding stage; Steps 3 and 4 specifically include: The position update formula of the nth rabbit is: Where, is the walking length of the rabbit, where j=1,2,...,D, is the position of a random individual in the Angora rabbit population, where is a random integer between [1,N] and and is a random number between (0,1), is a random number that obeys the standard normal distribution, β1 and β2 are fixed exploration factors, exp() represents the exponential function with e as the base, The symbol for rounding up.

6. The communication signal modulation identification method based on the Angora rabbit search mechanism according to claim 5 is characterized in that: Steps 3 and 5 specifically include: the position update formula of the nth Angora rabbit individual is: Where, represents the location of the burrow built by the nth rabbit at the tth iteration, is a random number between (0,1); Step 36 specifically includes: the random solution generation formula for the nth Angora rabbit is: Where, is a random number between (0,1), is the location of the random solution of the nth Angora rabbit; According to the fitness function Calculate the fitness value of the nth Angora rabbit at a random position, and calculate the fitness value of the nth Angora rabbit after completing the position update like but 7. The communication signal modulation identification method based on the Angora rabbit search mechanism according to claim 6 is characterized in that: Step 37 specifically includes: after completing the position update, the position of the nth Angora rabbit individual in the t+1th iteration is As the hyperparameter of the support vector machine model, the training set is input to train the support vector machine model. After the training is completed, the test set is input into the support vector machine model to obtain the recognition accuracy, and then the fitness value of each Angora rabbit individual in the t+1th iteration is obtained; the position of the individual with the best fitness value in the Angora rabbit population is set as the optimal position, and the optimal position of the Angora rabbit population in the t+1th iteration is recorded; if the fitness value of the optimal position of the Angora rabbit population in the t+1th iteration is better than the fitness value of the optimal position of the Angora rabbit population in the tth iteration, then the optimal position of the Angora rabbit population in the t+1th iteration is equal to the optimal position of the Angora rabbit population in the t+1th iteration; otherwise, the optimal position of the Angora rabbit population in the t+1th iteration is equal to the optimal position of the Angora rabbit population in the tth iteration.

8. A communication signal modulation recognition system based on Angora rabbit search mechanism, characterized in that: include: a data acquisition module configured to construct a digital modulation signal data set; A data preprocessing module is configured to preprocess the data in the digital modulation signal data set to obtain a training set and a test set; A hyperparameter optimization solution module is configured to optimize and solve the hyperparameters of the support vector machine based on the training set and the test set using the Angora rabbit search mechanism, wherein the hyperparameters include a penalty factor and a Gaussian kernel; and includes: Step 3. Build a support vector machine using a directed acyclic graph method and set the objective function and constraints for hyperparameter optimization; the objective function is set as follows: In the formula, min means minimization; w ij is the weight vector of the hyperplane of class i and class j; C is the penalty factor; ζ ij represents the slack variable of the i-th and j-th support vector machines; L represents the total number of categories; Indicates the Slack variables for the i-th and j-th samples; The constraints are set as follows: Where b ij is the bias of the hyperplane of class i and class j; A mapping function that maps data samples to high dimensions; Indicates the first samples, ω i and ω j Represents the category labels of category i and category j; Step 32: Initialize the position of each rabbit in the Angora rabbit population, use the inverse of the support vector machine test set classification accuracy as the fitness function of the Angora rabbit search mechanism, calculate the fitness value, and obtain the initial optimal position of the Angora rabbit population; wherein the position of each Angora rabbit is a hyperparameter of the support vector machine; Step 3. Energy contraction phase: Calculate the energy of each Angora rabbit and determine whether the rabbit will enter the detour foraging phase or the random hiding phase; Step 3 and 4: Foraging phase: Angora rabbits obtain sufficient food by searching for grass in the territories of random individuals in the population. Step 3.5: Random hiding stage: move closer to the rabbit with the best position to avoid being hunted by natural enemies; Step 36: Introduce a random solution generation strategy to update the random position of Angora rabbit individuals; Step 37: Calculate the fitness value of each Angora rabbit after the updated position, and update the position of the optimal solution; Step 38: Determine whether the maximum number of iterations has been reached. If not, return to step 33 and continue iterating. Otherwise, output the optimal position. a model training module configured to train a support vector machine with optimal hyperparameters using a training set; The modulation identification module is configured to perform modulation identification on the communication signal to be tested using the trained support vector machine.

Images