A communication signal feature extraction method based on complex network theory

By converting communication signals into complex networks and extracting their statistical features, the problem of low modulation recognition accuracy in low signal-to-noise ratio environment and a small number of label samples is solved, and more efficient communication signal feature extraction and recognition is achieved.

CN116996349BActive Publication Date: 2024-12-03SOUTHWEST JIAOTONG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310700102.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-13
Publication Date
2024-12-03
Estimated Expiration
2043-06-13

AI Technical Summary

Technical Problem

In the prior art, under a low signal-to-noise ratio environment and a small number of label samples, the accuracy of communication signal modulation recognition is not high, and the existing feature extraction methods are susceptible to noise and their performance is degraded for different communication application scenarios.

Method used

Convert communication signals into complex network fields, and use binning, symbolization processing, building state transfer matrix and binary undirected complex networks to extract statistical features to construct feature vectors to realize feature extraction of signals.

Benefits of technology

In a low signal-to-noise ratio environment and a small number of label samples, the accuracy of communication signal modulation recognition is improved, and a new feature extraction method is provided, bringing new horizons to the field of communication technology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116996349B_ABST
    Figure CN116996349B_ABST
Patent Text Reader

Abstract

The present invention discloses a communication signal feature extraction method based on complex network theory, comprising S1, collecting communication signals in a noisy environment; S2, binning and symbolizing the communication signals to construct a signal symbol sequence; S3, constructing a state transfer matrix according to different states of the symbol sequence; S4, constructing a binary undirected complex network of the communication signal according to the state transfer matrix; S5, extracting statistical features of the binary undirected complex network of the communication signal, constructing a feature vector of the communication signal, and completing feature extraction of the communication signal. The communication signal features extracted by the scheme are used as input of a classifier, thereby realizing effective recognition of multiple types of communication modulation signals in a low signal-to-noise ratio environment and with a small number of labeled samples.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a communication signal feature extraction method based on complex network theory. Background Art

[0002] Nowadays, with the rapid development of communication technology, the number of mobile devices has increased dramatically and the communication transmission rate has greatly increased. The modulation types of communication systems are increasing, and the electromagnetic environment is becoming more and more complex. Extracting effective communication signal features to achieve modulation recognition has an important position and role in the fields of communication electronic countermeasures and radio signal management. It is an important technical means to realize spectrum resource monitoring and management. At the same time, it is an indispensable technical foundation in both military and civilian fields. In the current modulation recognition technology, the pattern recognition method based on feature extraction and most of the modulation recognition methods based on deep learning are based on the key technology of feature extraction. Feature-based recognition methods include instantaneous features, high-order cumulants, information entropy features, etc. The stronger the distinguishability of the extracted feature parameters, the better the recognition performance; but the above features all have some defects, such as instantaneous features are easily affected by noise, high-order cumulants are insensitive to noise, and some signals cannot be recognized by high-order cumulants. The modulation recognition method based on deep learning is a recognition method based on feature extraction, which relies on training a large number of sample signals to extract high-dimensional features contained in the signal, and then recognize and classify it. However, the performance of the deep learning modulation recognition algorithm will drop sharply when there are a small number of labeled samples and in different communication application scenarios.

[0003] Communication signals obviously belong to time series, and time series are inextricably linked to complex networks. Time series are converted to the field of complex networks for processing and analysis: the conversion method is equivalent to the construction of a new model, and the model is analyzed by parameters; the calculation of complex network indicators is equivalent to statistical feature analysis. Currently, there are algorithms that convert time series to complex networks, such as periodic algorithms, recursive network algorithms, and visibility graph algorithms. These algorithms have brought new research perspectives to signal processing. At the same time, these algorithms have been applied to multiple fields, such as medical brain wave processing, innovative drug development and efficacy judgment, financial stock fluctuations, meteorology, seismology, etc. However, there are few reports on the application of complex network theory in the field of communication technology. The present invention proposes a communication signal feature extraction method based on complex network theory to solve the problem that the modulation recognition accuracy of the communication signal features extracted by the prior art is not high in a low signal-to-noise ratio environment and a small number of labeled samples. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a communication signal feature extraction method based on complex network theory, which solves the problem that the communication signal features extracted by the prior art have low modulation recognition accuracy in a low signal-to-noise ratio environment and with a small number of label samples.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: to provide a communication signal feature extraction method based on complex network theory, comprising the following steps:

[0006] S1, collect communication signals in a noisy environment;

[0007] S2, binning and symbolizing the communication signal to construct a signal symbol sequence;

[0008] S3. construct a state transfer matrix according to different states of the symbol sequence;

[0009] S4. Construct a binary undirected complex network of communication signals based on the state transfer matrix;

[0010] S5. Extract the statistical characteristics of the binary undirected complex network of the communication signal, construct the feature vector of the communication signal, and complete the feature extraction of the communication signal.

[0011] Further: the step S2 comprises the following sub-steps:

[0012] S21, performing min-max normalization processing on the communication signal, and mapping the processed data to the interval [0, 1];

[0013] S22. Sort the processed data from small to large, and divide it into m equal parts according to the number of values. Each part is a bin, and its expression is:

[0014]

[0015] Among them, B is the bin, k is the bin number, B k is the kth bin;

[0016] S23. According to the binning situation, the communication signal is mapped to obtain a symbol sequence S, which is expressed as:

[0017] S=[s1,s2,...,s L ],s L ∈[1,m]

[0018] Among them, S is a symbol sequence with a length of L, S L is the Lth symbol in the sequence; x i is the i-th numerical point of the communication signal, if x i After binning, it belongs to B k , then S L =k.

[0019] Further: the expression of the state transfer matrix N in step S3 is:

[0020]

[0021] Among them, element b ij Indicates state s i The subsequent state is s j The number of times.

[0022] Further: the dimension of the adjacency matrix A of the communication signal binary undirected complex network G(V, E) in step S4 is the same as the dimension of the state transfer matrix N;

[0023] Among them, V and E are the node set and edge set of the complex network respectively, and the expressions of A and V are:

[0024]

[0025] V={v1,v2,...,v m}

[0026] Among them, a ij The value of b ij The relationship between the values ​​of is: if b ij If the value of is greater than or equal to the threshold, then a ij The value of a is 1; otherwise, a ij The value of a is 0; ij Represents node v i With node v j Connection relationship, when a ij When the value is 1, it means that node v i With node v j connected; when a ij When the value is 0, it means that node v i With node v j connected.

[0027] Further: the step S5 comprises the following sub-steps:

[0028] S51, calculate the average degree value K of the binary undirected complex network G of communication signals, and the calculation formula is:

[0029]

[0030] in, is the average degree of all nodes in G, k i For node v i The degree value of

[0031] S52, calculating the average separation degree L between nodes in the binary undirected complex network G of communication signals as the average distance, and the calculation formula is:

[0032]

[0033] Among them, L is the average value of the shortest path distance of all nodes in G, d ij For node v i With node v j The number of edges connecting these two nodes on the shortest path;

[0034] S53. Calculate the average clustering coefficient of the binary undirected complex network G of communication signals <c>, and its calculation formula is:

[0035]

[0036] in, <c>is the average clustering coefficient of G, C i is the clustering coefficient of the node; a ik Represents node v i With node v k Connection relationship, a jk Represents node v j With node v k Connectivity;

[0037] S54. Calculate the global clustering coefficient G of the binary undirected complex network G of communication signals total , and its calculation formula is:

[0038]

[0039] Among them, G total is the global clustering coefficient of G, G Δ is the number of closed triples in the network, G Λ represents the number of open triples in the network;

[0040] S55. Calculate the matching coefficient r of the binary undirected complex network G of the communication signal, and its expression is:

[0041]

[0042]

[0043]

[0044]

[0045] Among them, r is the matching coefficient of G, q k is the probability that a neighboring node with degree k is selected again for a randomly selected node in the network, σ q is the variance of the redundancy q distribution, P(j, k) is the probability that the degrees of the nodes at the two ends of a randomly selected edge in the network are j and k respectively, k min and k max are the minimum and maximum degrees of nodes in the network, respectively. M is the total number of edges in the network. m(j, k) is the number of edges between a node with degree j and a node with degree k. μ(j, k) is a coefficient function. If j = k, the value of μ(j, k) is 2, otherwise it is 1.

[0046] S56. The average degree, average distance, average clustering coefficient, global clustering coefficient and matching coefficient of the binary undirected complex network G of the communication signal are taken as the statistical characteristics of G. According to the statistical characteristics of G, a characteristic vector F of the communication signal is constructed, and its expression is as follows:

[0047]

[0048] Wherein, F is the characteristic vector of the communication signal.

[0049] The beneficial effects of the present invention are:

[0050] 1. The communication signal feature extraction method based on complex network theory provided by the present invention converts the communication signal into the complex network field, and can extract the feature vector used to distinguish the communication signal modulation mode, effectively complete the communication modulation identification, and introduce new methods and new perspectives into the field of communication technology;

[0051] 2. Since the present invention converts the communication signal into the field of complex networks, extracts multiple statistical features of the network, and obtains the characteristic vector of the signal, it overcomes the problem that the communication signal features extracted by the prior art have low modulation recognition accuracy in a low signal-to-noise ratio environment and with a small number of labeled samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of the flow of the feature extraction method based on complex network theory provided by the present invention.

[0053] Figure 2 Schematic diagram of signal conversion to complex network provided by the present invention.

[0054] Figure 3 This is a time domain waveform diagram of a 2ASK signal under noise-free conditions provided by the present invention.

[0055] Figure 4 The present invention provides a 2ASK signal time domain waveform diagram with a signal-to-noise ratio of 0dB.

[0056] Figure 5 A complex network diagram of a 2ASK signal with a signal-to-noise ratio of 0 dB provided by the present invention.

[0057] Figure 6 This is a time domain waveform diagram of a BPSK signal under noise-free conditions provided by the present invention.

[0058] Figure 7 The present invention provides a BPSK signal time domain waveform diagram with a signal-to-noise ratio of 0dB.

[0059] Figure 8 A complex network diagram of a BPSK signal with a signal-to-noise ratio of 0 dB provided by the present invention.

[0060] Fig. 9 This is a time domain waveform diagram of a 2FSK signal under noise-free conditions provided by the present invention.

[0061] Fig.10 The present invention provides a 2FSK signal with a signal-to-noise ratio of 0dB in the time domain.

[0062] Fig.11 A complex network diagram of a 2FSK signal with a signal-to-noise ratio of 0 dB provided by the present invention.

[0063] Fig.12 The T-SNE visualization diagram of the six types of signal feature vectors with a signal-to-noise ratio of -3dB provided by the present invention.

[0064] Fig.13 The present invention provides a confusion matrix for communication modulation recognition based on Linear-SVM under the condition of -5dB with 100 training samples of each category.

[0065] Fig.14 This is a comparison chart of the modulation recognition accuracy of the three classifiers under different signal-to-noise ratios provided by the present invention. DETAILED DESCRIPTION

[0066] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0067] like Figure 1 and Figure 2 As shown, in one embodiment of the present invention, a communication signal feature extraction method based on complex network theory is provided, comprising the following steps:

[0068] S1, collect communication signals in a noisy environment;

[0069] S2, binning and symbolizing the communication signal to construct a signal symbol sequence;

[0070] S3. construct a state transfer matrix according to different states of the symbol sequence;

[0071] S4. Construct a binary undirected complex network of communication signals based on the state transfer matrix;

[0072] S5. Extract the statistical characteristics of the binary undirected complex network of the communication signal, construct the feature vector of the communication signal, and complete the feature extraction of the communication signal.

[0073] In this embodiment, the communication signal in step S1 includes communication signals of different modulation modes, such as binary amplitude modulation signal 2ASK, binary phase shift keying signal BPSK, binary frequency shift keying signal 2FSK, orthogonal amplitude modulation signal QAM, octal orthogonal amplitude modulation signal 8QAM and hexadecimal orthogonal amplitude modulation signal 16QAM.

[0074] In this embodiment, step S2 includes the following sub-steps:

[0075] S21, performing min-max normalization processing on the communication signal, and mapping the processed data to the interval [0, 1];

[0076] S22. Sort the processed data from small to large, and divide it into m equal parts according to the number of values. Each part is a bin, and its expression is:

[0077]

[0078] Among them, B is the bin, k is the bin number, B k is the kth bin;

[0079] S23. According to the binning situation, the communication signal is mapped to obtain a symbol sequence S, which is expressed as:

[0080] S=[s1,s2,...,s L ],s L ∈[1,m]

[0081] Among them, S is a symbol sequence with a length of L, S L is the Lth symbol in the sequence; x i is the i-th numerical point of the communication signal, if x i After binning, it belongs to B k , then S L =k.

[0082] In this embodiment, the expression of the state transfer matrix N in step S3 is:

[0083]

[0084] Among them, element b ij Indicates state s i The subsequent state is s j The number of times.

[0085] In the present embodiment, the dimension of the adjacency matrix A of the communication signal binary undirected complex network G(V, E) in step S4 is the same as the dimension of the state transfer matrix N;

[0086] Among them, V and E are the node set and edge set of the complex network respectively, and the expressions of A and V are:

[0087]

[0088] V={v1,v2,…,v m }

[0089] Among them, a ij The value of b ij The relationship between the values ​​of is: if b ij If the value of is greater than or equal to the threshold, then a ij The value of a is 1; otherwise, a ij The value of a is 0; ij Represents node v i With node v j Connection relationship, when a ij When the value is 1, it means that node v i With node v j connected; when a ij When the value is 0, it means that node v i With node v j connected.

[0090] In this embodiment, step S5 includes the following sub-steps:

[0091] S51. Calculate the average degree of the binary undirected complex network G of the communication signal The calculation formula is:

[0092]

[0093] in, is the average degree of all nodes in G, k i For node v i The degree value of

[0094] S52, calculating the average separation degree L between nodes in the binary undirected complex network G of communication signals as the average distance, and the calculation formula is:

[0095]

[0096] Among them, L is the average value of the shortest path distance of all nodes in G, d ij For node v i With node v j The number of edges connecting these two nodes on the shortest path;

[0097] S53. Calculate the average clustering coefficient of the binary undirected complex network G of communication signals <c>, and its calculation formula is:

[0098]

[0099] in, <c>is the average clustering coefficient of G, C i is the clustering coefficient of the node; a ik Represents node v i With node v k Connection relationship, a jk Represents node v j With node v k Connectivity;

[0100] S54. Calculate the global clustering coefficient G of the binary undirected complex network G of communication signals total , and its calculation formula is:

[0101]

[0102] Among them, G total is the global clustering coefficient of G, G Δ is the number of closed triples in the network, G Λ represents the number of open triples in the network;

[0103] S55. Calculate the matching coefficient r of the binary undirected complex network G of the communication signal, and its expression is:

[0104]

[0105]

[0106]

[0107]

[0108] Among them, r is the matching coefficient of G, q k is the probability that a neighboring node with degree k is selected again for a randomly selected node in the network, σ q is the variance of the redundancy q distribution, P(j, k) is the probability that the degrees of the nodes at the two ends of a randomly selected edge in the network are j and k respectively, k min and k max are the minimum and maximum degrees of nodes in the network, respectively. M is the total number of edges in the network. m(j, k) is the number of edges between a node with degree j and a node with degree k. μ(j, k) is a coefficient function. If j = k, the value of μ(j, k) is 2, otherwise it is 1.

[0109] S56. The average degree, average distance, average clustering coefficient, global clustering coefficient and matching coefficient of the binary undirected complex network G of the communication signal are taken as the statistical characteristics of G. According to the statistical characteristics of G, a characteristic vector F of the communication signal is constructed, and its expression is as follows:

[0110]

[0111] Wherein, F is the characteristic vector of the communication signal.

[0112] In one embodiment of the present invention, a simulated communication signal is used, and the simulation conditions are as follows: the number of code elements is 500, the baud rate is 8000 Baud, and the carrier frequency of 2FSK is f c1 =400kHz,f c2 =400kHz, the carrier frequencies of other modulation methods are all f c =400kHz; sampling frequency f s =8MHz; Add additive white Gaussian noise to the communication signal.

[0113] In this embodiment, a communication signal with a signal-to-noise ratio of [-5, 5] dB (each type of modulation signal under different signal-to-noise ratios has 200 samples) is used, and a sample size ratio of 1:1 is set to construct a training set and a test set, and the training set and the test set are subjected to mean-variance normalization, and the characteristic vector of the communication signal is scaled to a distribution with a mean of 0 and a variance of 1. Three commonly used classifiers are selected: K-nearest neighbor classifier (KNN), Gaussian radial basis kernel function support vector machine (RBF-SVM), and linear kernel function support vector machine (Linear-SVM), and classification model training is performed, and a comparative experiment on recognition accuracy is performed on the modulation recognition classification task.

[0114] like Figure 3-Figure 11 As shown, in this embodiment, there are respectively a time domain waveform diagram of a 2ASK signal under noise-free conditions, a time domain waveform diagram of a 2ASK signal with a signal-to-noise ratio of 0dB, a complex network diagram of a 2ASK signal with a signal-to-noise ratio of 0dB, a time domain waveform diagram of a BPSK signal under noise-free conditions, a time domain waveform diagram of a BPSK signal with a signal-to-noise ratio of 0dB, a complex network diagram of a BPSK signal with a signal-to-noise ratio of 0dB, a time domain waveform diagram of a 2FSK signal under noise-free conditions, a time domain waveform diagram of a 2FSK signal with a signal-to-noise ratio of 0dB, and a complex network diagram of a 2FSK signal with a signal-to-noise ratio of 0dB;

[0115] The comparative experiment in this embodiment is shown in Table 1:

[0116] Table 1 Recognition accuracy of different classifiers

[0117]

[0118] From Table 6-1 and Fig.14 It can be seen that the classification effects of using the signal features extracted by the present invention to perform modulation recognition using different classifiers are slightly different, but under the condition that the signal-to-noise ratio is -3dB or above, the recognition accuracy of each classifier remains above 90%; in particular, the recognition effect of Linear-SVM is always higher than that of the other two classifiers;

[0119] like Fig.12 and Fig.13 As shown, in this embodiment, 10, 20, ..., 100 samples are selected from each category to form a training set (a total of 6 categories); the sample size of the test set is fixed at 600 (100 per category), and Linear-SVM is used as the classifier. The classification model is trained using training set samples with different signal-to-noise ratios, and a comparative experiment on the recognition accuracy is conducted on the modulation recognition classification task. The experimental results are shown in Table 2:

[0120] Table 2 Recognition accuracy of different sample numbers for each category

[0121]

[0122] From the data in 2, we can see that as the number of training samples decreases, the modulation recognition accuracy of Linear-SVM gradually decreases; but when the signal-to-noise ratio is -3dB and the number of training samples in each category is reduced to 10, the recognition accuracy of the classifier is always no less than 85%.

[0123] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.< / c> < / c> < / c> < / c>

Claims

1. A communication signal feature extraction method based on complex network theory, characterized in that: The following steps are included: S1, collect communication signals in a noisy environment; S2, binning and symbolizing the communication signal to construct a signal symbol sequence; The step S2 comprises the following sub-steps: S21, performing min-max normalization processing on the communication signal, and mapping the processed data to the interval [0, 1]; S22. Sort the processed data from small to large, and divide it into m equal parts according to the number of values, and each part is a bin, and its expression is: Among them, B is the bin, k is the bin number, B k is the kth bin; S23. According to the binning situation, the communication signal is mapped to obtain a symbol sequence S, which is expressed as: S=[s1,s2,...,s L ],s L ∈[1,m] Among them, S is a symbol sequence with a length of L, S L is the Lth symbol in the sequence; x i is the i-th numerical point of the communication signal, if x i After binning, it belongs to B k , then S L = k; S3. construct a state transfer matrix according to different states of the symbol sequence; S4. Construct a binary undirected complex network of communication signals based on the state transfer matrix; S5, extracting statistical features of the binary undirected complex network of the communication signal, constructing a feature vector of the communication signal, and completing feature extraction of the communication signal; The step S5 comprises the following sub-steps: S51, calculate the average degree value K of the binary undirected complex network G of communication signals, and the calculation formula is: Among them, K is the average degree of all nodes in G, k i For node v i The degree value of S52, calculating the average separation degree L between nodes in the binary undirected complex network G of communication signals as the average distance, and the calculation formula is: Among them, L is the average value of the shortest path distance of all nodes in G, d ij For node v i With node v j The number of edges connecting these two nodes on the shortest path; S53. Calculate the average clustering coefficient of the binary undirected complex network G of communication signals <c> , and its calculation formula is:< / c> in, <c>is the average clustering coefficient of G, C i is the clustering coefficient of the node; a ik Represents node v i With node v k Connection relationship, a jk Represents node v j With node v k Connectivity;< / c> S54. Calculate the global clustering coefficient G of the binary undirected complex network G of communication signals total , and its calculation formula is: Among them, G total is the global clustering coefficient of G, G Δ is the number of closed triples in the network, G Λ represents the number of open triples in the network; S55. Calculate the matching coefficient r of the binary undirected complex network G of the communication signal, and its expression is: Among them, r is the matching coefficient of G, q k is the probability that a neighboring node with degree k is selected again for a randomly selected node in the network, σ q is the variance of the redundancy q distribution, P(j, k) is the probability that the degrees of the nodes at the two ends of a randomly selected edge in the network are j and k respectively, k min and k max are the minimum and maximum degrees of nodes in the network, respectively. M is the total number of edges in the network. m(j, k) is the number of edges between a node with degree j and a node with degree k. μ(j, k) is a coefficient function. If j = k, the value of μ(j, k) is 2, otherwise it is 1. S56. The average degree, average distance, average clustering coefficient, global clustering coefficient and matching coefficient of the binary undirected complex network G of the communication signal are taken as the statistical characteristics of G. According to the statistical characteristics of G, a characteristic vector F of the communication signal is constructed, and its expression is as follows: Wherein, F is the characteristic vector of the communication signal.

2. The method for extracting communication signal features based on complex network theory according to claim 1, characterized in that: The expression of the state transfer matrix N in step S3 is: Among them, element b ij Indicates state s i The subsequent state is s j The number of times.

3. The communication signal feature extraction method based on complex network theory according to claim 2 is characterized in that: The dimension of the adjacency matrix A of the communication signal binary undirected complex network G(V, E) in step S4 is the same as the dimension of the state transfer matrix N; Among them, V and E are the node set and edge set of the complex network respectively, and the expressions of A and V are: V={v1,v2,…,v m } Among them, a ij The value of b ij The relationship between the values ​​of is: if b ij If the value of is greater than or equal to the threshold, then a ij The value of a is 1; otherwise, a ij The value of a is 0; ij Represents node v i With node v j Connection relationship, when a ij When the value is 1, it means that node v i With node v j connected; when a ij When the value is 0, it means that node v i Not with node v j connected.

Citation Information

Patent Citations

  • Wireless frame behavior feature extraction method and system based on MEMD decomposition

    CN113205076A

  • Alzheimer's disease recognition method based on brain imaging genome characteristics

    CN114596253A