Wireless signal interference detection method based on generative adversarial network

Abstract

The invention discloses a wireless signal interference detection method based on a generative adversarial network. The method comprises the following steps: S1, constructing a generative adversarial network model for wireless signal interference detection; s2, designing a dynamic network optimization mechanism based on the disturbance characteristics of the wireless signal data; s3, mapping interference characteristics of different domains to a unified characteristic space by adopting a cross-domain interference mode migration technology; s4, sparse features of the interference signals are extracted through the compressed sensing technology, and time domain and frequency domain distribution of the interference mode is restored through a reverse mapping algorithm; s5, on the basis of a collaborative optimization mechanism of generation and discrimination bidirectional feedback, discriminator feedback information is introduced into the generator to optimize the generated sample; and S6, adopting an adaptive data enhancement technology based on dynamic adversarial sample generation, expanding the training data set through a random disturbance interpolation and feature mixing strategy, and outputting an interference detection result. The method has the advantages of high adaptability, high detection precision and excellent generalization ability.

Description

Technical Field

[0001] The present invention relates to the technical field of signal detection, and in particular to a wireless signal interference detection method based on a generative adversarial network. Background Art

[0002] With the rapid development and widespread application of wireless communication technology, wireless signals play a vital role in modern communication systems. However, the impact of complex wireless signal interference on communication quality is becoming increasingly prominent, especially in dynamic and diverse wireless communication environments. The continuous evolution of various interference patterns makes interference detection an important technical challenge. The existence of wireless signal interference not only reduces the efficiency of the communication system, but also may cause signal interruption or information leakage, seriously affecting the reliability and security of wireless communication.

[0003] In the existing technology, traditional wireless signal interference detection methods usually rely on signal analysis models based on feature engineering or simple machine learning algorithms. These methods have significant limitations when facing complex wireless communication environments and diverse interference patterns, which are specifically reflected in the following aspects:

[0004] 1. Lack of adaptability to complex interference patterns: Traditional interference detection methods are usually based on static feature extraction or fixed model architecture, which makes it difficult to effectively capture dynamically changing and highly complex interference patterns, resulting in insufficient ability to identify new interference signals.

[0005] 2. Insufficient generalization ability: The detection models in the existing technology perform poorly in different wireless communication scenarios or cross-domain signals, and lack adaptability to different interference environments, resulting in limited robustness and reliability in practical applications.

[0006] 3. Low data utilization efficiency: Faced with massive wireless signal data, traditional methods lack efficient data processing and feature extraction mechanisms and are unable to fully mine key interference information in the data, resulting in reduced detection accuracy.

[0007] 4. Slow response speed: Due to reliance on manual intervention or static rules, traditional methods often have response delays when processing dynamic interference signals and cannot meet the requirements of real-time detection and rapid response.

[0008] 5. Weak interference characteristic characterization capability: Existing methods mostly use simple feature analysis and lack in-depth characterization and modeling of interference signal characteristics, making it impossible to accurately classify and comprehensively analyze interference patterns.

[0009] Therefore, how to provide a wireless signal interference detection method based on a generative adversarial network is a problem that those skilled in the art urgently need to solve. Summary of the invention

[0010] One object of the present invention is to propose a wireless signal interference detection method based on a generative adversarial network. The present invention makes full use of technologies such as a generative adversarial network model, cross-domain interference pattern migration, adaptive network optimization, sparse representation and inverse mapping, and describes in detail a technical solution for efficiently realizing complex interference pattern detection. The method has the advantages of strong adaptability, high detection accuracy and excellent generalization ability.

[0011] A wireless signal interference detection method based on a generative adversarial network according to an embodiment of the present invention includes the following steps:

[0012] S1. Construct a generative adversarial network model for wireless signal interference detection. The generator is used to generate signal data that simulates complex interference patterns, and the discriminator is used to evaluate the difference in interference characteristics between the generated signal and the real signal.

[0013] S2. Design a dynamic network optimization mechanism based on the disturbance characteristics of wireless signal data, and dynamically adjust the network structure through signal disturbance intensity, adaptive weight allocation and multi-layer feature fusion;

[0014] S3, using cross-domain interference pattern migration technology to map the interference characteristics of different domains to a unified feature space, and conduct adversarial training by using the signal distribution difference between the source domain and the target domain;

[0015] S4, based on the sparse representation and inverse mapping method of interference signals, the sparse features of interference signals are extracted through compressed sensing technology, and the time domain and frequency domain distribution of interference patterns are restored using the inverse mapping algorithm;

[0016] S5. Based on the collaborative optimization mechanism of bidirectional feedback between generation and discrimination, the discriminator feedback information is introduced into the generator to optimize the generated samples, and the output of the generator is used to update the detection capability of the discriminator;

[0017] S6. Adopt adaptive data enhancement technology based on dynamic adversarial sample generation, expand the training data set through random perturbation interpolation and feature mixing strategy, and output interference detection results.

[0018] Optionally, the S1 specifically includes:

[0019] S11. Design the generator network structure. The generator uses random noise input vector and dynamically extracted interference characteristic conditions As input, generate signal data that simulates a complex interference pattern:

[0020] x g =G(z,c;θ g );

[0021] Among them, G is the generator network, θ grepresents the parameters of the generator, the interference characteristic condition c is obtained from the original signal data through the adaptive feature extraction module, and c dynamically combines the frequency domain and time domain characteristics to enhance the interference adaptability of the generated signal;

[0022] S12. Design the discriminator network structure. The discriminator combines the multi-resolution feature discrimination mechanism, inputs the signal data x and the interference characteristic condition c, and outputs the probability that the signal belongs to the real signal:

[0023] D(x,c;θ d )=σ(f(x,c;α,θ d ));

[0024] Among them, σ is the activation function, the signal perturbation sensitivity weight introduced by the discrimination mechanism is α, f is the weighted feature mapping function, θ d are the parameters of the discriminator;

[0025] S13. Define the generative adversarial loss function and expand the generator optimization objective to interference feature consistency evaluation:

[0026]

[0027] Among them, λ is the weight factor, L c represents the consistency loss function of the generated signal interference characteristics;

[0028] S14. Optimize the generator and discriminator through a dynamic parameter update mechanism, and use a multi-batch training strategy to adjust θ in real time g and θ d , so that the signal generated by the generator gradually approaches the interference distribution of the real signal, and the error of the discriminator converges to the specified range.

[0029] Optionally, the S2 specifically includes:

[0030] S21. Based on the disturbance characteristics of wireless signal data, a multi-scale disturbance analysis model is constructed. By analyzing the multiple disturbance patterns of the signal in the time domain, frequency domain and amplitude dimensions, a characteristic mapping matrix M is generated. perturb , where M perturb The elements of represent the weighted associations between different disturbance characteristics;

[0031] S22, Adaptive network optimization algorithm driven by disturbance characteristics, dynamically calculates signal disturbance intensity Δ and disturbance gradient change rate The perturbation-driven weight update rule is:

[0032]

[0033] in, is the weight of the previous round, η is the learning rate, γ and κ represent the influence factors of perturbation intensity and gradient change on weight update respectively;

[0034] S23, based on the characteristic mapping matrix M perturb Extract multi-layer perturbation features {F1,F2,...,F k}, combined with the dynamic fusion strategy to define the overall feature representation F c :

[0035]

[0036] Among them, β i is the dynamically adjusted feature weight factor, T(F i ) represents the feature transformation function;

[0037] S24. In the process of network optimization, the disturbance-driven mechanism is used to adjust the network topology structure. By introducing a dynamic connection module, the connection density between network layers is optimized, nodes are dynamically increased or decreased, and an optimized network architecture with adaptive capabilities is constructed.

[0038] S25. The model training of the generative adversarial network is completed through the adaptive network optimization mechanism driven by the perturbation characteristics. The perturbation-driven feature module is embedded in the generator, and the feature-sensitive evaluation unit is introduced in the discriminator to keep the generated signal consistent with the real signal.

[0039] Optionally, the S3 specifically includes:

[0040] S31, extract the characteristic distribution of source domain signal and target domain signal, and define the source domain signal distribution as P s (x s ), the target domain signal distribution is P t (x t ), and the two domain signals are mapped to a unified feature space through the feature extraction function f(·) to obtain the feature representation Z s and Z t :

[0041] Z s =f(x s θ f );

[0042] Z t =f(x t θ f );

[0043] Among them, θ f Parameters of the feature extraction function;

[0044] S32, based on the cross-domain signal distribution difference measurement function D(P s ,P t), and optimize the feature extraction function f(·) through adversarial training to minimize the difference between the source domain and target domain feature distribution:

[0045]

[0046] Among them, N s and N t are the number of samples in the source domain and the target domain, μ s and μ t are the means of the source domain and target domain features respectively;

[0047] S33, build a domain discriminator network, input feature representation Z, output the probability distribution of domain categories, and optimize the discriminator network parameters θ through adversarial training d :

[0048]

[0049] Where D represents the domain discriminator function;

[0050] S34, optimize the parameter θ of the feature extraction function f(·) f , minimizing the difference in cross-domain feature distribution and the inconsistency of domain discriminator output results:

[0051]

[0052] Among them, λ is a regulation factor, balancing the influence of distribution difference measurement and adversarial training;

[0053] S35. Complete cross-domain feature space mapping through adversarial training process, so that the source domain and target domain feature distributions achieve minimum differentiation in a unified feature space, and generate an optimized model that meets the interference characteristics requirements of the target domain.

[0054] Optionally, the S4 specifically includes:

[0055] S41. Construct a signal processing model based on compressed sensing to project the input interference signal x into a low-dimensional sparse space:

[0056] y=Φx+;

[0057] Among them, y is the compressed signal, Φ is the compressed sensing matrix, and is the noise interference component;

[0058] S42. Extracting sparse coefficients of the signal using adaptive sparse representation technology:

[0059] x = Ψ·θ;

[0060] Among them, Ψ is the sparse basis matrix, θ is the sparse coefficient;

[0061] S43, jointly optimize the sparse coefficient θ by minimizing the signal reconstruction error and the sparse constraint:

[0062]

[0063] Among them, λ is the regularization parameter, which controls the balance between sparsity and reconstruction error;

[0064] S44, based on the optimized sparse coefficient θ, reverse mapping is performed to restore the high-dimensional distribution characteristics of the signal and construct a restored signal And extract the time domain disturbance pattern and frequency domain interference characteristics;

[0065] S45. Combine the multi-scale feature extraction method, dynamically adjust the feature fusion weight, and generate the interference pattern characteristic model M interf , for time domain characteristics and frequency domain characteristics Conduct comprehensive modeling:

[0066]

[0067] Among them, g is the characteristic mapping function, θ m are model parameters;

[0068] S46, based on the constructed interference pattern characteristic model M interf , calculate the multi-dimensional characteristic indicators of the signal, including the time domain disturbance intensity distribution, frequency domain energy density distribution and cross-domain correlation characteristics, through the characteristic model parameter θ m The optimization output is a complete description of the interference pattern, and the interference pattern characterization result is converted into a standardized feature vector.

[0069] Optionally, the S5 specifically includes:

[0070] S51, construct the feedback-driven optimization mechanism of the generator, and receive the confidence output D(G(z,c;θ) of the generated sample by the discriminator in real time through the dynamic feedback module g );θ d ), the characteristic distribution of the samples generated by the generator is optimized by weighted fusion of the discriminator feedback information. The optimization goal of the generator is:

[0071]

[0072] Where f(·) is the signal feature extraction function, α is the feedback adjustment coefficient, and dynamically adjusts the contribution of the discriminator feedback to the optimization of the generated samples;

[0073] S52, embed the generated sample feature enhancement module in the discriminator, evaluate the feature difference between the generated sample and the real signal through the feature consistency measurement function Δ(G(x), x), and optimize the discriminator's ability to distinguish signal distribution based on this:

[0074]

[0075] Among them, β is a feature consistency measurement factor, which strengthens the sensitivity of the discriminator to the differences in the features of the generated samples;

[0076] S53. A two-way optimization framework for generation and discrimination is established. The generator dynamically adjusts the generated samples to make their characteristic distribution close to the real signal. At the same time, the discriminator uses the generated samples to improve the recognition accuracy of complex interference patterns:

[0077]

[0078] Among them, λ is the importance balance factor optimized by the generator and the discriminator;

[0079] S54, adopting a collaborative training mechanism based on adversarial dynamic weights, dynamically adjusting the training weights of the generator and the discriminator through round-by-round iterations, so that the interference signal samples generated by the generator gradually approach the real signal distribution in the multi-dimensional feature space;

[0080] S55. After the collaborative optimization is completed, the feature generalization ability of the generator is tested so that the signal samples generated by it maintain a consistent distribution in different interference scenarios, and the discriminator is comprehensively evaluated so that it can accurately detect different types of interference patterns;

[0081] S56. Use the optimized generator to generate diverse high-complexity interference signal samples, and use the optimized discriminator to accurately classify unknown interference characteristics to achieve source tracing analysis and real-time detection of interference patterns.

[0082] Optionally, the S6 specifically includes:

[0083] S61. Construct a dynamic adversarial sample generation module to generate diverse interference samples through an adaptive generation method based on perturbation interpolation:

[0084] x aug =αx i +(1-α)x j +;

[0085] Among them, x i and x j are signal samples from different sources, α is a dynamically assigned weight factor, which is adjusted according to the characteristic distribution of the input samples, and is the introduced random noise disturbance;

[0086] S62, based on the adaptive enhancement strategy of multi-feature mixing, the multi-dimensional features of the signal sample are mapped to a unified enhanced feature space, and the feature fusion process is dynamically adjusted through the nonlinear feature transformation function g(·) and the characteristic adaptive weight β to generate the enhanced sample feature F aug ;

[0087] S63. Combine adversarial samples, mixed feature samples and real signal samples to construct an adaptive enhanced dataset D aug ,The construction of the enhanced dataset is based on a hierarchical feature coverage strategy, which makes the generated samples widely distributed in the multi-dimensional feature space:

[0088] D aug =D real ∪D gen ∪D mix ;

[0089] Among them, D real is the real signal sample set, D gen is the generated adversarial sample set, D mix is a mixed feature sample set;

[0090] S64. Using adaptive enhancement dataset D aug The interference detection model is trained with the optimization goal of minimizing the classification error of the enhanced data set, and the model's adaptability to samples of different categories is adjusted through a dynamic loss function;

[0091] S65. Based on the trained interference detection model, the input signal is classified into interference patterns in real time, and a multi-dimensional characteristic description of the interference signal is output, including the probability distribution of the interference type, the signal characteristic index and its time-frequency domain distribution information.

[0092] The beneficial effects of the present invention are:

[0093] (1) The present invention combines generative adversarial networks, dynamic adversarial sample generation, adaptive network optimization, and cross-domain interference feature migration technology to achieve deep learning and dynamic adaptation of complex wireless signal interference patterns, significantly improving the system's ability to detect dynamically changing and highly complex interference signals, especially for the accurate identification and classification of new interference patterns. By integrating sparse representation and inverse mapping methods, it can efficiently extract key interference features from massive signal data and adapt to diverse wireless communication environments.

[0094] (2) The present invention optimizes the training efficiency of the interference detection model by constructing an adaptive enhanced data set and combining adversarial sample generation with feature mixing strategies, significantly improving the generalization ability and robustness of the detection model for different interference scenarios, and meeting the real-time detection requirements in complex wireless communication environments.

[0095] (3) The present invention uses a collaborative optimization mechanism of bidirectional feedback between generation and discrimination to dynamically adjust the parameters of the generator and discriminator, thereby improving the detection accuracy and discrimination ability of complex interference signals. Combined with the multi-dimensional deep characterization of signal interference characteristics, a comprehensive interference pattern characteristic analysis is provided, enabling the system to efficiently identify and classify different types of interference signals, while achieving traceability analysis and real-time management of interference patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0097] Figure 1 This is an overall framework diagram of a wireless signal interference detection method based on a generative adversarial network proposed in the present invention. DETAILED DESCRIPTION

[0098] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0099] refer to Figure 1 , a wireless signal interference detection method based on a generative adversarial network comprises the following steps:

[0100] S1. Construct a generative adversarial network model for wireless signal interference detection. The generator is used to generate signal data that simulates complex interference patterns, and the discriminator is used to evaluate the difference in interference characteristics between the generated signal and the real signal.

[0101] In this implementation, S1 specifically includes:

[0102] S11. Design the generator network structure. The generator uses random noise input vector and dynamically extracted interference characteristic conditions As input, generate signal data that simulates a complex interference pattern:

[0103] x g =G(z,c;θ g );

[0104] Among them, G is the generator network, θ g represents the parameters of the generator, the interference characteristic condition c is obtained from the original signal data through the adaptive feature extraction module, and c dynamically combines the frequency domain and time domain characteristics to enhance the interference adaptability of the generated signal;

[0105] S12. Design the discriminator network structure. The discriminator combines the multi-resolution feature discrimination mechanism, inputs the signal data x and the interference characteristic condition c, and outputs the probability that the signal belongs to the real signal:

[0106] D(x,c;θ d )=σ(f(x,c;α,θ d ));

[0107] Among them, σ is the activation function, the signal perturbation sensitivity weight introduced by the discrimination mechanism is α, f is the weighted feature mapping function, θ d are the parameters of the discriminator;

[0108] S13. Define the generative adversarial loss function and expand the generator optimization objective to interference feature consistency evaluation:

[0109]

[0110] Among them, λ is the weight factor, L c represents the consistency loss function of the generated signal interference characteristics;

[0111] S14. Optimize the generator and discriminator through a dynamic parameter update mechanism, and use a multi-batch training strategy to adjust θ in real time g and θ d , so that the signal generated by the generator gradually approaches the interference distribution of the real signal, and the error of the discriminator converges to the specified range.

[0112] S2. Design a dynamic network optimization mechanism based on the disturbance characteristics of wireless signal data, and dynamically adjust the network structure through signal disturbance intensity, adaptive weight allocation and multi-layer feature fusion;

[0113] In this implementation, S2 specifically includes:

[0114] S21. Based on the disturbance characteristics of wireless signal data, a multi-scale disturbance analysis model is constructed. By analyzing the multiple disturbance patterns of the signal in the time domain, frequency domain and amplitude dimensions, a characteristic mapping matrix M is generated. perturb , where M perturb The elements of represent the weighted associations between different disturbance characteristics;

[0115] S22, Adaptive network optimization algorithm driven by disturbance characteristics, dynamically calculates signal disturbance intensity Δ and disturbance gradient change rate The perturbation-driven weight update rule is:

[0116]

[0117] in, is the weight of the previous round, η is the learning rate, γ and κ represent the influence factors of perturbation intensity and gradient change on weight update respectively;

[0118] S23, based on the characteristic mapping matrix M perturb Extract multi-layer perturbation features {F1,F2,...,F k}, combined with the dynamic fusion strategy to define the overall feature representation F c :

[0119]

[0120] Among them, β i is the dynamically adjusted feature weight factor, T(F i ) represents the feature transformation function;

[0121] S24. In the process of network optimization, the disturbance-driven mechanism is used to adjust the network topology structure. By introducing a dynamic connection module, the connection density between network layers is optimized, nodes are dynamically increased or decreased, and an optimized network architecture with adaptive capabilities is constructed.

[0122] S25. The model training of the generative adversarial network is completed through the adaptive network optimization mechanism driven by the perturbation characteristics. The perturbation-driven feature module is embedded in the generator, and the feature-sensitive evaluation unit is introduced in the discriminator to keep the generated signal consistent with the real signal.

[0123] S3, using cross-domain interference pattern migration technology to map the interference characteristics of different domains to a unified feature space, and conduct adversarial training by using the signal distribution difference between the source domain and the target domain;

[0124] In this implementation, S3 specifically includes:

[0125] S31, extract the characteristic distribution of source domain signal and target domain signal, and define the source domain signal distribution as P s (x s ), the target domain signal distribution is P t (x t ), and the two domain signals are mapped to a unified feature space through the feature extraction function f(·) to obtain the feature representation Z s and Z t :

[0126] Z s =f(x s θ f );

[0127] Z t =f(x t θ f );

[0128] Among them, θ fParameters of the feature extraction function;

[0129] S32, based on the cross-domain signal distribution difference measurement function D(P s ,P t ), and optimize the feature extraction function f(·) through adversarial training to minimize the difference between the source domain and target domain feature distribution:

[0130]

[0131] Among them, N s and N t are the number of samples in the source domain and the target domain, μ s and μ t are the means of the source domain and target domain features respectively;

[0132] S33, build a domain discriminator network, input feature representation Z, output the probability distribution of domain categories, and optimize the discriminator network parameters θ through adversarial training d :

[0133]

[0134] Where D represents the domain discriminator function;

[0135] S34, optimize the parameter θ of the feature extraction function f(·) f , minimizing the difference in cross-domain feature distribution and the inconsistency of domain discriminator output results:

[0136]

[0137] Among them, λ is a regulation factor, balancing the influence of distribution difference measurement and adversarial training;

[0138] S35. Complete cross-domain feature space mapping through adversarial training process, so that the source domain and target domain feature distributions achieve minimum differentiation in a unified feature space, and generate an optimized model that meets the interference characteristics requirements of the target domain.

[0139] S4, based on the sparse representation and inverse mapping method of interference signals, the sparse features of interference signals are extracted through compressed sensing technology, and the time domain and frequency domain distribution of interference patterns are restored using the inverse mapping algorithm;

[0140] In this implementation, S4 specifically includes:

[0141] S41. Construct a signal processing model based on compressed sensing to project the input interference signal x into a low-dimensional sparse space:

[0142] y=Φx+;

[0143] Among them, y is the compressed signal, Φ is the compressed sensing matrix, and is the noise interference component;

[0144] S42. Extracting sparse coefficients of the signal using adaptive sparse representation technology:

[0145] x = Ψ·θ;

[0146] Among them, Ψ is the sparse basis matrix, θ is the sparse coefficient;

[0147] S43, jointly optimize the sparse coefficient θ by minimizing the signal reconstruction error and the sparse constraint:

[0148]

[0149] Among them, λ is the regularization parameter, which controls the balance between sparsity and reconstruction error;

[0150] S44, based on the optimized sparse coefficient θ, reverse mapping is performed to restore the high-dimensional distribution characteristics of the signal and construct a restored signal And extract the time domain disturbance pattern and frequency domain interference characteristics;

[0151] S45. Combine the multi-scale feature extraction method, dynamically adjust the feature fusion weight, and generate the interference pattern characteristic model M interf , for time domain characteristics and frequency domain characteristics Conduct comprehensive modeling:

[0152]

[0153] Among them, g is the characteristic mapping function, θ m are model parameters;

[0154] S46, based on the constructed interference pattern characteristic model M interf , calculate the multi-dimensional characteristic indicators of the signal, including the time domain disturbance intensity distribution, frequency domain energy density distribution and cross-domain correlation characteristics, through the characteristic model parameter θ m The optimization output is a complete description of the interference pattern, and the interference pattern characterization result is converted into a standardized feature vector.

[0155] S5. Based on the collaborative optimization mechanism of bidirectional feedback between generation and discrimination, the discriminator feedback information is introduced into the generator to optimize the generated samples, and the output of the generator is used to update the detection capability of the discriminator;

[0156] In this implementation, S5 specifically includes:

[0157] S51, construct the feedback-driven optimization mechanism of the generator, and receive the confidence output D(G(z,c;θ) of the generated sample by the discriminator in real time through the dynamic feedback module g );θ d), the characteristic distribution of the samples generated by the generator is optimized by weighted fusion of the discriminator feedback information. The optimization goal of the generator is:

[0158]

[0159] Where f(·) is the signal feature extraction function, α is the feedback adjustment coefficient, and dynamically adjusts the contribution of the discriminator feedback to the optimization of the generated samples;

[0160] S52, embed the generated sample feature enhancement module in the discriminator, evaluate the feature difference between the generated sample and the real signal through the feature consistency measurement function Δ(G(x), x), and optimize the discriminator's ability to distinguish signal distribution based on this:

[0161]

[0162] Among them, β is a feature consistency measurement factor, which strengthens the sensitivity of the discriminator to the differences in the features of the generated samples;

[0163] S53. A two-way optimization framework for generation and discrimination is established. The generator dynamically adjusts the generated samples to make their characteristic distribution close to the real signal. At the same time, the discriminator uses the generated samples to improve the recognition accuracy of complex interference patterns:

[0164]

[0165] Among them, λ is the importance balance factor optimized by the generator and the discriminator;

[0166] S54, adopting a collaborative training mechanism based on adversarial dynamic weights, dynamically adjusting the training weights of the generator and the discriminator through round-by-round iterations, so that the interference signal samples generated by the generator gradually approach the real signal distribution in the multi-dimensional feature space;

[0167] S55. After the collaborative optimization is completed, the feature generalization ability of the generator is tested so that the signal samples generated by it maintain a consistent distribution in different interference scenarios, and the discriminator is comprehensively evaluated so that it can accurately detect different types of interference patterns;

[0168] S56. Use the optimized generator to generate diverse high-complexity interference signal samples, and use the optimized discriminator to accurately classify unknown interference characteristics to achieve source tracing analysis and real-time detection of interference patterns.

[0169] S6. Adopt adaptive data enhancement technology based on dynamic adversarial sample generation, expand the training data set through random perturbation interpolation and feature mixing strategy, and output interference detection results.

[0170] In this implementation, S6 specifically includes:

[0171] S61. Construct a dynamic adversarial sample generation module to generate diverse interference samples through an adaptive generation method based on perturbation interpolation:

[0172]

[0173] Among them, x i and x j are signal samples from different sources, α is a dynamically assigned weight factor, which is adjusted according to the characteristic distribution of the input samples, and is the introduced random noise disturbance;

[0174] S62, based on the adaptive enhancement strategy of multi-feature mixing, the multi-dimensional features of the signal sample are mapped to a unified enhanced feature space, and the feature fusion process is dynamically adjusted through the nonlinear feature transformation function g(·) and the characteristic adaptive weight β to generate the enhanced sample feature F aug ;

[0175] S63. Combine adversarial samples, mixed feature samples and real signal samples to construct an adaptive enhanced dataset D aug ,The construction of the enhanced dataset is based on a hierarchical feature coverage strategy, which makes the generated samples widely distributed in the multi-dimensional feature space:

[0176] D aug =D real ∪D gen ∪D mix ;

[0177] Among them, D real is the real signal sample set, D gen is the generated adversarial sample set, D mix is a mixed feature sample set;

[0178] S64. Using adaptive enhancement dataset D aug The interference detection model is trained with the optimization goal of minimizing the classification error of the enhanced data set, and the model's adaptability to samples of different categories is adjusted through a dynamic loss function;

[0179] S65. Based on the trained interference detection model, the input signal is classified into interference patterns in real time, and a multi-dimensional characteristic description of the interference signal is output, including the probability distribution of the interference type, the signal characteristic index and its time-frequency domain distribution information.

[0180] Embodiment 1:

[0181] In order to verify the feasibility of the present invention, the present invention is applied to a public wireless communication network in a large city. The network covers multiple important commercial areas, including high-traffic transportation hubs, commercial centers and residential areas. Due to the complex communication environment and the large number of users, operators have found frequent communication interruptions, especially during peak hours, when the base station signal quality has significantly decreased, and it is suspected that there are complex signal interference problems. In order to improve the reliability and anti-interference capability of wireless communication, the operator decided to deploy the wireless signal interference detection method based on the generative adversarial network of the present invention.

[0182] The system first receives and processes real-time signal data from five major base stations in the network. These signal data are preprocessed, including denoising, normalization, and interference feature annotation. The generator module of the generative adversarial network generates a batch of simulated complex interference samples through dynamic perturbation interpolation, which covers possible high-complexity interference patterns. At the same time, the discriminator receives the output of the generator and compares it with the real signal, optimizing the model parameters in real time to ensure that the generated samples are closer to the real interference pattern.

[0183] In actual operation, the system detected abnormal frequency fluctuations at a certain base station during non-peak hours at night. The characteristic distribution of these abnormal signals is significantly different from that of normal signals, which is manifested as abnormal concentration of energy in the high-frequency band. Through cross-domain feature migration technology, the system maps the characteristics of abnormal signals to a unified feature space, and combines it with a dynamic adversarial sample generation module to accurately restore the complex characteristics of the interference signal. The system further uses sparse representation and inverse mapping technology to extract the key characteristics of abnormal signals. After multiple rounds of optimization, it is confirmed that the signal comes from an external signal injection device that attempts to maliciously interfere with the base station.

[0184] During the entire implementation process, the system not only realizes real-time detection of complex interference signals, but also performs cross-domain adaptability testing on interference in multi-domain signal environments. By comparing the traditional detection method with the method of the present invention, the effect data is shown in Table 1 below:

[0185] Table 1 Comparison of wireless signal interference detection effects

[0186]

[0187] The data in the table shows that before the system was deployed, the traditional method detected a large number of suspicious signal interferences every day, most of which were complex signals that could not be accurately classified due to unclear characteristics. After the deployment of the present invention, the number of interference detections per day was greatly reduced to 2 times, the system's high-complexity interference recognition rate was significantly improved to 96.8%, and the cross-domain signal adaptability test results reached 98.5%. The system's false alarm rate was reduced from 7.3% to 1.2%, and the detection time was shortened from an average of 12 minutes to 1.5 minutes.

[0188] In a specific case, the system detected that the signal of a certain base station fluctuated frequently during peak hours. By generating simulated samples through the dynamic adversarial sample generation module and analyzing them, it was found that the frequency domain characteristics of the signal were significantly different from those of the normal communication signal, and it was suspected to be interfered by external signal injection. After extracting the key characteristics of the signal using sparse characterization technology, the system restored the interference pattern through inverse mapping, and finally located the interference source device and prevented its further impact. The success of this case shows that the present invention has significant advantages in processing complex interference signals and achieving real-time response.

[0189] Through actual data verification, the present invention not only solves the problem of difficult classification of complex interference signals, but also significantly improves the efficiency and reliability of interference detection. Operator reports show that since the deployment of the system, communication interruptions have been greatly reduced and network reliability has been significantly improved. The successful application of the present invention in complex wireless communication environments has proved its important value in the field of interference detection and communication assurance.

[0190] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A wireless signal interference detection method based on a generative adversarial network, characterized in that: The steps include: S1. Construct a generative adversarial network model for wireless signal interference detection. The generator is used to generate signal data that simulates complex interference patterns, and the discriminator is used to evaluate the difference in interference characteristics between the generated signal and the real signal. S2. Design a dynamic network optimization mechanism based on the disturbance characteristics of wireless signal data, and dynamically adjust the network structure through signal disturbance intensity, adaptive weight allocation and multi-layer feature fusion; S3, using cross-domain interference pattern migration technology to map the interference characteristics of different domains to a unified feature space, and conduct adversarial training by using the signal distribution difference between the source domain and the target domain; S4, based on the sparse representation and inverse mapping method of interference signals, the sparse features of interference signals are extracted through compressed sensing technology, and the time domain and frequency domain distribution of interference patterns are restored using the inverse mapping algorithm; S5. Based on the collaborative optimization mechanism of bidirectional feedback between generation and discrimination, the discriminator feedback information is introduced into the generator to optimize the generated samples, and the output of the generator is used to update the detection capability of the discriminator; S6. Adopt adaptive data enhancement technology based on dynamic adversarial sample generation, expand the training data set through random perturbation interpolation and feature mixing strategy, and output interference detection results.

2. According to claim 1, a wireless signal interference detection method based on a generative adversarial network is characterized in that: The S1 specifically includes: S11. Design the generator network structure. The generator uses random noise input vector and dynamically extracted interference characteristic conditions As input, generate signal data that simulates a complex interference pattern: x g =G(z,c;θ g ); Among them, G is the generator network, θ g represents the parameters of the generator, the interference characteristic condition c is obtained from the original signal data through the adaptive feature extraction module, and c dynamically combines the frequency domain and time domain characteristics to enhance the interference adaptability of the generated signal; S12. Design the discriminator network structure. The discriminator combines the multi-resolution feature discrimination mechanism, inputs the signal data x and the interference characteristic condition c, and outputs the probability that the signal belongs to the real signal: D(x,c;θ d )=σ(f(x,c;α,θ d )); Among them, σ is the activation function, the signal perturbation sensitivity weight introduced by the discrimination mechanism is α, f is the weighted feature mapping function, θ d are the parameters of the discriminator; S13. Define the generative adversarial loss function and expand the generator optimization objective to interference feature consistency evaluation: Among them, λ is the weight factor, L c represents the consistency loss function of the generated signal interference characteristics; S14. Optimize the generator and discriminator through a dynamic parameter update mechanism, and use a multi-batch training strategy to adjust θ in real time g and θ d , so that the signal generated by the generator gradually approaches the interference distribution of the real signal, and the error of the discriminator converges to the specified range.

3. The wireless signal interference detection method based on a generative adversarial network according to claim 1 is characterized in that: The S2 specifically includes: S21. Based on the disturbance characteristics of wireless signal data, a multi-scale disturbance analysis model is constructed. By analyzing the multiple disturbance patterns of the signal in the time domain, frequency domain and amplitude dimensions, a characteristic mapping matrix M is generated. perturb , where M perturb The elements of represent the weighted associations between different disturbance characteristics; S22, Adaptive network optimization algorithm driven by disturbance characteristics, dynamically calculates signal disturbance intensity Δ and disturbance gradient change rate The perturbation-driven weight update rule is: in, is the weight of the previous round, η is the learning rate, γ and κ represent the influence factors of perturbation intensity and gradient change on weight update respectively; S23, based on the characteristic mapping matrix M perturb Extract multi-layer perturbation features {F1,F2,...,F k }, combined with the dynamic fusion strategy to define the overall feature representation F c : Among them, β i is the dynamically adjusted feature weight factor, T(F i ) represents the feature transformation function; S24. In the process of network optimization, the disturbance-driven mechanism is used to adjust the network topology structure. By introducing a dynamic connection module, the connection density between network layers is optimized, nodes are dynamically increased or decreased, and an optimized network architecture with adaptive capabilities is constructed. S25. The model training of the generative adversarial network is completed through the adaptive network optimization mechanism driven by the perturbation characteristics. The perturbation-driven feature module is embedded in the generator, and the feature-sensitive evaluation unit is introduced in the discriminator to keep the generated signal consistent with the real signal.

4. The method for detecting wireless signal interference based on a generative adversarial network according to claim 1, characterized in that: The S3 specifically includes: S31, extract the characteristic distribution of source domain signal and target domain signal, and define the source domain signal distribution as P s (x s ), the target domain signal distribution is P t (x t ), and the two domain signals are mapped to a unified feature space through the feature extraction function f(·) to obtain the feature representation Z s and Z t : Z s =f(x s ;θ f ); Z t =f(x t ;θ f ); Among them, θ f Parameters of the feature extraction function; S32, based on the cross-domain signal distribution difference measurement function D(P s ,P t ), and optimize the feature extraction function f(·) through adversarial training to minimize the difference between the source domain and target domain feature distribution: Among them, N s and N t are the number of samples in the source domain and the target domain, μ s and μ t are the means of the source domain and target domain features respectively; S33, build a domain discriminator network, input feature representation Z, output the probability distribution of domain categories, and optimize the discriminator network parameters θ through adversarial training d : Where D represents the domain discriminator function; S34, optimize the parameter θ of the feature extraction function f(·) f , minimizing the difference in cross-domain feature distribution and the inconsistency of domain discriminator output results: Among them, λ is a regulation factor, balancing the influence of distribution difference measurement and adversarial training; S35. Complete cross-domain feature space mapping through adversarial training process, so that the source domain and target domain feature distributions achieve minimum differentiation in a unified feature space, and generate an optimized model that meets the interference characteristics requirements of the target domain.

5. The method for detecting wireless signal interference based on a generative adversarial network according to claim 1, characterized in that: The S4 specifically includes: S41. Construct a signal processing model based on compressed sensing to project the input interference signal x into a low-dimensional sparse space: y=Φx+; Among them, y is the compressed signal, Φ is the compressed sensing matrix, and is the noise interference component; S42. Extracting sparse coefficients of the signal using adaptive sparse representation technology: x = Ψ·θ; Among them, Ψ is the sparse basis matrix, θ is the sparse coefficient; S43, jointly optimize the sparse coefficient θ by minimizing the signal reconstruction error and the sparse constraint: Among them, λ is the regularization parameter, which controls the balance between sparsity and reconstruction error; S44, based on the optimized sparse coefficient θ, reverse mapping is performed to restore the high-dimensional distribution characteristics of the signal and construct a restored signal And extract the time domain disturbance pattern and frequency domain interference characteristics; S45. Combine the multi-scale feature extraction method, dynamically adjust the feature fusion weight, and generate the interference pattern characteristic model M interf , for time domain characteristics and frequency domain characteristics Conduct comprehensive modeling: Among them, g is the characteristic mapping function, θ m are model parameters; S46, based on the constructed interference pattern characteristic model M interf , calculate the multi-dimensional characteristic indicators of the signal, including the time domain disturbance intensity distribution, frequency domain energy density distribution and cross-domain correlation characteristics, through the characteristic model parameter θ m The optimization output is a complete description of the interference pattern, and the interference pattern characterization result is converted into a standardized feature vector.

6. The method for detecting wireless signal interference based on a generative adversarial network according to claim 1, characterized in that: The S5 specifically includes: S51, construct the feedback-driven optimization mechanism of the generator, and receive the confidence output D(G(z,c;θ) of the generated sample by the discriminator in real time through the dynamic feedback module g );θ d ), the characteristic distribution of the samples generated by the generator is optimized by weighted fusion of the discriminator feedback information. The optimization goal of the generator is: Where f(·) is the signal feature extraction function, α is the feedback adjustment coefficient, and dynamically adjusts the contribution of the discriminator feedback to the optimization of the generated samples; S52, embed the generated sample feature enhancement module in the discriminator, evaluate the feature difference between the generated sample and the real signal through the feature consistency measurement function Δ(G(x), x), and optimize the discriminator's ability to distinguish signal distribution based on this: Among them, β is a feature consistency measurement factor, which strengthens the sensitivity of the discriminator to the differences in the features of the generated samples; S53. A two-way optimization framework for generation and discrimination is established. The generator dynamically adjusts the generated samples to make their characteristic distribution close to the real signal. At the same time, the discriminator uses the generated samples to improve the recognition accuracy of complex interference patterns: Among them, λ is the importance balance factor optimized by the generator and the discriminator; S54, adopting a collaborative training mechanism based on adversarial dynamic weights, dynamically adjusting the training weights of the generator and the discriminator through round-by-round iterations, so that the interference signal samples generated by the generator gradually approach the real signal distribution in the multi-dimensional feature space; S55. After the collaborative optimization is completed, the feature generalization ability of the generator is tested so that the signal samples generated by it maintain a consistent distribution in different interference scenarios, and the discriminator is comprehensively evaluated so that it can accurately detect different types of interference patterns; S56. Use the optimized generator to generate a variety of high-complexity interference signal samples, and use the optimized discriminator to accurately classify unknown interference characteristics to achieve source analysis and real-time detection of interference patterns.

7. The method for detecting wireless signal interference based on a generative adversarial network according to claim 1, characterized in that: The S6 specifically includes: S61. Construct a dynamic adversarial sample generation module to generate diverse interference samples through an adaptive generation method based on perturbation interpolation: x aug =αx i +(1-a)x j +; Among them, x i and x j are signal samples from different sources, α is a dynamically assigned weight factor, which is adjusted according to the characteristic distribution of the input samples, and is the introduced random noise disturbance; S62, based on the adaptive enhancement strategy of multi-feature mixing, the multi-dimensional features of the signal sample are mapped to a unified enhanced feature space, and the feature fusion process is dynamically adjusted through the nonlinear feature transformation function g(·) and the characteristic adaptive weight β to generate the enhanced sample feature F aug ; S63. Combine adversarial samples, mixed feature samples and real signal samples to construct an adaptive enhanced dataset D aug ,The construction of the enhanced dataset is based on a hierarchical feature coverage strategy, which makes the generated samples widely distributed in the multi-dimensional feature space: D aug =D real ∪D gen ∪D mix ; Among them, D real is the real signal sample set, D gen is the generated adversarial sample set, D mix is a mixed feature sample set; S64. Using adaptive enhancement dataset D aug The interference detection model is trained with the optimization goal of minimizing the classification error of the enhanced data set, and the model's adaptability to samples of different categories is adjusted through a dynamic loss function; S65. Based on the trained interference detection model, the input signal is classified into interference patterns in real time, and a multi-dimensional characteristic description of the interference signal is output, including the probability distribution of the interference type, the signal characteristic index and its time-frequency domain distribution information.

Images