Wireless communication anti-interference method and system based on data analysis

Abstract

The invention discloses a wireless communication anti-interference method and system based on data analysis, and belongs to the technical field of communication anti-interference, in the wireless communication anti-interference method and system based on data analysis, the active defense capability of the system to unknown interference is improved through virtual interference generation and physical suppression by linkage cooperation of a generative adversarial network and an adaptive beam forming algorithm. The generator continuously simulates a novel interference signal injection training environment, so that a beam forming algorithm learns in advance to form accurate null in an interference direction, and the response time when actual interference occurs is greatly shortened. Meanwhile, interference space features detected by the beam forming module in an actual scene reversely optimize a signal construction rule of the generator, and the matching degree of a generated sample and a real electromagnetic environment is enhanced. The dynamic mutual promotion mechanism not only enhances the timeliness of interference suppression, but also reduces the problems of signal quality fluctuation and excessive consumption of hardware resources caused by passive response in the traditional scheme.

Description

Technical Field

[0001] The present invention belongs to the technical field of communication anti-interference, and specifically relates to a wireless communication anti-interference method and system based on data analysis. Background Art

[0002] Wireless communication anti-interference refers to the use of specific technical means to enhance the ability of wireless communication systems to resist interference from internal or external unwanted signals in complex electromagnetic environments, thereby ensuring the reliability, effectiveness, and continuity of communications. Interference sources may include other wireless devices, industrial equipment, natural phenomena (such as lightning), or hostile signals (in military scenarios). There are many types of anti-interference technologies, the most common of which include: spread spectrum communications (such as direct sequence spread spectrum DSSS and frequency hopping spread spectrum FHSS), which reduce the impact of interference by expanding signal bandwidth or dynamically changing frequency; multi-antenna technologies (such as MIMO), which use spatial diversity or beamforming to improve signal quality; and forward error correction coding, adaptive modulation, and channel estimation. The combined application of these technologies aims to minimize the negative impact of interference on communication link quality, data transmission rate, and system capacity, ensuring the stable operation of wireless communications in various environments.

[0003] However, traditional beamforming algorithms require parameter adjustments only after interference actually occurs, resulting in delayed responses and degraded communication quality. Static interference signature libraries struggle to adapt to dynamic changes in the electromagnetic environment, leading to a decline in suppression effectiveness after long-term operation. Summary of the Invention

[0004] The purpose of the present invention is to provide a wireless communication anti-interference method and system based on data analysis in order to solve the above-mentioned problems.

[0005] The technical solution adopted by the present invention is as follows: a wireless communication anti-interference method and system based on data analysis, the method comprising the following steps:

[0006] S1: Real-time collection of spectrum data of wireless communication frequency bands to monitor signal strength and noise distribution;

[0007] S2: Extract spectrum features through time-frequency transformation to identify the time and frequency domain patterns of potential interference signals;

[0008] S3: Build a dynamic interference feature library to store interference waveforms and statistical characteristics in different scenarios;

[0009] S4: Design a generative adversarial network to simulate unknown interference signals and generate adversarial training samples to inject into the feature library;

[0010] S5: Train a deep neural network classifier to distinguish normal communication signals from various interference patterns;

[0011] S6: Deploy an online learning module to update the interference identification model parameters based on real-time channel feedback;

[0012] S7: Establish a multi-dimensional interference assessment matrix to quantify the impact of interference intensity on communication quality;

[0013] S8: Develop adaptive beamforming algorithms to dynamically adjust antenna radiation patterns based on interference distribution;

[0014] S9: Implement intelligent power control strategy and adopt non-uniform symbol energy distribution scheme during interference peak period.

[0015] In a preferred embodiment, in step S1, a software-defined radio device is used to scan the 400MHz-6GHz frequency band with a period of 10ms, configured with a 100MHz instantaneous bandwidth and 14-bit quantization accuracy, and a dynamic range of not less than 90dB. The data acquisition module records the I / Q signal at a 256-fold oversampling rate, synchronously triggers the noise floor calibration program, and eliminates hardware nonlinear errors through the Kalman filter. The raw data stream is Hilbert transformed to generate a time-frequency spectrum matrix, and the signal strength indicator RSSI and noise variance σ of each 5MHz subband are calculated in real time. 2 , the abnormal frequency trigger threshold is set to the background noise mean plus three times the standard deviation.

[0016] In a preferred embodiment, in step S2, the intercepted signal is subjected to a short-time Fourier transform (SFT), using a Blackman-Harris seventh-order window function with a frame length of 256 samples and an overlap ratio of 75%. Time-domain feature extraction includes zero-crossing rate, peak-to-average ratio, and pulse width distribution, while frequency-domain feature calculation includes normalized spectrum centroid and bandwidth occupancy factor. An improved Viterbi algorithm is used to align the time-frequency ridges of multi-component signals. A hidden Markov model is used to identify periodic interference patterns. A support vector machine is used to construct a binary classifier to distinguish between burst and continuous interference. The feature dimensions are compressed to 32 dimensions before being input to the next-level module.

[0017] In a preferred embodiment, in step S3, the dynamic database utilizes a hierarchical storage architecture. The first layer stores the original time-frequency graphs indexed by timestamp, the second layer stores the reduced feature vectors and labels, and the third layer records the statistical mean, variance, and cross-correlation matrix. Data updates follow a sliding window principle, retaining the last 72 hours of data while performing incremental clustering every 15 minutes. The density-based clustering algorithm (DBSCAN) is used to remove outliers, with core parameters set to a minimum neighborhood of 5 and a neighborhood radius of 0.35. The interference waveform template library has a capacity cap of 10,000 entries, and uses a least recently used strategy to replace old data.

[0018] In a preferred embodiment, in step S4, an interference signal simulation framework is constructed based on a generative adversarial network. First, a real interference sample is used to train the generator to capture the frequency domain energy distribution and time domain waveform distortion characteristics. The discriminator distinguishes between real samples and synthetic samples through spectral similarity measurement and mutual information constraint. The generator jointly optimizes the adversarial nature of the generated signal by minimizing the Wasserstein distance and maximizing the discriminator's misjudgment probability. At the same time, random phase perturbations and impulse noise factors are embedded in the latent space to enhance sample diversity. The dynamically generated interference samples are verified in the time and frequency domains and stored in the feature library for classifier robustness training.

[0019] The calculation formula of the generator loss function is:

[0020]

[0021] Where: D represents the discriminator output score, G(z) is the interference signal output by the generator, z is the potential space noise vector, Φ(·) is the signal spectrum energy distribution operator, I real represents the real interference sample, α represents the control of adversarial strength, and β represents the adjustment of spectrum feature matching weight.

[0022] In a preferred embodiment, in step S5, a deep neural network is constructed using a hybrid CNN-LSTM architecture. The input layer receives a 256×256 time-frequency spectrum, the convolutional layer is configured with 64 5×5 filters, the LSTM hidden layer has 128 units, and the output layer is activated with Softmax to generate five types of interference probability distributions. Focal Loss is used as the loss function to mitigate class imbalance, and the optimizer is Adam with Nesterov momentum, with an initial learning rate of 0.001 and a batch size of 32. The training set is partitioned using 5-fold temporal cross-validation, with an early stopping monitoring window of 20 epochs. After model convergence, the convolutional layer weights are frozen for online learning.

[0023] In a preferred embodiment, in step S6, the model update trigger condition is set to a KL divergence between the current batch of data and the historical distribution exceeding 0.3. An elastic weight hardening algorithm is used to protect important parameters, and the learning rate of the frozen layer is reduced to 1% of its original value. The incremental learning module loads new data buffers every 30 minutes and performs momentum-corrected stochastic gradient descent with a momentum coefficient of 0.9 and a learning rate decay factor of 0.95. Before parameter updates, Jacobi matrix condition number testing is performed to ensure system stability, and abnormal conditions automatically roll back to the previous three model snapshots.

[0024] In a preferred embodiment, step S7 defines four evaluation metrics: signal-to-interference-and-noise ratio degradation ΔSINR, packet error rate (K_PER), delay jitter variance σ_Jitter, and bandwidth utilization decrease rate η_BW. Each metric is normalized and weighted using the entropy weighting method to construct a linear weighted scoring model. Double exponential smoothing is used in real-time calculations to predict metric trends. When the combined score exceeds a threshold of 7.5, the anti-interference decision engine is triggered. Evaluation results are updated every 500ms and written to the quality monitoring report.

[0025] In a preferred embodiment, in step S8, an adversarial beamforming framework is deployed, the interference signal generated by S4 is used as the dynamic environment input, and the beam weight matrix is trained to form a null in the direction of the generated signal interference. is mapped to a joint spatial angle-delay distribution P(θ,τ), and the beam weights are updated by solving the following optimization problem:

[0026]

[0027] where a(θ k ,τ k ) is the array response vector corresponding to the interference sample G(zk), TV(w) is the total variation constraint on the beam weights, and λ is the smoothing factor. The algorithm uses the virtual interference spatial distribution generated by S4 as prior knowledge, enabling beamforming to adapt to unknown interference patterns in advance.

[0028] Steps S4 and S8 cooperate with each other to achieve:

[0029] Data closure: The interference spatial features actually detected in S8 are fed back into the S4 generator to expand its potential space dimension. Achieving co-evolution of interference generation and suppression

[0030] Increasing difficulty: Dynamically adjust the power level of the S4 generated signal according to the S8 null depth indicator αt=α0·e ηt , where η is the learning rate, forming adversarial enhancement training

[0031] Physical verification: adding array response reversibility constraints to S4 generated samples Ensure that it complies with the physical laws of electromagnetic wave propagation.

[0032] In a preferred embodiment, in step S9, non-uniform symbol energy allocation is implemented in the OFDM system and is activated when the interference power exceeds -95dBm. The cost function is defined to minimize the inter-symbol interference and peak-to-average ratio as follows:

[0033]

[0034] Among them E k is the kth subcarrier energy, is the equivalent noise power, Γ is the target signal-to-noise ratio, and μ is the peak-to-average ratio constraint coefficient. The solution is solved using the interior point method with 20 iterations. The power adjustment period is set to 10 OFDM symbols. The adjustment result is used to compensate for the nonlinear characteristics of the power amplifier through a predistortion feedback loop.

[0035] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0036] 1. In this invention, the closed-loop coordination of generative adversarial networks and dynamic beamforming significantly improves the system's real-time defense capability against unknown interference. The generator actively constructs diverse interference samples, enabling the classifier to identify new interference patterns that have not appeared in historical data in advance. At the same time, the beamforming algorithm uses these virtual interference samples for pre-training, quickly forming deep nulls when real interference occurs, reducing the signal loss rate to less than a quarter of that of traditional methods. This linkage mechanism of virtual adversarial and physical suppression enables the communication system to maintain stable transmission when the interference environment suddenly changes, avoiding the risk of communication interruption caused by traditional solutions that rely on post-adjustment.

[0037] 2. In this invention, long-term adaptability in complex electromagnetic environments is enhanced through dynamic feedback of interference generation and spatial suppression. The generator reversely optimizes the time-frequency characteristics of the interference samples based on the measured results of beamforming, forcing the beamforming algorithm to continuously evolve its spatial filtering accuracy, forming a spiraling countermeasure enhancement effect. This self-driven optimization process enables the system to automatically match base station deployment changes with terminal movement trajectories, while ensuring a bit error rate below 1e-6, while reducing the ineffective radiation power of the antenna array, significantly reducing the risk of adjacent channel interference, and extending the continuous operation time of the device in areas with strong interference.

[0038] 2. In the present invention, the generative adversarial network actively constructs unknown interference samples for enhanced training and the adaptive beamforming algorithm dynamically suppresses the spatial propagation of interference, which improves the system's active defense capability against unknown interference through closed-loop feedback of virtual interference generation and physical suppression. The generator continuously simulates new interference signals to inject into the training environment, so that the beamforming algorithm learns in advance to form precise nulls in the interference direction, greatly shortening the response time when actual interference occurs. At the same time, the beamforming module reversely optimizes the generator's signal construction rules based on the interference space characteristics detected in the actual scene, enhancing the matching degree between the generated samples and the real electromagnetic environment. This dynamic mutual promotion mechanism not only enhances the timeliness of interference suppression, but also improves the long-term stability of the system in complex and changing scenarios through continuous adversarial training, while reducing the signal quality fluctuations and excessive consumption of hardware resources caused by passive response of traditional solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the process principle of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] Example

[0042] Reference Figure 1 A wireless communication anti-interference method and system based on data analysis, the method comprising the following steps:

[0043] S1: Real-time collection of spectrum data of wireless communication frequency bands to monitor signal strength and noise distribution;

[0044] S2: Extract spectrum features through time-frequency transformation to identify the time and frequency domain patterns of potential interference signals;

[0045] S3: Build a dynamic interference feature library to store interference waveforms and statistical characteristics in different scenarios;

[0046] S4: Design a generative adversarial network to simulate unknown interference signals and generate adversarial training samples to inject into the feature library;

[0047] S5: Train a deep neural network classifier to distinguish normal communication signals from various interference patterns;

[0048] S6: Deploy an online learning module to update the interference identification model parameters based on real-time channel feedback;

[0049] S7: Establish a multi-dimensional interference assessment matrix to quantify the impact of interference intensity on communication quality;

[0050] S8: Develop adaptive beamforming algorithms to dynamically adjust antenna radiation patterns based on interference distribution;

[0051] S9: Implement intelligent power control strategy and adopt non-uniform symbol energy distribution scheme during interference peak period.

[0052] In step S1, a software-defined radio device is used to scan the 400MHz-6GHz frequency band with a 10ms cycle, configured with a 100MHz instantaneous bandwidth and 14-bit quantization accuracy, with a dynamic range of no less than 90dB. The data acquisition module records the I / Q signals at a 256x oversampling rate, synchronously triggers the noise floor calibration procedure, and eliminates hardware nonlinear errors through the Kalman filter. The raw data stream is Hilbert transformed to generate a time-frequency spectrum matrix, and the signal strength indicator RSSI and noise variance σ are calculated in real time for each 5MHz subband. 2 , the abnormal frequency trigger threshold is set to the background noise mean plus three times the standard deviation.

[0053] In step S2, the intercepted signal is subjected to a short-time Fourier transform (SFT). The window function uses a Blackman-Harris seventh-order window, a frame length of 256 samples, and an overlap ratio of 75%. Time-domain feature extraction includes zero-crossing rate, peak-to-average ratio, and pulse width distribution. Frequency-domain feature calculation includes normalized spectrum centroid and bandwidth occupancy factor. An improved Viterbi algorithm is used to align the time-frequency ridges of the multi-component signals. A hidden Markov model is used to identify periodic interference patterns. A support vector machine is used to construct a binary classifier to distinguish between burst and continuous interference. The feature dimensions are compressed to 32 before being input to the next-level module.

[0054] In step S3, the dynamic database uses a hierarchical storage architecture. The first layer stores the original time-frequency graphs indexed by timestamp, the second layer stores the reduced feature vectors and labels, and the third layer records the statistical mean, variance, and cross-correlation matrix. Data updates follow the sliding window principle, retaining the last 72 hours of data while performing incremental clustering every 15 minutes. The density clustering algorithm DBSCAN is used to remove outliers, with core parameters set to a minimum neighborhood of 5 points and a neighborhood radius of 0.35. The interference waveform template library has a capacity cap of 10,000 entries, and uses a least recently used strategy to replace old data.

[0055] In step S4, an interference signal simulation framework is constructed based on a generative adversarial network. First, the generator is trained with real interference samples to capture the frequency domain energy distribution and time domain waveform distortion characteristics. The discriminator distinguishes real samples from synthetic samples through spectral similarity measurement and mutual information constraints. The generator jointly optimizes the adversarial nature of the generated signal by minimizing the Wasserstein distance and maximizing the discriminator's misjudgment probability. At the same time, random phase perturbations and impulse noise factors are embedded in the latent space to enhance sample diversity. The dynamically generated interference samples are verified in the time and frequency domains and stored in the feature library for classifier robustness training.

[0056] The calculation formula of the generator loss function is:

[0057]

[0058] Where: D represents the discriminator output score, G(z) is the interference signal output by the generator, z is the potential space noise vector, Φ(·) is the signal spectrum energy distribution operator, I real represents a real interference sample, α controls the adversarial strength, and β adjusts the spectral feature matching weight. This formula innovatively combines adversarial training with physical layer signal feature constraints. It ensures the physical plausibility of generated samples through spectral energy alignment, while leveraging adversarial mechanisms to break through the boundaries of historical data distribution.

[0059] In step S5, a deep neural network using a hybrid CNN-LSTM architecture was constructed. The input layer received a 256×256 time-frequency spectrum graph, the convolutional layer was configured with 64 5×5 filters, the LSTM hidden layer had 128 units, and the output layer had a Softmax activation to generate probability distributions for five types of interference. Focal Loss was used as the loss function to mitigate class imbalance, and the optimizer used was Adam with Nesterov momentum, with an initial learning rate of 0.001 and a batch size of 32. The training set was partitioned using 5-fold temporal cross-validation, with an early stopping monitoring window of 20 epochs. After model convergence, the convolutional layer weights were frozen for online learning.

[0060] In step S6, the model update trigger condition is set to a KL divergence between the current batch of data and the historical distribution exceeding 0.3. An elastic weight hardening algorithm is used to protect key parameters, and the learning rate of the frozen layer is reduced to 1% of its original value. The incremental learning module loads new data buffers every 30 minutes and performs momentum-corrected stochastic gradient descent with a momentum coefficient of 0.9 and a learning rate decay factor of 0.95. Before parameter updates, Jacobi matrix condition number checks are performed to ensure system stability. In the event of an abnormal state, the system automatically rolls back to the previous three model snapshots.

[0061] In step S7, four evaluation metrics are defined: signal-to-interference-and-noise ratio degradation ΔSINR, packet error rate (K_PER), delay jitter variance σ_Jitter, and bandwidth utilization decrease rate η_BW. Each metric is normalized and weighted using the entropy weighting method to construct a linear weighted scoring model. Double exponential smoothing is used in real-time calculations to predict metric trends. When the combined score exceeds a threshold of 7.5, the anti-interference decision engine is triggered. Evaluation results are updated every 500ms and written to the quality monitoring report.

[0062] In step S8, the adversarial beamforming framework is deployed, the interference signal generated by S4 is used as the dynamic environment input, and the beam weight matrix is trained to form a null in the direction of the generated signal interference. is mapped to a joint spatial angle-delay distribution P(θ,τ), and the beam weights are updated by solving the following optimization problem:

[0063]

[0064] where a(θ k ,τ k ) is the array response vector corresponding to the interference sample G(zk), TV(w) is the total variation constraint on the beam weights, and λ is the smoothing factor. The algorithm uses the virtual interference spatial distribution generated by S4 as prior knowledge, enabling beamforming to adapt to unknown interference patterns in advance.

[0065] Steps S4 and S8 cooperate with each other to achieve:

[0066] Data closure: The interference spatial features actually detected in S8 are fed back into the S4 generator to expand its potential space dimension. Achieving co-evolution of interference generation and suppression

[0067] Increasing difficulty: Dynamically adjust the power level of the S4 generated signal according to the S8 null depth indicator αt=α0·e ηt , where η is the learning rate, forming adversarial enhancement training

[0068] Physical verification: adding array response reversibility constraints to S4 generated samples Ensure that it complies with the physical laws of electromagnetic wave propagation.

[0069] In step S9, non-uniform symbol energy allocation is implemented in the OFDM system and is activated when the interference power exceeds -95dBm. The cost function is defined to minimize the inter-symbol interference and peak-to-average ratio as follows:

[0070]

[0071] Among them E k is the kth subcarrier energy, is the equivalent noise power, Γ is the target signal-to-noise ratio, and μ is the peak-to-average ratio constraint coefficient. The solution is solved using the interior point method with 20 iterations. The power adjustment period is set to 10 OFDM symbols. The adjustment result is used to compensate for the nonlinear characteristics of the power amplifier through a predistortion feedback loop.

[0072] A wireless communication anti-interference system based on data analysis, which runs the wireless communication anti-interference method based on data analysis according to any one of claims 1 to 9 when in use.

[0073] From the above we can know:

[0074] In this invention, the closed-loop coordination of generative adversarial networks and dynamic beamforming significantly improves the system's real-time defense capabilities against unknown interference. The generator actively constructs diverse interference samples, enabling the classifier to identify new interference patterns that have not appeared in historical data in advance. At the same time, the beamforming algorithm uses these virtual interference samples for pre-training, quickly forming deep nulls when real interference occurs, reducing the signal loss rate to less than a quarter of that of traditional methods. This linkage mechanism of virtual adversarial and physical suppression enables the communication system to maintain stable transmission when the interference environment suddenly changes, avoiding the risk of communication interruption caused by traditional solutions that rely on post-adjustment.

[0075] This invention enhances long-term adaptability in complex electromagnetic environments through dynamic feedback between interference generation and spatial suppression. The generator reversely optimizes the time-frequency characteristics of interference samples based on beamforming measurements, forcing the beamforming algorithm to continuously evolve its spatial filtering accuracy, creating a spiraling countermeasure enhancement effect. This self-driven optimization process enables the system to automatically match base station deployment changes with terminal movement trajectories, while ensuring a bit error rate below 1e-6. This reduces the ineffective radiation power of the antenna array, significantly minimizing the risk of adjacent channel interference, and extending the device's continuous operation time in areas of strong interference.

[0076] In the present invention, the generative adversarial network actively constructs unknown interference samples for enhanced training and the adaptive beamforming algorithm dynamically suppresses the spatial propagation of interference, thereby improving the system's active defense capability against unknown interference through closed-loop feedback of virtual interference generation and physical suppression. The generator continuously simulates new interference signals and injects them into the training environment, enabling the beamforming algorithm to learn in advance to form precise nulls in the interference direction, significantly shortening the response time when actual interference occurs. At the same time, the beamforming module reversely optimizes the generator's signal construction rules based on the interference space characteristics detected in the actual scenario, enhancing the matching degree between the generated samples and the real electromagnetic environment. This dynamic mutual promotion mechanism not only enhances the timeliness of interference suppression, but also improves the long-term stability of the system in complex and changing scenarios through continuous adversarial training, while reducing the signal quality fluctuations and excessive consumption of hardware resources caused by passive response in traditional solutions.

[0077] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0078] The above description is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wireless communication anti-interference method and system based on data analysis, characterized by: The method comprises the following steps: S1: Real-time collection of spectrum data of wireless communication frequency bands to monitor signal strength and noise distribution; S2: Extract spectrum features through time-frequency transformation to identify the time and frequency domain patterns of potential interference signals; S3: Build a dynamic interference feature library to store interference waveforms and statistical characteristics in different scenarios; S4: Design a generative adversarial network to simulate unknown interference signals and generate adversarial training samples to inject into the feature library; S5: Train a deep neural network classifier to distinguish normal communication signals from various interference patterns; S6: Deploy an online learning module to update the interference identification model parameters based on real-time channel feedback; S7: Establish a multi-dimensional interference assessment matrix to quantify the impact of interference intensity on communication quality; S8: Develop adaptive beamforming algorithms to dynamically adjust antenna radiation patterns based on interference distribution; S9: Implement intelligent power control strategy and adopt non-uniform symbol energy distribution scheme during interference peak period.

2. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S1, a software-defined radio device is used to scan the 400MHz-6GHz frequency band with a period of 10ms, with a configuration of 100MHz instantaneous bandwidth and 14-bit quantization accuracy, and a dynamic range of not less than 90dB; a data acquisition module records I / Q signals at a 256-fold oversampling rate, synchronously triggers a noise floor calibration program, and eliminates hardware nonlinear errors through a Kalman filter; the raw data stream is Hilbert transformed to generate a time-frequency spectrum matrix, and the signal strength indicator RSSI and noise variance σ of each 5MHz subband are calculated in real time. 2 , the abnormal frequency trigger threshold is set to the background noise mean plus three times the standard deviation.

3. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S2, a short-time Fourier transform is performed on the intercepted signal, and a Blackman-Harris seventh-order window is selected as the window function, with a frame length of 256 sampling points and an overlap rate of 75%; Time domain feature extraction includes zero-crossing rate, peak-to-average ratio and pulse width distribution, and frequency domain feature calculation includes normalized spectrum centroid and bandwidth occupancy factor. An improved Viterbi algorithm is used to align the time-frequency ridges of multi-component signals, and a hidden Markov model is used to identify periodic interference patterns. A support vector machine is used to construct a binary classifier to distinguish between burst and continuous interference. The feature dimension is compressed to 32 dimensions and then input into the lower-level module.

4. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S3, the dynamic database adopts a layered storage architecture, the first layer stores the original time-frequency graph by timestamp index, the second layer stores the eigenvectors and labels after dimensionality reduction, and the third layer records the statistical mean, variance and cross-correlation matrix; data update follows the sliding window principle, retaining the data of the last 72 hours while performing incremental clustering every 15 minutes, using the density clustering algorithm DBSCAN to eliminate outliers, and the core parameters are set to a minimum neighborhood point number of 5 and a neighborhood radius of 0.35; the interference waveform template library is set with a capacity upper limit of 10,000, and the least recently used strategy is used to replace old data.

5. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S4, an interference signal simulation framework is constructed based on a generative adversarial network. First, a generator is trained using real interference samples to capture the frequency domain energy distribution and time domain waveform distortion characteristics. The discriminator distinguishes real samples from synthetic samples through spectral similarity measurement and mutual information constraints. The generator jointly optimizes the adversarial nature of the generated signal by minimizing the Wasserstein distance and maximizing the discriminator's misjudgment probability. At the same time, random phase perturbations and impulse noise factors are embedded in the latent space to enhance sample diversity. The dynamically generated interference samples are verified in the time and frequency domains and then stored in a feature library for classifier robustness training. The calculation formula of the generator loss function is: Where: D represents the discriminator output score, G(z) is the interference signal output by the generator, z is the potential space noise vector, Φ(·) is the signal spectrum energy distribution operator, I real represents the real interference sample, α represents the control of adversarial strength, and β represents the adjustment of spectrum feature matching weight.

6. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S5, a deep neural network is constructed using a CNN-LSTM hybrid structure, the input layer receives a 256×256 time-frequency spectrum, the convolution layer is configured with 64 5×5 filters, the number of LSTM hidden layer units is 128, and the output layer Softmax activation generates five types of interference probability distributions; the loss function uses Focal Loss to alleviate category imbalance, the optimizer uses Nesterov momentum Adam, the initial learning rate is 0.001, and the batch size is 32; the training set is divided into five-fold time series cross-validation, the early stopping method monitoring window is set to 20 epochs, and the convolution layer weights are frozen after the model converges for online learning.

7. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S6, the model update trigger condition is set to the KL divergence between the current batch data and the historical distribution exceeding 0.3, the elastic weight solidification algorithm is used to protect important parameters, and the frozen layer learning rate is reduced to 1% of the original value; The incremental learning module loads a new data buffer every 30 minutes and performs momentum-corrected stochastic gradient descent with a momentum coefficient of 0.9 and a learning rate decay factor of 0.

95. Before updating parameters, the Jacobi matrix condition number test is performed to ensure system stability, and abnormal conditions automatically roll back to the model snapshots of the previous three versions.

8. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S7, four evaluation indicators are defined: signal-to-interference-and-noise ratio degradation ΔSINR, packet error rate rising slope K_PER, delay jitter variance σ_Jitter, and bandwidth utilization decrease rate η_BW. After normalization, each indicator is weighted using the entropy weight method to construct a linear weighted scoring model. In real-time calculation, a double exponential smoothing method is used to predict the indicator change trend. When the comprehensive score exceeds a threshold of 7.5, the anti-interference decision engine is triggered. The evaluation results are refreshed every 500 ms and written to the quality monitoring report.

9. The wireless communication anti-interference method and system based on data analysis according to claim 1, characterized in that: In step S8, the adversarial beamforming framework is deployed, the interference signal generated by S4 is used as the dynamic environment input, and the beam weight matrix is trained to form a null in the direction of the generated signal interference; in the specific implementation, the interference sample output by the generator is mapped to a joint spatial angle-delay distribution P(θ,τ), and the beam weights are updated by solving the following optimization problem: where a(θ k ,τ k ) is the array response vector corresponding to the interference sample G(zk), TV(w) is the total variation constraint term of the beam weight, and λ is the smoothing factor. The algorithm uses the virtual interference spatial distribution generated by S4 as prior knowledge to enable beamforming to adapt to unknown interference patterns in advance. In step S9, non-uniform symbol energy allocation is implemented in the OFDM system and is activated when the interference power is detected to be greater than -95dBm. The cost function is defined to minimize the inter-symbol interference and peak-to-average ratio as follows: Among them E k is the kth subcarrier energy, is the equivalent noise power, Γ is the target signal-to-noise ratio, and μ is the peak-to-average ratio constraint coefficient; the solution is solved by the interior point method with 20 iterations, and the power adjustment period is set to 10 OFDM symbol lengths. The adjustment result is used to compensate for the nonlinear characteristics of the power amplifier through the predistortion feedback loop.

10. A wireless communication anti-interference system based on data analysis, characterized by: When in use, the system runs the wireless communication anti-interference method based on data analysis according to any one of claims 1 to 9.

Images