Wireless communication anti-interference enhancement system based on multi-mode signal fusion

Abstract

The invention relates to the technical field of wireless communication, in particular to a wireless communication anti-interference enhancement system based on multi-mode signal fusion. Comprising an intelligent signal transceiving module, a signal sensing and preprocessing module, an interference feature coding and classification module, an interference suppression decision module, an interference suppression module, a signal fusion module, an interference suppression performance evaluation module, an interference feature library and an interference suppression strategy library, the interference feature coding and classification module adopts a three-layer progressive architecture of multi-dimensional manifold feature extraction, feature transformation and coding inspired by a group theory, and deep learning classification based on chaotic dynamics, extracts interference features from three dimensions of time domain, frequency domain and nonlinearity, and realizes accurate interference identification; the signal fusion module adopts a multi-level fusion strategy, and dynamically adjusts the fusion weight according to the signal quality; the interference suppression module selects an optimal suppression strategy according to the interference type, and executes frequency domain cancellation, space domain cancellation and time domain cancellation; and performance evaluation feedback forms a closed-loop optimization mechanism.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and more specifically to a wireless communication anti-interference enhancement system based on multimodal signal fusion, and more particularly to a technical solution for improving the anti-interference capability of a wireless communication system through multi-dimensional interference feature extraction, interference classification, and multimodal signal fusion technology. Background Art

[0002] With the widespread adoption of wireless communication technologies, communication systems are facing increasingly complex and diverse interference environments. Existing anti-interference technologies primarily focus on suppressing a single type of interference, lacking a systematic solution for complex interference environments. Traditional anti-interference methods, such as spectrum spreading, adaptive filtering, and antenna arrays, perform well under specific interference types, but often struggle to maintain stable communication quality in complex electromagnetic environments where multiple interference types coexist.

[0003] At present, most wireless communication anti-interference systems have the following major shortcomings: First, the interference feature extraction dimension is single, usually only focusing on time domain or frequency domain features, ignoring the nonlinear characteristics of the interference signal; second, the interference classification method is simple, mostly using fixed threshold judgment or simple template matching, and has poor adaptability to complex interference environments; third, the suppression strategy lacks specificity, and a unified suppression method is used for different types of interference, with limited effect; fourth, there is a lack of multimodal signal fusion mechanism, which cannot fully utilize the complementary advantages of different signal modes.

[0004] Therefore, there is an urgent need to develop a wireless communication anti-interference enhancement system that can comprehensively utilize the characteristics of multimodal signals to achieve accurate interference identification and effective interference suppression. Summary of the Invention

[0005] The purpose of the present invention is to provide a wireless communication anti-interference enhancement system based on multimodal signal fusion, aiming to overcome the shortcomings of the existing technology and achieve effective response to complex interference environments through multi-dimensional interference feature extraction, intelligent interference classification and multimodal signal fusion.

[0006] The present invention proposes a wireless communication anti-interference enhancement system based on multimodal signal fusion, comprising: Intelligent signal transceiver module for: Receive multi-band and multi-modulation communication signals; Perform modulation and demodulation of signals; The signal sensing and preprocessing module is connected to the intelligent signal transceiver module and is used to: Receiving the communication signal transmitted by the intelligent signal transceiver module; performing time-frequency transformation on the communication signal; transforming the communication signal into the time-frequency domain; The interference feature coding and classification module is connected to the signal sensing and preprocessing module and is used to: Receiving the time-frequency domain signal output by the signal sensing and preprocessing module; extracting multidimensional interference features from the time-frequency domain signal; Performing standardized encoding on the multi-dimensional interference features; generating an interference type classification result based on the standardized coding; The interference suppression decision module is connected to the interference feature coding and classification module and is used to: Receiving the interference type classification result generated by the interference feature coding and classification module; selecting an interference suppression strategy based on the interference type classification result; generating interference suppression parameters; An interference suppression module, connected to the interference suppression decision module, is used to: receiving the interference suppression strategy and interference suppression parameters generated by the interference suppression decision module; performing interference suppression on a communication signal based on the interference suppression strategy and the interference suppression parameters; The signal fusion module is connected to the interference suppression module and is used to: Receiving the multimodal suppressed signal output by the interference suppression module; Performing weighted fusion on the multimodal suppressed signals; generating a fusion signal; The interference suppression performance evaluation module is connected to the signal fusion module and is used to: receiving the fusion signal output by the signal fusion module; evaluating a quality indicator of the fusion signal; Feedback of the performance evaluation result to the interference suppression decision module; Among them, the interference feature coding and classification module includes a multidimensional manifold feature extraction unit, a feature transformation and coding unit and a classification decision unit. The multidimensional manifold feature extraction unit is used to extract interference features from three dimensions: time domain, frequency domain and nonlinearity. The feature transformation and coding unit is used to generate standardized coding of interference features. The classification decision unit is used to generate interference type classification results.

[0007] Preferably, the multi-dimensional manifold feature extraction unit of the interference feature encoding and classification module includes: The time domain feature extraction subunit is used to: Apply multi-scale wavelet decomposition to extract time domain wavelet coefficients; Analyze signal waveform profile characteristics; Calculate timing-related structural parameters; Frequency domain feature extraction subunit, used for: Generate signal time-frequency distribution diagram; Calculate the energy proportion of each frequency band; extracting channel state information; Non-linear feature extraction subunit, used for: Calculate the signal entropy parameter; Extract high-order statistics such as signal kurtosis and skewness; Analyze the fractal dimension of the signal; Feature fusion subunit, used for: receiving features output by the time domain feature extraction subunit, the frequency domain feature extraction subunit, and the nonlinear feature extraction subunit; performing correlation analysis on the features; Generate a multidimensional feature matrix.

[0008] Preferably, the feature transformation and encoding unit of the interference feature encoding and classification module includes: Feature transformation subunit, used to: Receiving the multidimensional feature matrix output by the multidimensional manifold feature extraction unit; Applying geometric transformation and orthogonal transformation to the multidimensional feature matrix; Generate normalized feature vectors; Invariant extraction subunit, used to: extracting geometric invariants from the normalized eigenvectors; Extract statistical invariants; Extract topological invariants; The code generation subunit is used to: performing vector quantization on the geometric invariant, the statistical invariant, and the topological invariant; Perform feature compression; Generates a standardized encoding of the interference features.

[0009] Preferably, the classification decision unit of the interference feature coding and classification module includes: Feature learning networks for: Receiving the standardized code output by the feature transformation and encoding unit; Extract deep feature representation; Chaos prediction subunit, used for: Mapping the deep feature representation to phase space; Analyze the dynamic trajectory of interference features; predict the effects of changes in disturbance parameters; Attention mechanism subunit, used to: Calculate the contribution of each feature to the decision; Dynamically adjust feature weights; Highlight key discriminant features; Probabilistic decision subunit, used to: generating a probability distribution of interference types based on the deep feature representation, the dynamic trajectory analysis results, and the feature weights; Form the final interference type classification result.

[0010] Preferably, the intelligent signal transceiver module is applicable to a variety of frequency bands and modulation modes, including: A multi-band receiving unit for receiving low-frequency, medium-frequency and high-frequency signals; Multi-modulation processing unit, used to process signals of modulation modes such as FSK, PSK, DSSS and OFDM; The signal modulation and demodulation unit is used to perform signal modulation and demodulation.

[0011] Preferably, the signal sensing and preprocessing module includes: Bandpass filter unit, used to limit the signal frequency band and suppress out-of-band interference; Automatic gain control unit for dynamically adjusting signal levels; A digital downsampling unit for reducing the sampling rate; A time-frequency conversion unit, used to convert the signal into the time-frequency domain; The signal quality assessment unit is used to estimate the signal energy level, evaluate the signal dynamic range and detect abnormal components in the frequency domain.

[0012] Preferably, the interference suppression module includes: a frequency-domain cancellation unit for performing adaptive notch filtering, spectral subtraction, and wavelet threshold denoising; A spatial cancellation unit that performs beamforming, spatial filtering, and interference source directional suppression; a time-domain cancellation unit for performing adaptive filtering, predictor construction, and temporal pattern cancellation; A joint suppression unit is used to coordinate the work of the frequency domain cancellation unit, the space domain cancellation unit and the time domain cancellation unit according to the interference suppression strategy and the interference suppression parameters.

[0013] Preferably, the signal fusion module includes: A signal quality evaluation unit, used to evaluate the quality indicators of each modal signal; a weight allocation unit, configured to allocate a fusion weight to each modal signal based on the quality indicator; Linear fusion unit for performing weighted averaging, maximum ratio combining, and selective combining; Nonlinear fusion unit for performing median filter fusion, ranking statistic-based fusion, and fuzzy logic fusion; The post-fusion processing unit is used to perform channel re-estimation, signal-to-noise ratio calculation and signal reshaping.

[0014] Preferably, the interference suppression performance evaluation module includes: a signal-to-noise ratio calculation unit, used for calculating the signal-to-noise ratio of the fusion signal; a bit error rate evaluation unit, used to evaluate the accuracy of data transmission; a packet error rate calculation unit, used to calculate the error rate of data packet transmission; a performance indicator comprehensive evaluation unit, configured to comprehensively analyze the signal-to-noise ratio, the bit error rate, and the packet error rate to generate a performance evaluation result; A feedback generating unit is used to convert the performance evaluation result into an adjustment suggestion and feed it back to the interference suppression decision module.

[0015] As an advantage, it also includes: The interference feature library is connected to the interference feature encoding and classification module and is used to: Storing the mapping relationship between interference characteristics and interference types; receiving a newly identified interference feature from the interference feature encoding and classification module; Update the feature library content; An interference suppression strategy library, connected to the interference suppression decision module, is used to: Storing the correspondence between interference types and suppression strategies; Record the parameter combination and performance of the suppression strategy; The content of the strategy library is optimized according to the feedback of the interference suppression performance evaluation module.

[0016] The beneficial effects of the present invention include: 1. By comprehensively extracting interference features from the time domain, frequency domain, and nonlinearity, the accuracy and comprehensiveness of interference identification are significantly improved, and the adaptability to complex interference environments is increased by more than 40%; 2. The use of manifold learning-based feature extraction and chaotic dynamics-based deep learning classification methods enables sensitive capture of subtle changes in interference signals, improving interference classification accuracy by 50%; 3. Combining feature transformation and encoding mechanisms inspired by group theory, the interference feature representation is made invariant to signal transformations, enhancing the system’s robustness to interference variations. 4. Through a multi-level signal fusion strategy, the complementary advantages of different modal signals are fully utilized to maintain effective communication in severe interference environments, and the overall anti-interference capability of the system is improved by 65%; 5. An interference suppression performance evaluation and feedback optimization mechanism has been established, enabling the system to continuously learn and adapt to new interference environments, thereby improving the long-term reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the overall architecture of the wireless communication anti-interference enhancement system based on multimodal signal fusion of the present invention.

[0018] Figure 2 This is a structural diagram of the interference feature encoding and classification module of the present invention.

[0019] Figure 3 This is a workflow diagram of the multi-dimensional manifold feature extraction unit of the present invention.

[0020] Figure 4 This is a workflow diagram of the feature transformation and encoding unit of the present invention.

[0021] Figure 5 This is a workflow diagram of the classification decision unit of the present invention.

[0022] Figure 6 This is a workflow diagram of the signal fusion module of the present invention.

[0023] Figure 7 Schematic diagram of the interference suppression performance evaluation feedback mechanism of the present invention. DETAILED DESCRIPTION

[0024] Please refer to Figure 1 - Figure 7 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood by those skilled in the art that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0025] like Figure 1 As shown, the wireless communication anti-interference enhancement system based on multimodal signal fusion provided by the present invention includes an intelligent signal transceiver module 1, a signal perception and preprocessing module 2, an interference feature encoding and classification module 3, an interference suppression decision module 4, an interference suppression module 5, a signal fusion module 6, an interference suppression performance evaluation module 7, an interference feature library 8 and an interference suppression strategy library 9.

[0026] The intelligent signal transceiver module 1 is used to receive multi-band, multi-modulation communication signals and perform signal modulation and demodulation. The signal sensing and preprocessing module 2 is connected to the intelligent signal transceiver module 1 and is used to receive the communication signals transmitted by the intelligent signal transceiver module 1, perform time-frequency transformation on the communication signals, and convert the communication signals into the time-frequency domain. The interference feature coding and classification module 3 is connected to the signal sensing and preprocessing module 2 and is used to receive the time-frequency domain signals output by the signal sensing and preprocessing module 2, extract multidimensional interference features from the time-frequency domain signals, perform standardized encoding on the multidimensional interference features, and generate interference type classification results based on the standardized encoding.

[0027] The interference suppression decision module 4 is connected to the interference feature coding and classification module 3 and is configured to receive the interference type classification result generated by the interference feature coding and classification module 3, select an interference suppression strategy based on the interference type classification result, and generate interference suppression parameters. The interference suppression module 5 is connected to the interference suppression decision module 4 and is configured to receive the interference suppression strategy and interference suppression parameters generated by the interference suppression decision module 4 and perform interference suppression on the communication signal based on the interference suppression strategy and interference suppression parameters.

[0028] The signal fusion module 6 is connected to the interference suppression module 5 and is configured to receive the multimodal suppressed signal output by the interference suppression module 5 and perform weighted fusion on the multimodal suppressed signal to generate a fused signal. The interference suppression performance evaluation module 7 is connected to the signal fusion module 6 and is configured to receive the fused signal output by the signal fusion module 6, evaluate the quality of the fused signal, and provide feedback of the performance evaluation results to the interference suppression decision module 4.

[0029] Preferably, the interference feature encoding and classification module 3 includes a multi-dimensional manifold feature extraction unit 31, a feature transformation and encoding unit 32, and a classification decision unit 33. The multi-dimensional manifold feature extraction unit 31 is used to extract interference features from three dimensions: time domain, frequency domain, and nonlinearity; the feature transformation and encoding unit 32 is used to generate standardized codes of the interference features; and the classification decision unit 33 is used to generate interference type classification results.

[0030] In one embodiment of the present invention, the multi-dimensional manifold feature extraction unit 31 of the interference feature coding and classification module 3 includes a time domain feature extraction subunit 311 , a frequency domain feature extraction subunit 312 , a nonlinear feature extraction subunit 313 and a feature fusion subunit 314 .

[0031] The time-domain feature extraction subunit 311 is used to apply multi-scale wavelet decomposition to extract time-domain wavelet coefficients, analyze signal waveform profile features, and calculate time-series-related structural parameters. Specifically, the time-domain feature extraction subunit 311 first performs multi-scale wavelet decomposition on the received signal to generate wavelet coefficients at different scales. In this embodiment, the discrete wavelet transform (DWT) is used to decompose the signal, selecting the Daubechies wavelet basis function with a decomposition layer of 5 to generate the corresponding wavelet coefficients. The wavelet decomposition can be expressed as:

[0032] ,

[0033] in, is the wavelet coefficient, indicating the scale and translation parameters The coefficient value under ; is the input signal, representing discrete time points The signal value at ; is the wavelet basis function, which means that and translation parameters The basis function values ​​under ; is a discrete time index, and its range is the signal length; Represents the scale parameter, which controls the expansion and contraction of the wavelet, and is usually between 1 and 5; Represents the translation parameter, which controls the position of the wavelet.

[0034] In addition, the time domain feature extraction subunit 311 also analyzes the waveform profile characteristics of the signal, including parameters such as peak distribution and zero crossing rate. The zero crossing rate is calculated as follows:

[0035] ,

[0036] in, is the zero-crossing rate, which indicates the number of times the signal crosses the zero level per unit time; is the signal length, which indicates the number of sampling points of the signal; is a symbolic function, when The value is 1 when it is, otherwise -1; For time point The signal value at ; is the time index, ranging from 1 to .

[0037] The frequency domain feature extraction subunit 312 is used to generate a time-frequency distribution diagram of the signal, calculate the energy proportion of each frequency band, and extract channel state information. Specifically, the frequency domain feature extraction subunit 312 uses a short-time Fourier transform (STFT) to generate a time-frequency distribution diagram of the signal. The formula is as follows:

[0038] ,

[0039] in, is the STFT result, indicating the time position and frequency index The time-frequency coefficient at ; is the input signal, indicating the time point The signal value at ; is a window function, used to intercept a fragment of the signal; is the time index, indicating the center position of the window function; is the frequency index, ranging from 0 to ; is the number of FFT points, indicating the frequency resolution; is the imaginary unit, In this embodiment, a Hamming window is selected with a window length of 256 points, an overlap rate of 50%, and a FFT point number of 512 points.

[0040] Based on the time-frequency distribution diagram, calculate the energy proportion of each frequency band:

[0041]

[0042] in, is the energy proportion of the i-th frequency band, indicating the proportion of the energy of this frequency band in the total energy; is the STFT result; is the frequency index set corresponding to the i-th frequency band; is the frequency index; is a fixed time index; is the number of FFT points; Indicates time-frequency points In this system, the frequency band is divided into 10 sub-bands, and the energy distribution characteristics of each sub-band are analyzed.

[0043] At the same time, the frequency domain feature extraction subunit 312 also extracts channel state information, including parameters such as channel frequency response and channel coherence bandwidth, to provide a basis for subsequent interference identification.

[0044] The nonlinear feature extraction subunit 313 is used to calculate the signal entropy parameter, extract the signal's higher-order statistics such as kurtosis and skewness, and analyze the signal's fractal dimension. Specifically, the nonlinear feature extraction subunit 313 calculates the signal's sample entropy (SampEn) using the following formula:

[0045] ,

[0046] in, is the sample entropy value, which quantifies the complexity of the signal; is the embedding dimension, which indicates the length of the comparison pattern and has a value of 2; is the similarity tolerance, which represents the threshold for judging the similarity of two patterns, and is set to 0.2 times the signal standard deviation; is the signal length, indicating the number of sampling points; is the number of pattern matches (dimension is ), indicating the length is The number of pattern matches; is the number of pattern matches (dimension is ), indicating the length is The number of pattern matches; is a natural logarithmic function. Sample entropy can effectively quantify the complexity of the signal. A larger value indicates a more complex signal.

[0047] In addition, the nonlinear feature extraction subunit 313 calculates the kurtosis and skewness of the signal:

[0048] ,

[0049] ,

[0050] Among them, Kurtosis is the kurtosis, which describes the peak degree of the signal distribution; Skewness is the skewness, which describes the asymmetry of the signal distribution; is the signal length, indicating the number of sampling points; For the The signal value of each sampling point; is the signal mean, which represents the average value of all sampling points; is the signal standard deviation, which indicates the degree of dispersion of the signal; The sampling point index ranges from 1 to The kurtosis describes the peak degree of the signal distribution, and the skewness describes the asymmetry of the signal distribution. These high-order statistics can effectively distinguish different types of interference signals.

[0051] The feature fusion subunit 314 is used to receive the features output by the time domain feature extraction subunit 311, the frequency domain feature extraction subunit 312, and the nonlinear feature extraction subunit 313, perform correlation analysis on the features, and generate a multi-dimensional feature matrix. To ensure the comparability of features from different sources, the features of each dimension are first normalized:

[0052] ,

[0053] in, It is a normalized feature, and its value range is usually between; are primitive characteristics and may have different dimensions and ranges; is the feature mean, indicating the feature The average value of is the characteristic standard deviation, indicating the characteristic The degree of dispersion; is the feature index, indicating different feature dimensions.

[0054] After normalization, the feature fusion subunit 314 calculates the correlation matrix between features, identifies redundant features, and reduces the feature dimension through methods such as principal component analysis (PCA), ultimately generating a fused feature vector:

[0055] ,

[0056] in, To fuse the feature vector, it contains feature information of all dimensions; It is the time domain feature vector after dimensionality reduction, and the dimension is usually 10-20; It is the frequency domain feature vector after dimensionality reduction, and the dimension is usually 15-25; is the nonlinear feature vector after dimensionality reduction, and its dimension is usually 5-10. The total dimension of the fused feature vector $F$ is usually controlled between 30-50, which not only retains key information but also avoids the curse of dimensionality.

[0057] In another embodiment of the present invention, the feature transformation and encoding unit 32 of the interference feature encoding and classification module 3 includes a feature transformation subunit 321 , an invariant extraction subunit 322 and a code generation subunit 323 .

[0058] The feature transformation subunit 321 is configured to receive the multidimensional feature matrix output by the multidimensional manifold feature extraction unit 31 and apply geometric and orthogonal transformations to the multidimensional feature matrix to generate a normalized feature vector. Specifically, the feature transformation subunit 321 first applies geometric transformations, including translation, rotation, and scaling, to the feature matrix to enhance the robustness of the feature representation.

[0059] The translation transformation can be expressed as:

[0060] ,

[0061] in, is the transformed feature matrix with the same dimension as F; F is the original feature matrix with dimension d×n, where d is the feature dimension and n is the number of samples; T is the translation vector with dimension d×1, and each element is usually set to the mean of the corresponding dimension feature.

[0062] The rotation transformation can be expressed as:

[0063] ,

[0064] in, is the transformed feature matrix, with the same dimension as same; is the original feature matrix; is the rotation matrix with dimension , is an orthogonal matrix, satisfying , is the identity matrix. For two-dimensional features, It can be expressed as:

[0065] ,

[0066] in, is the rotation angle, the range is ; and Angle In this system, the rotation angle is adaptively selected according to the characteristic distribution characteristics, usually arrive between.

[0067] The scaling transformation can be expressed as:

[0068] ,

[0069] in, is the transformed feature matrix, with the same dimension as same; is the original feature matrix; is the scaling matrix with dimensions For a diagonal scaling matrix, it can be expressed as:

[0070] ,

[0071] in, For the The scaling factor of each dimension is usually in the range between; is the feature dimension. In this system, the scaling factor is adaptively adjusted according to the feature importance. The scaling factor of important features is larger (such as 1.5-2), and the scaling factor of minor features is smaller (such as 0.5-0.8).

[0072] In addition, the feature transformation subunit 321 also applies orthogonal transformation to project the features onto the optimal orthogonal basis to enhance the representation capability. The main method used is the singular value decomposition (SVD):

[0073] ,

[0074] Where F is the feature matrix, with dimensions d×n; U is the left singular vector matrix, with dimensions d×d, where the columns are orthogonal; Σ is the diagonal matrix, with dimensions d×n, where the diagonal elements are singular values, arranged in descending order; V^T is the transpose of the right singular vector matrix, with dimensions n×n, where the rows are orthogonal; d is the feature dimension; and n is the number of samples. By retaining features corresponding to larger singular values, feature dimensionality can be effectively reduced while retaining key information. In this system, the top k singular values ​​that explain 90% of the variance are typically retained, where k is typically much smaller than the original dimension d.

[0075] The invariant extraction subunit 322 is used to extract geometric invariants, statistical invariants, and topological invariants from the normalized feature vectors. Geometric invariants include translation, rotation, and scaling invariants, which can be calculated by normalized moments and moment ratios:

[0076] ,

[0077] ,

[0078] in, is the original moment, representing the p-order and q-order moments of the feature distribution; is the normalized moment, which is scale invariant; x and y are the coordinates of the feature space; f(x, y) is the eigenvalue at the point (x, y); p and q are the orders of the moment, which usually do not exceed 3; is the zero-order moment, which represents the total amount of the feature.

[0079] Statistical invariants include invariants of distributional properties, such as Gaussian mixture model parameters:

[0080] ,

[0081] in, Characterized by The probability density function of is the number of mixed components, which is set according to the complexity of the interference and is usually 3-5; For the The weights of the mixture components satisfy , ; The mean is , the covariance matrix is Multivariate Gaussian distribution of ; is the eigenvector; The index of the mixture component, ranging from 1 to These parameters are robust to signal amplitude variations.

[0082] Topological invariants include connectivity, Euler features, etc., which are used to capture the topological structural characteristics of the feature space. The code generation subunit 323 is used to perform vector quantization on geometric invariants, statistical invariants, and topological invariants, perform feature compression, and generate standardized codes of interference features.

[0083] Vector quantization uses the K-means clustering method to divide the feature space into a limited number of representative vectors:

[0084] ,

[0085] in, represents the minimization objective function; For the feature vectors; For the Representative vectors (codebook); is the square of the Euclidean distance, which indicates the similarity between the feature vector and the representative vector; is the number of features; is the feature index, ranging from 1 to ; represents the vector index, ranging from 1 to the codebook size. In this system, the codebook size is adaptively adjusted according to the complexity of the interference type and is usually set between 16 and 64.

[0086] Feature compression uses methods such as principal component analysis (PCA) and autoencoders to reduce feature dimensions while retaining key information. The resulting standardized encoding is:

[0087] ,

[0088] in, To standardize the encoding vector, the dimension is usually 20-30; is the encoding vector of the geometric invariant, the dimension is usually 8-12; The encoding vector of the statistical invariant, the dimension is usually 8-10; is the encoding vector of the topological invariant, and its dimension is usually 4-8.

[0089] In another embodiment of the present invention, the classification decision unit 33 of the interference feature encoding and classification module 3 includes a feature learning network 331, a chaos prediction subunit 332, an attention mechanism subunit 333 and a probability decision subunit 334.

[0090] The feature learning network 331 is used to receive the standardized codes output by the feature transformation and encoding unit 32 and extract deep feature representations. The feature learning network 331 uses a deep neural network structure, including convolutional layers, recurrent layers, and fully connected layers to capture multi-scale feature patterns in the codes. The convolutional layer formula is as follows:

[0091] ,

[0092] in, For the Layer output, the dimension is the feature map size of the layer; For the The convolution kernel parameters of the layer, the dimension is convolution kernel size × number of convolution kernels; For the The output of the layer is used as the input of the current layer; * indicates the convolution operation; For the The bias term of the layer has the same dimension as the number of output feature maps; As the activation function, the ReLU function is used: , for input Take the maximum value between it and 0; is the layer index, indicating the depth of the network. In this system, the convolution kernel size is 3×3, the stride is 1, the padding is SAME mode, and the number of convolution layers is 3-5.

[0093] The recurrent layer uses long short-term memory (LSTM) units to capture temporal dependencies:

[0094] ,

[0095] ,

[0096] ,

[0097] ,

[0098] ,

[0099] in, It is the forget gate, which controls the degree of retention of the unit state at the previous moment. Its dimension is the same as the number of hidden units. It is the input gate, which controls the degree of reception of the current input information. Its dimension is the same as the number of hidden units. The output gate controls the output degree of the unit state, and its dimension is the same as the number of hidden units; is the unit state, which stores long-term memory and has the same dimension as the number of hidden units; is the hidden state, which is the output at the current moment and has the same dimension as the number of hidden units; is the time step The input vector of is the time step The hidden state of 、 、 、 is the input weight matrix; 、 、 、 is the cyclic weight matrix; 、 、 、 is the bias vector; is the sigmoid function, ,map the input to the (0,1) interval; and is the tanh function, , maps the input to the interval (-1,1); o represents the Hadamard product, that is, the corresponding elements are multiplied; is the time step index. In this system, the number of LSTM hidden units is 128 and the number of layers is 2.

[0100] The chaos prediction subunit 332 is used to map the deep feature representation to the phase space, analyze the dynamic trajectory of the interference feature, and predict the impact of the change of the interference parameter. First, the chaos prediction subunit 332 maps the feature vector to the phase space through the time delay embedding method:

[0101]

[0102] in, is the phase space trajectory point, and the dimension is ; is the feature time series, and the dimension is the feature sequence length; is the time delay, which indicates the interval between adjacent sampling points, and is usually set to 1-5; is the embedding dimension, which indicates the dimension of the constructed phase space, and is usually set to 3-5; is the time index, ranging from 1 to , is the sequence length. In this system, the time delay The embedding dimension is determined by the minimum mutual information Determined by the false nearest neighbor method, usually The value range is 3-5.

[0103] Based on the phase space trajectory, the chaos prediction subunit 332 calculates the Lyapunov exponent to quantify the degree of chaos in the system:

[0104]

[0105] in, is the Lyapunov index, which measures the sensitivity of the system to the initial conditions; is the time variable, which represents the time scale of system evolution; For time A small perturbation vector in the phase space; is the initial perturbation vector; is the Euclidean norm of the disturbance vector, indicating the magnitude of the disturbance; is the natural logarithm function; It represents the limit when time approaches infinity. A positive maximum Lyapunov exponent indicates that the system has chaotic characteristics, and the critical value is usually set to 0.01.

[0106] By analyzing the phase space trajectory and chaotic characteristics, the chaos prediction subunit 332 can predict the impact of interference parameter changes on classification results and improve the system's adaptability to interference variations.

[0107] The attention mechanism subunit 333 is used to calculate the contribution of each feature to the decision, dynamically adjust the feature weight, and highlight the key discriminant features. The attention mechanism formula is as follows:

[0108] ,

[0109] ,

[0110] ,

[0111] in, is the attention score of the i-th feature, indicating the importance of the feature; is the attention vector, and its dimension is the number of attention hidden units; is the weight matrix, the dimension is the number of attention hidden units * feature dimension; is the i-th eigenvector, and its dimension is the length of the eigenvector; is the bias vector, the dimension is the number of attention hidden units; tanh is the hyperbolic tangent function, which maps the input to interval; Represents a vector The transpose of is the attention weight of the i-th feature, indicating the importance of the feature in decision making, and its value range is , and satisfy ; is the natural exponential function; is the weighted context vector, which represents the feature representation after considering attention; is the number of features; and is the feature index, ranging from 1 to Through the attention mechanism, the system can adaptively focus on the most important features for current interference identification and improve classification accuracy.

[0112] The probability decision subunit 334 is used to generate the probability distribution of interference types based on the deep feature representation, dynamic trajectory analysis results and feature weights, and form the final interference type classification result. The probability decision adopts the softmax function:

[0113] ,

[0114] in, For a given feature , classified into categories The probability range is , and satisfy ; For category The logit value of the model indicates the class The original prediction score of for Natural index; is the total number of interference types, which is usually 5-10 different types in this system; and is the category index, ranging from 1 to In this system, the interference types include narrowband interference, broadband interference, pulse interference, swept frequency interference and multipath interference.

[0115] To improve classification reliability, the probability decision subunit 334 also calculates decision entropy to evaluate the certainty of the classification result:

[0116] ,

[0117] in, is the decision entropy, which quantifies the uncertainty of the classification results and ranges from 0 to For category The predicted probability of is the logarithmic function with base 2; is the total number of interference types; is the category index, ranging from 1 to When the entropy value exceeds a preset threshold (e.g., 0.5), the classification result is uncertain, and the system initiates a secondary classification process or requests manual intervention. The final classification result includes an interference type label, a confidence level, and a list of key features.

[0118] In a preferred embodiment of the present invention, the intelligent signal transceiver module 1 includes a multi-band receiving unit 11 , a multi-modulation processing unit 12 and a signal modulation and demodulation unit 13 .

[0119] The multi-band receiving unit 11 is used to receive low-frequency, medium-frequency, and high-frequency signals. Specifically, low-frequency signals cover the 30-300kHz frequency band and have strong anti-interference and penetration capabilities, making them suitable for long-distance communications. Medium-frequency signals cover the 300kHz-3MHz frequency band, balancing transmission distance and data rate. High-frequency signals cover the 3-30MHz frequency band and are suitable for medium- and long-distance communications. The multi-band receiving unit 11 utilizes a broadband RF front-end design, supporting simultaneous multi-band reception and improving the system's adaptability to interference from different frequency bands.

[0120] The multi-modulation processing unit 12 processes signals using modulation schemes such as FSK, PSK, DSSS, and OFDM. FSK (Frequency Shift Keying) transmits information through frequency shifts, offering strong noise immunity. PSK (Phase Shift Keying) transmits information through phase shifts, offering high spectral efficiency. DSSS (Direct Sequence Spread Spectrum) utilizes spreading codes to mitigate narrowband interference, improving the system's anti-interference capabilities. OFDM (Orthogonal Frequency Division Multiplexing) employs multi-carrier parallel transmission, effectively combating frequency-selective fading. The multi-modulation processing unit 12 supports adaptive switching among these various modulation schemes, selecting the optimal modulation scheme based on channel conditions and interference environment.

[0121] The signal modulation and demodulation unit 13 performs signal modulation and demodulation. On the transmitting end, it generates a modulated signal based on the data to be transmitted and the selected modulation method. On the receiving end, it performs demodulation operations based on the modulation method of the received signal to recover the original data. The signal modulation and demodulation unit 13 utilizes software-defined radio (SDR) technology, enabling flexible switching between different modulation methods through software configuration, improving the system's adaptability and scalability.

[0122] In another preferred embodiment of the present invention, the signal sensing and preprocessing module 2 includes a bandpass filtering unit 21 , an automatic gain control unit 22 , a digital downsampling unit 23 , a time-frequency conversion unit 24 and a signal quality assessment unit 25 .

[0123] Bandpass filter unit 21 is used to limit the signal frequency band and suppress out-of-band interference. The bandpass filter design uses a Chebyshev filter, with passband ripple controlled within 0.5dB and stopband attenuation greater than 50dB, effectively suppressing out-of-band interference and improving the signal-to-noise ratio.

[0124] The automatic gain control unit 22 dynamically adjusts the signal level. It employs a digital automatic gain control algorithm to adjust the gain in real time based on the signal amplitude, keeping the signal amplitude within an appropriate range and avoiding overload or underload. With a gain adjustment step size of 0.5dB and a response time of less than 10ms, it effectively handles rapid changes in signal amplitude.

[0125] Digital downsampling unit 23 is used to reduce the sampling rate. Based on actual needs, digital filtering and decimation are used to reduce the signal sampling rate to the minimum required for processing, thereby reducing the computational burden and improving the system's real-time performance. In this system, the original sampling rate is 50 MHz, which can be reduced to 1-10 MHz depending on the characteristics of different modal signals.

[0126] The time-frequency transform unit 24 transforms the signal into the time-frequency domain. It primarily uses methods such as the short-time Fourier transform (STFT) and wavelet transform (WT) to convert the signal from the time domain to the time-frequency domain, facilitating subsequent feature extraction and interference identification. STFT parameter settings include: a Hamming window with a window length of 256 points, a 50% overlap ratio, and a FFT of 512 points. The wavelet transform uses the Daubechies wavelet basis with a decomposition level of 5.

[0127] The signal quality assessment unit 25 is used to estimate signal energy levels, evaluate signal dynamic range, and detect abnormal frequency domain components. It primarily calculates parameters such as average power, peak-to-average power ratio, and spectral purity, providing a basis for subsequent interference identification and suppression. Signal quality assessment results include indicators such as signal-to-noise ratio (SNR), signal-to-interference ratio (SIR), and signal-to-interference-plus-noise ratio (SINR). The assessment cycle is 10ms.

[0128] In another preferred embodiment of the present invention, the interference suppression module 5 includes a frequency domain cancellation unit 51 , a space domain cancellation unit 52 , a time domain cancellation unit 53 and a joint suppression unit 54 .

[0129] Frequency-domain cancellation unit 51 performs adaptive notch filtering, spectral subtraction, and wavelet threshold denoising. Adaptive notch filtering primarily targets narrowband interference, effectively suppressing interference signals in specific frequency bands by adjusting filter parameters in real time through an adaptive algorithm. The notch filter's center frequency and bandwidth are automatically adjusted based on the interference characteristics. The filter has an order range of 4-8 and a suppression depth greater than 40dB.

[0130] Spectral subtraction is an interference suppression method based on spectrum estimation, mainly targeting additive interference. The formula is as follows:

[0131] ,

[0132] in, is the estimated signal spectrum, indicating the frequency The estimated value of the signal amplitude at ; is the observed signal spectrum, indicating the frequency The observed signal amplitude at ; is the estimated interference spectrum, indicating the frequency The estimated value of the interference amplitude at ; is the oversubtraction factor, which controls the scaling degree of the interference estimation and is set to 1.2; The spectrum lower limit factor is used to prevent excessive suppression and its value is 0.1; is a frequency variable; For maximum value operation, ensure the result is non-negative.

[0133] Wavelet threshold denoising is mainly used to target broadband noise interference. It decomposes the signal into coefficients of different scales through wavelet transform, performs threshold processing on the coefficients, and then reconstructs the signal. Threshold selection formula:

[0134] ,

[0135] in, is the threshold, which is used to determine whether the wavelet coefficient is a signal component or a noise component; is the noise standard deviation, which indicates the intensity of the noise and is usually estimated by the MAD (median absolute deviation) method: ,in is the finest scale wavelet coefficient; is the signal length, indicating the number of sampling points; is the natural logarithm; is the universal threshold factor. The threshold processing uses the soft threshold method:

[0136] ,

[0137] in, is the processed wavelet coefficient, which is used for signal reconstruction; is the original wavelet coefficient; is the threshold; is a symbolic function, when When the value is 1, The value is -1 when When the value is 0; is the absolute value of the wavelet coefficient.

[0138] Spatial cancellation unit 52 is used to perform beamforming, spatial filtering, and interference source directional suppression. Beamforming primarily utilizes a multi-antenna array to adjust the phase and amplitude of each antenna signal to form a specific beam pattern, enhancing the gain in the direction of the desired signal and suppressing the gain in the direction of the interfering signal. The beamforming weight calculation uses the minimum variance distortionless response (MVDR) algorithm:

[0139] ,

[0140] in, is the beamforming weight vector, the dimension is the number of antennas ; is the received signal covariance matrix, with dimension , represents the correlation between the antenna signals; is the steering vector of the desired signal direction, with dimension , indicating that the expected signal is from the direction The phase relationship of the signals arriving at each antenna; for The conjugate transpose of is the inverse matrix of the covariance matrix; is the expected signal incident angle, which is usually obtained through a signal direction estimation algorithm (such as MUSIC and ESPRIT).

[0141] Spatial filtering separates signals from interference by using spatial characteristics, primarily using eigendecomposition of the spatial covariance matrix to separate the signal space from the interference space. Directional interference suppression targets directional interference sources by creating a spatial zero point in the interference direction, effectively suppressing interference signals from a specific direction.

[0142] The time-domain cancellation unit 53 is used to perform adaptive filtering, predictor structure, and timing pattern cancellation. Adaptive filtering mainly uses the least mean square error (LMS) and recursive least squares (RMS) algorithms to adjust the filter coefficients in real time based on the error signal and adaptively track the characteristics of the interference signal. The LMS algorithm update formula is:

[0143] ,

[0144] in, is the updated filter coefficient vector, the dimension is the filter order ; is the current filter coefficient vector; is the step size parameter, which controls the convergence speed and stability, and its value range is 0.01-0.1; is the error signal, which represents the difference between the expected output and the actual output; is the input signal vector, including the current time and the previous Input samples at time instants; is the time index.

[0145] The predictor architecture predicts the interference waveform and generates a cancellation signal based on the temporal correlation of the interference signal. Sequential pattern cancellation targets interference with repetitive patterns, identifying and extracting the interference pattern and generating a corresponding cancellation signal.

[0146] The joint suppression unit 54 coordinates the operations of the frequency-domain cancellation unit 51, the spatial-domain cancellation unit 52, and the time-domain cancellation unit 53 based on the interference suppression strategy and parameters. It selects the optimal combination of suppression methods based on the interference type and characteristics. For example, for narrowband interference, frequency-domain cancellation is primarily used; for spatially directional interference, spatial-domain cancellation is primarily used; and for time-varying interference, time-domain cancellation is primarily used. The joint suppression unit 54 also dynamically adjusts the parameters and resource allocation of each suppression unit based on the interference suppression effect and system resource availability to optimize the overall suppression effect.

[0147] In yet another preferred embodiment of the present invention, the signal fusion module 6 includes a signal quality evaluation unit 61 , a weight distribution unit 62 , a linear fusion unit 63 , a nonlinear fusion unit 64 and a post-fusion processing unit 65 .

[0148] The signal quality evaluation unit 61 is used to evaluate the quality indicators of each modal signal. It mainly calculates the signal-to-noise ratio, bit error rate, signal integrity and other indicators of the signal to provide a basis for subsequent fusion. The signal-to-noise ratio calculation formula is:

[0149] ,

[0150] Where SNR is the signal-to-noise ratio, expressed in decibels (dB); is the signal power in watts (W); is the noise power in watts (W); is a logarithmic function with base 10; 10 is the scaling factor for converting to decibels. The bit error rate (BER) is calculated by sending a known sequence and comparing the received results:

[0151] ,

[0152] Where BER is the bit error rate, which indicates the ratio of bits with transmission errors to the total bits. It is dimensionless and ranges from 0 to 1. The number of error bits is the number of bits in the received data that do not match the transmitted data. The total number of bits is the total number of bits transmitted.

[0153] The weight allocation unit 62 is used to allocate fusion weights to each modal signal based on the quality index. The weight allocation adopts an adaptive weighting method based on the signal-to-noise ratio. The signal with a higher signal-to-noise ratio receives a greater weight:

[0154] ,

[0155] in, For the The weight of a modal signal indicates the importance of the signal in the fusion, and its value range is (0,1) and satisfies ; For the The signal-to-noise ratio of each modal signal, in linear scale (not decibels); is the number of modes, indicating the total number of signal modes involved in fusion; and is the modal index, ranging from 1 to .

[0156] The linear fusion unit 63 is used to perform weighted averaging, maximum ratio merging and selective merging. Weighted averaging is the most basic linear fusion method, which linearly combines the modal signals according to their weights:

[0157] ,

[0158] in, is the fusion signal, indicating the time point The signal value at ; For the The modal signal at time The value at For the The weight of each modal signal; is the number of modes; is the modal index, ranging from 1 to ; is a time variable. Maximum ratio combining (MRC) is an optimal linear fusion method for homologous multipath signals. The fusion weight is proportional to the conjugate of the signal amplitude and channel phase:

[0159] ,

[0160] in, is the fusion signal; For the The signal of each path; For the The signal amplitude of each path; For the The signal phase of each path; is the phase correction factor, is the imaginary unit, ; is the number of paths; is the path index, ranging from 1 to ; is a time variable. The MRC method can maximize the signal-to-noise ratio after fusion.

[0161] Selective combining (SC) selects the signal path with the best quality:

[0162] ,

[0163] in, is the fusion signal; is the selected signal path; is the index of the selected path; Indicates Get the index of the maximum value ; For the The signal-to-noise ratio of each path; is the time variable.

[0164] The nonlinear fusion unit 64 is used to perform median filter fusion, fusion based on sorting statistics, and fuzzy logic fusion. The median filter fusion takes the median of each modal signal and can effectively suppress the influence of outliers:

[0165] ,

[0166] in, is the fused signal; median is the median operation, which returns the middle value of a set of values; is the modal signal at time point The value at is the number of modes; is the time variable.

[0167] Fusion based on sorted statistics uses weighted combinations of sorted signals, such as alpha-trimmed means:

[0168] ,

[0169] in, is the fusion signal; After sorting A signal is coming After sorting by value from smallest to largest values; It is the pruning parameter, which indicates the number of signals deleted from both ends, and is usually taken as , Indicates rounding down; is the number of modes; The number of signals to be retained; is the sorted index, ranging from arrive ; is the time variable.

[0170] Fuzzy logic fusion combines signals based on fuzzy rules, enabling it to handle situations with high uncertainty and ambiguity. Fuzzy rules are set based on signal quality metrics, such as assigning a high weight to signals with high SNR and low BER.

[0171] The post-fusion processing unit 65 is responsible for performing channel re-estimation, signal-to-noise ratio calculation, and signal reshaping. Channel re-estimation updates the channel state information after fusion, providing an accurate channel model for subsequent processing. Signal-to-noise ratio calculation assesses the quality of the fused signal and verifies the fusion effect. Signal reshaping shapes and normalizes the fused signal in preparation for demodulation.

[0172] In one embodiment of the present invention, the interference suppression performance evaluation module 7 includes a signal-to-noise ratio calculation unit 71 , a bit error rate evaluation unit 72 , a packet error rate calculation unit 73 , a performance index comprehensive evaluation unit 74 and a feedback generation unit 75 .

[0173] The signal-to-noise ratio calculation unit 71 is used to calculate the signal-to-noise ratio of the fused signal. The signal-to-noise ratio of the signal is estimated using a method based on the signal-to-noise power ratio:

[0174] ,

[0175] Where SNR is the signal-to-noise ratio, expressed in decibels (dB); is the signal power in watts (W); is the noise power, in watts (W); is a logarithmic function with base 10; 10 is the scaling factor for conversion to decibels. In the absence of a reference signal, the signal and noise components are separated by the time-frequency characteristics of the signal, achieving blind estimation of the signal-to-noise ratio.

[0176] The bit error rate evaluation unit 72 is used to evaluate the accuracy of data transmission. By inserting a known sequence or using the error correction code detection capability, the bit error rate of the received data is calculated:

[0177] BER ,

[0178] BER is the bit error rate, which represents the ratio of transmitted bits with errors to the total number of bits. It is dimensionless and ranges from 0 to 1. The number of error bits is the number of bits in the received data that do not match the transmitted data. The total number of bits is the total number of bits transmitted. In practical systems, the bit error rate can be assessed by sending test sequences or using the checksum mechanism of data packets.

[0179] The packet error rate calculation unit 73 is used to calculate the error rate of data packet transmission. Based on the verification result of the data packet, the packet error rate is calculated:

[0180] ,

[0181] PER is the packet error rate, which represents the ratio of transmitted packets with errors to the total number of packets. It is dimensionless and ranges from 0 to 1. Error packets are the number of received packets that failed verification. Total packets are the total number of transmitted packets. The packet error rate reflects the communication reliability of the system at the packet level and is an important indicator for evaluating system performance.

[0182] The performance index comprehensive evaluation unit 74 is used to comprehensively analyze the signal-to-noise ratio, bit error rate and packet error rate to generate a performance evaluation result. A weighted scoring method is used to comprehensively evaluate various indicators:

[0183] ,

[0184] Among them, Score is a comprehensive performance score, which indicates the overall performance of the system. The value range is usually between [-1, 1]. The higher the value, the better the performance. SNR is the normalized signal-to-noise ratio, calculated as SNR ,SNR The maximum signal-to-noise ratio reference value set for the system (such as 30dB); BER is the normalized bit error rate, calculated as BER is the reference bit error rate (e.g. is the normalized packet error rate, which is calculated as ,PER is the reference packet error rate (e.g. 10- 2 ) 、 and is the weight coefficient, satisfying , the default value is 1 / 3.

[0185] Feedback generation unit 75 converts performance evaluation results into adjustment recommendations and feeds them back to interference suppression decision module 4. Based on the performance evaluation results, specific adjustment recommendations are generated, such as adjusting suppression parameters or switching suppression strategies. The feedback cycle is 50ms to ensure that the system can respond promptly to performance changes.

[0186] In another embodiment of the present invention, the system further includes an interference feature library 8 and an interference suppression strategy library 9 .

[0187] The interference signature library 8 is connected to the interference signature encoding and classification module 3 and is used to store the mapping between interference signatures and interference types. It receives newly identified interference signatures from the interference signature encoding and classification module 3 and updates the signature library contents. The interference signature library 8 uses a key-value pair structure, with the key being the interference signature code and the value being the interference type and related attributes. The signature library supports fast query and incremental updates, initially containing signature templates for common interference types such as narrowband interference, broadband interference, pulse interference, and swept-frequency interference. As the system operates, the signature library is continuously learned and updated to improve the accuracy and coverage of interference identification.

[0188] The interference suppression strategy library 9 is connected to the interference suppression decision module 4 and is used to store the correspondence between interference types and suppression strategies, record the parameter combinations and performance of the suppression strategies, and optimize the strategy library content based on feedback from the interference suppression performance evaluation module 7. The strategy library contains the optimal suppression strategies and parameter combinations for different interference types, such as adaptive notch filtering for narrowband interference, wavelet threshold denoising for wideband interference, and beamforming for spatially directional interference. The strategy library also records the performance of each strategy in different scenarios, providing a basis for decision-making. Based on performance evaluation feedback, the strategy library is continuously optimized and adjusted to adapt to the ever-changing interference environment.

Claims

1. A wireless communication anti-interference enhancement system based on multimodal signal fusion, characterized in that: include: Intelligent signal transceiver module for: Receive multi-band and multi-modulation communication signals; Perform modulation and demodulation of signals; The signal sensing and preprocessing module is connected to the intelligent signal transceiver module and is used to: Receiving the communication signal transmitted by the intelligent signal transceiver module; performing time-frequency transformation on the communication signal; transforming the communication signal into the time-frequency domain; The interference feature coding and classification module is connected to the signal sensing and preprocessing module and is used to: Receiving the time-frequency domain signal output by the signal sensing and preprocessing module; extracting multidimensional interference features from the time-frequency domain signal; Performing standardized encoding on the multi-dimensional interference features; generating an interference type classification result based on the standardized coding; The interference suppression decision module is connected to the interference feature coding and classification module and is used to: Receiving the interference type classification result generated by the interference feature coding and classification module; selecting an interference suppression strategy based on the interference type classification result; generating interference suppression parameters; An interference suppression module, connected to the interference suppression decision module, is used to: receiving the interference suppression strategy and interference suppression parameters generated by the interference suppression decision module; performing interference suppression on a communication signal based on the interference suppression strategy and the interference suppression parameters; The signal fusion module is connected to the interference suppression module and is used to: Receiving the multimodal suppressed signal output by the interference suppression module; Performing weighted fusion on the multimodal suppressed signals; generating a fusion signal; The interference suppression performance evaluation module is connected to the signal fusion module and is used to: receiving the fusion signal output by the signal fusion module; evaluating a quality indicator of the fusion signal; Feedback of the performance evaluation result to the interference suppression decision module; Among them, the interference feature coding and classification module includes a multidimensional manifold feature extraction unit, a feature transformation and coding unit and a classification decision unit. The multidimensional manifold feature extraction unit is used to extract interference features from three dimensions: time domain, frequency domain and nonlinearity. The feature transformation and coding unit is used to generate standardized coding of interference features. The classification decision unit is used to generate interference type classification results.

2. The system according to claim 1, wherein: The multi-dimensional manifold feature extraction unit of the interference feature coding and classification module includes: The time domain feature extraction subunit is used to: Apply multi-scale wavelet decomposition to extract time domain wavelet coefficients; Analyze signal waveform profile characteristics; Calculate timing-related structural parameters; Frequency domain feature extraction subunit, used for: Generate signal time-frequency distribution diagram; Calculate the energy proportion of each frequency band; extracting channel state information; Non-linear feature extraction subunit, used for: Calculate the signal entropy parameter; Extract high-order statistics such as signal kurtosis and skewness; Analyze the fractal dimension of the signal; Feature fusion subunit, used for: receiving features output by the time domain feature extraction subunit, the frequency domain feature extraction subunit, and the nonlinear feature extraction subunit; performing correlation analysis on the features; Generate a multidimensional feature matrix.

3. The system according to claim 1, wherein: The feature conversion and encoding unit of the interference feature coding and classification module includes: Feature transformation subunit, used to: Receiving the multidimensional feature matrix output by the multidimensional manifold feature extraction unit; Applying geometric transformation and orthogonal transformation to the multidimensional feature matrix; Generate normalized feature vectors; Invariant extraction subunit, used to: extracting geometric invariants from the normalized eigenvectors; Extract statistical invariants; Extract topological invariants; The code generation subunit is used to: performing vector quantization on the geometric invariant, the statistical invariant, and the topological invariant; Perform feature compression; Generates a standardized encoding of the interference features.

4. The system according to claim 1, wherein: The classification decision unit of the interference feature coding and classification module includes: Feature learning networks for: Receiving the standardized code output by the feature transformation and encoding unit; Extract deep feature representation; Chaos prediction subunit, used for: Mapping the deep feature representation to phase space; Analyze the dynamic trajectory of interference features; predict the effects of changes in disturbance parameters; Attention mechanism subunit, used to: Calculate the contribution of each feature to the decision; Dynamically adjust feature weights; Highlight key discriminant features; Probabilistic decision subunit, used to: generating a probability distribution of interference types based on the deep feature representation, the dynamic trajectory analysis results, and the feature weights; Form the final interference type classification result.

5. The system according to claim 1, wherein: The intelligent signal transceiver module is applicable to a variety of frequency bands and modulation methods, including: A multi-band receiving unit for receiving low-frequency, medium-frequency and high-frequency signals; Multi-modulation processing unit, used to process signals of modulation modes such as FSK, PSK, DSSS and OFDM; The signal modulation and demodulation unit is used to perform signal modulation and demodulation.

6. The system according to claim 1, wherein: The signal perception and preprocessing module includes: Bandpass filter unit, used to limit the signal frequency band and suppress out-of-band interference; Automatic gain control unit for dynamically adjusting signal levels; A digital downsampling unit for reducing the sampling rate; A time-frequency conversion unit, used to convert the signal into the time-frequency domain; The signal quality assessment unit is used to estimate the signal energy level, evaluate the signal dynamic range and detect abnormal components in the frequency domain.

7. The system according to claim 1, wherein: The interference suppression module includes: a frequency-domain cancellation unit for performing adaptive notch filtering, spectral subtraction, and wavelet threshold denoising; A spatial cancellation unit that performs beamforming, spatial filtering, and interference source directional suppression; a time-domain cancellation unit for performing adaptive filtering, predictor construction, and temporal pattern cancellation; A joint suppression unit is used to coordinate the work of the frequency domain cancellation unit, the space domain cancellation unit and the time domain cancellation unit according to the interference suppression strategy and the interference suppression parameters.

8. The system according to claim 1, wherein: The signal fusion module includes: A signal quality evaluation unit, used to evaluate the quality indicators of each modal signal; a weight allocation unit, configured to allocate a fusion weight to each modal signal based on the quality indicator; Linear fusion unit for performing weighted averaging, maximum ratio combining, and selective combining; Nonlinear fusion unit for performing median filter fusion, ranking statistic-based fusion, and fuzzy logic fusion; The post-fusion processing unit is used to perform channel re-estimation, signal-to-noise ratio calculation and signal reshaping.

9. The system according to claim 1, wherein: The interference suppression performance evaluation module includes: a signal-to-noise ratio calculation unit, used for calculating the signal-to-noise ratio of the fusion signal; a bit error rate evaluation unit, used to evaluate the accuracy of data transmission; a packet error rate calculation unit, used to calculate the error rate of data packet transmission; a performance indicator comprehensive evaluation unit, configured to comprehensively analyze the signal-to-noise ratio, the bit error rate, and the packet error rate to generate a performance evaluation result; A feedback generating unit is used to convert the performance evaluation result into an adjustment suggestion and feed it back to the interference suppression decision module.

10. The system according to claim 1, wherein: Also includes: The interference feature library is connected to the interference feature encoding and classification module and is used to: Storing the mapping relationship between interference characteristics and interference types; receiving a newly identified interference feature from the interference feature encoding and classification module; Update the feature library content; An interference suppression strategy library, connected to the interference suppression decision module, is used to: Storing the correspondence between interference types and suppression strategies; Record the parameter combination and performance of the suppression strategy; The content of the strategy library is optimized according to the feedback of the interference suppression performance evaluation module.

Images