Integrated airborne communication antenna device and communication method

Abstract

The present invention relates to an integrated airborne communication antenna device and a communication method. The method: CSI data acquisition and spatiotemporal focus modulation are performed on the communication channel to obtain CSI spatiotemporal feature data; residual vector quantization operation is performed to generate a quantized data stream, and recursive coding processing is performed to obtain a target data stream; the target data stream is mapped to a three-dimensional tensor space, a real-valued tensor model is constructed, and iterative operations are performed to obtain a CSI feature vector; the CSI feature vector is decomposed by a fractional wavelet transform to obtain a radio frequency feature set, and the radio frequency feature set is input into a federated fractional scattering network for feature processing to obtain an identity authentication result; the amplitude and phase weight coefficients of the antenna array are calculated to generate beam parameters, and compensation matrix operations are performed on the beam parameters according to the CSI feature vector to obtain a precoded signal. The present invention realizes the dynamic adjustment of antenna beam parameters and improves the accuracy and anti-interference capability of beamforming.

Description

Technical Field

[0001] The present invention relates to the technical field of airborne communications, and in particular to an integrated airborne communication antenna device and a communication method. Background Art

[0002] Due to the high-speed mobility and complex electromagnetic environment of airborne platforms, traditional communication antenna equipment faces severe challenges in channel state information collection, feature extraction and identity authentication. Especially in high-dynamic scenarios, the rapid change of channel state information leads to unstable communication quality, and traditional CSI collection and processing methods are difficult to meet actual needs.

[0003] In the process of airborne communication, precise control and directional optimization of antenna beams are key factors in ensuring communication quality. In the existing technology, due to the lack of effective feature extraction and authentication mechanisms, it is difficult to accurately obtain channel feature information, resulting in insufficient beamforming accuracy and susceptibility to malicious interference and deception attacks, which seriously affects communication security and reliability. At the same time, traditional channel feature processing methods have high computational complexity and are difficult to meet the needs of real-time communication. Summary of the invention

[0004] The main purpose of the present invention is to provide an integrated airborne communication antenna device and a communication method. The present invention realizes the dynamic adjustment of antenna beam parameters and improves the accuracy and anti-interference capability of beamforming.

[0005] To achieve the above object, the present invention provides a communication method of an integrated airborne communication antenna device, comprising the following steps:

[0006] Perform CSI data collection and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data;

[0007] Performing a residual vector quantization operation on the CSI spatiotemporal feature data to generate a quantized data stream, and performing recursive encoding processing on the quantized data stream to obtain a target data stream;

[0008] Mapping the target data stream to a three-dimensional tensor space, constructing a real-valued tensor model, and applying unitary parallel factor decomposition and regularized least squares iterative operations to the real-valued tensor model to obtain a CSI feature vector;

[0009] Performing fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and inputting the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result;

[0010] The amplitude and phase weight coefficients of the antenna array are calculated according to the identity authentication result and the target data stream to generate beam parameters, and a compensation matrix operation is performed on the beam parameters according to the CSI eigenvector to obtain a precoded signal.

[0011] The present invention also provides an integrated airborne communication antenna device, comprising:

[0012] An acquisition module is used to perform CSI data acquisition and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data;

[0013] A processing module, configured to perform a residual vector quantization operation on the CSI spatiotemporal feature data to generate a quantized data stream, and perform recursive encoding processing on the quantized data stream to obtain a target data stream;

[0014] A computing module, used for mapping the target data stream into a three-dimensional tensor space, constructing a real-valued tensor model, and applying unitary parallel factor decomposition and regularized least squares iterative operations to the real-valued tensor model to obtain a CSI feature vector;

[0015] A decomposition module, used for performing fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and inputting the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result;

[0016] A generation module is used to calculate the amplitude and phase weight coefficients of the antenna array according to the identity authentication result and the target data stream, generate beam parameters, and perform compensation matrix operations on the beam parameters according to the CSI eigenvector to obtain a precoded signal.

[0017] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0019] In summary, the technical solution provided by the present invention introduces spatiotemporal focus modulation technology, combined with CSI data acquisition and feature fusion mechanism, so that the system can accurately capture channel characteristics in a high dynamic environment, thereby improving the acquisition accuracy of channel state information; adopts residual vector quantization and recursive coding processing methods to achieve efficient compression and transmission of CSI data, reducing the system calculation complexity and storage overhead; innovatively combines three-dimensional tensor space mapping with unitary parallel factor decomposition to enhance the system's ability to extract and express multidimensional features and improve the accuracy of feature extraction; through the coordinated application of fractional wavelet transform decomposition and federated fractional scattering network, efficient feature processing and identity authentication are achieved while protecting data privacy; based on the dual constraint mechanism of identity authentication results and CSI feature vectors, combined with compensation matrix operations, the accuracy and anti-interference ability of beamforming are improved; the deep learning architecture of multi-layer convolution processing and feature fusion modules is adopted to improve the system's feature learning and classification capabilities and enhance the reliability of identity authentication; by introducing adaptive calculation and compensation optimization mechanism, dynamic adjustment of antenna beam parameters is achieved, thereby improving the system's adaptability in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the steps of a communication method of an integrated airborne communication antenna device in one embodiment of the present invention;

[0021] Figure 2 It is a structural block diagram of an integrated airborne communication antenna device in one embodiment of the present invention;

[0022] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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.

[0025] Reference Figure 1 This embodiment provides a communication method for an integrated airborne communication antenna device, comprising the following steps:

[0026] S1, perform CSI data collection and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data;

[0027] Among them, the communication channel is sampled in time sequence to obtain the CSI original sequence. The received signal is sampled at the signal receiving end to accurately restore the time-varying characteristics of the channel. The CSI original sequence obtained by sampling contains the amplitude and phase information of the channel. The CSI original sequence is subjected to amplitude extraction and phase extraction to separate the amplitude part and phase part of the signal from the original sequence. The amplitude reflects the strength of the signal, while the phase contains the relative information of the signal propagation path. Through these two extraction steps, the amplitude data and phase data of the CSI are obtained respectively. The amplitude data and the phase data are feature fused to obtain a comprehensive CSI feature sequence. The CSI feature sequence is subjected to time domain feature decomposition. The change characteristics of the signal in the time dimension are extracted through time domain analysis. The time domain characteristics of the signal are modeled through time-frequency analysis methods or signal decomposition algorithms to obtain time domain data. At the same time, spatial domain feature decomposition analyzes the spatial propagation characteristics of the signal from a spatial perspective. In the environment of a multi-antenna array, the signal correlation between different antennas is obtained through spatial domain decomposition to obtain spatial domain data. Focus modulation operations are performed on the time domain data and the spatial domain data. Focus modulation is a signal processing technology that adjusts the time domain and spatial domain characteristics of the signal to focus the signal in space and improve the quality and stability of the signal. By calculating the modulation coefficient, the time domain data and the spatial domain data are weighted and combined to obtain the eigenvector. The calculation of the modulation coefficient is based on the time and space characteristics of the signal. By optimizing these coefficients, the signal strength and propagation path are enhanced, so that the signal can achieve the best transmission effect in a specific spatial area. Perform an orthogonal transformation on the eigenvector. Through a series of linear transformations, the original eigenvector is converted into a set of orthogonal basis vectors. The orthogonal basis vectors have good mathematical properties, making the representation of the signal more concise and without losing information. The orthogonal basis vectors are applied to the CSI feature sequence to complete the projection of the original CSI feature sequence, and the original features are mapped to a new coordinate system to better highlight the main features of the signal and remove unnecessary noise. The projection matrix is ​​feature reconstructed. Through the reconstruction algorithm, a data sequence close to the original features is recovered from the projection matrix. The signal is restored from a compressed or decomposed state to a form as close to the original as possible through a specific mathematical method. The reconstructed sequence is nonlinearly mapped, and the signal is transformed by a nonlinear function to mine the potential information in the signal, help reveal the nonlinear relationship in the signal, improve the expressiveness of the feature, and obtain the mapping data. The mapping data is subjected to singular value decomposition and dimensionality transformation. Singular value decomposition is a matrix decomposition method that extracts the main components of the signal by decomposing the signal matrix into the product of multiple matrices. By performing singular value decomposition on the mapping data, the dimension of the data is reduced and redundant information is removed. Dimension transformation further reduces or increases the dimension of the data to meet the subsequent feature processing requirements, and finally obtains CSI spatiotemporal feature data.

[0028] S2, performing residual vector quantization operation on the CSI spatiotemporal feature data to generate a quantized data stream, and performing recursive encoding processing on the quantized data stream to obtain a target data stream;

[0029] Specifically, the CSI spatiotemporal feature data is divided into blocks, and the large-scale signal data is divided into multiple smaller feature data groups, which is convenient for subsequent quantization and encoding operations and reduces the complexity of data processing. Each feature data group represents the state information of a specific area or a certain moment of the channel and can be quantized and encoded independently. The block-based data is normalized by scaling the values ​​of each data group to standardize the range of the data, which is convenient for subsequent operations and obtains a normalized vector. A codebook is generated for the normalized vector. The codebook is a reference set used to map the input vector to the output vector during the quantization process and consists of multiple possible output vectors. By generating an initial codebook, a preliminary reference set is provided for the quantization process. The generated initial codebook is adjusted according to the actual channel environment to meet the needs of subsequent quantization. At the same time, the quantization parameter threshold is set according to the initial codebook to obtain a quantization control matrix. The function of the quantization parameter threshold is to set a critical value to determine which feature data should be quantized into which codeword, thereby helping the quantization module to make a choice during processing. The quantization control matrix is ​​a matrix used to indicate different parameter settings during the quantization process and control the behavior of the entire quantization process. The normalized vector is input to the first-level quantization module for vector quantization. The first-level quantization module maps the normalized vector to the closest codeword according to the generated codebook and outputs the first-level quantized data. The first-level quantized data and the normalized vector are subjected to difference operation to calculate the target residual data. The target residual data represents the quantization error and contains the part of information loss. The target residual data is input to the second-level quantization module for quantization to obtain the second-level quantized data. The second-level quantization module uses vector quantization to map the residual data to a codeword in the codebook. Through the two-level quantization method, the data error is reduced and the quantization accuracy and efficiency are improved. The first-level quantized data and the second-level quantized data are cascaded to obtain a quantized data stream. The quantized data stream is hierarchically encoded. The quantized data is divided into multiple levels for encoding, and each level uses a different encoding method to give priority to encoding different important data. The hierarchical encoding method can improve the robustness of the data and ensure that important data is transmitted first when the channel quality is poor or the transmission conditions are not ideal. The encoded data sequence is processed by data interleaving to obtain an interleaving matrix. By rearranging the order of data, the data is distributed more evenly during transmission, avoiding long-term continuous error propagation and improving the decoding success rate. The interleaving matrix is ​​recursively processed according to the quantization control matrix. The data is gradually optimized by repeating the processing steps many times. The recursive algorithm optimizes the representation and transmission effect of the signal by continuously adjusting the coding parameters and quantization thresholds. The recursive process uses historical information to correct the current data to achieve better transmission effects and obtain encoded data. The encoded data is dynamically updated in the codebook.During the data transmission process, the codebook is adjusted in real time according to the channel changes and feedback information to adapt to the new channel state. The coded data is reconstructed according to the updated codebook to obtain a reconstructed sequence. The quantized data is restored to a state as close to the original data as possible through the inverse operation to ensure the quality of the signal. After the reconstruction is completed, the reconstructed sequence is checked and corrected. By detecting and correcting the errors that occur during the data transmission process, the correctness and integrity of the final data are ensured to obtain the target data stream.

[0030] S3, maps the target data stream to a three-dimensional tensor space, constructs a real-valued tensor model, and applies unitary parallel factorization and regularized least squares iterative operations to the real-valued tensor model to obtain the CSI feature vector;

[0031] It should be noted that the target data stream is processed in the time dimension to obtain a time series. The time series captures the characteristics of the signal changing over time and reflects the dynamic change process of the channel. The time series is mapped in the spatial dimension. By spatially mapping the data in the time series, it is converted into a spatial matrix so that the data can be effectively represented in space. The spatial matrix is ​​sliced ​​and divided into multiple smaller fragments according to a certain dimension. Each slice represents a subset of the spatial matrix at different time or spatial positions. The slice set is reconstructed in three dimensions to fully represent the relationship between time, space and data to obtain a three-dimensional tensor. The three-dimensional tensor is aligned and normalized. Data alignment is to ensure that the data in the three-dimensional tensor can be correctly aligned and matched so that further processing will not be affected by misalignment or errors. Normalization is to eliminate the scale differences of data of different dimensions and ensure that the data in the tensor is within a uniform magnitude range, making subsequent processing more stable and efficient. After alignment and normalization, a standardized tensor is obtained. Based on the standardized tensor, factorization is performed to decompose a multidimensional data tensor into multiple factor matrices of lower dimensions, extract the potential features in the high-dimensional data, and represent the data with fewer parameters to improve the computational efficiency and obtain a real-valued tensor model. The real-valued tensor model is subjected to forward and backward operations. Forward operation refers to the prediction and calculation of data by the model under the condition of known input to obtain the forward tensor. The forward tensor contains the prediction results of the model on the input data and is used to preliminarily evaluate the performance of the model. Backward operation refers to the adjustment of model parameters by the back propagation algorithm under the condition of given target output to obtain the backward tensor. The backward tensor reflects the error between the model output and the target output. By analyzing the error, the model can adjust itself to obtain more accurate results in subsequent operations. Unitary decomposition is performed on the forward tensor and the backward tensor. Unitary decomposition is a mathematical operation that decomposes the tensor into a set of unitary matrices. Unitary matrices have good mathematical properties and can provide more stable solutions in the optimization process. Through unitary decomposition, a unitary characteristic matrix is ​​obtained to describe the intrinsic relationship between the dimensions in the tensor. Parallel factorization based on unitary characteristic matrix. Parallel factorization is a technique that decomposes a tensor into multiple factor matrices through parallel computing. It can process large-scale data in a shorter time and improve computational efficiency. Through parallel factorization, multiple decomposition parameters are obtained, representing the main features of different dimensions in the tensor. The decomposition parameters are input into the iterative optimizer for optimization. The iterative optimizer adjusts the decomposition parameters through multiple iterations to gradually improve the accuracy of the model. In each iteration, the optimizer updates the decomposition parameters according to the current error and parameter adjustment direction until the optimal solution is reached. The purpose of iterative optimization is to continuously correct the decomposition parameters so that the model can obtain the best fitting effect in the shortest possible time.During the iteration process, the iterative sequence is tested for convergence to evaluate whether the optimization process has reached a stable state. If the optimization process has converged, it means that the parameters of the model no longer change significantly, and it is considered that an optimal solution has been found. The convergence index is evaluated by calculating the change in the error or the change in the objective function. If these changes are less than the set threshold, it means that the optimization process has stabilized. According to the results of the convergence index, the decomposition parameters are optimized and calculated to obtain the final CSI feature vector.

[0032] S4, performing fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and inputting the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result;

[0033] Specifically, the CSI feature vector is subjected to fractional wavelet transform. As a multi-scale signal processing method, fractional wavelet transform analyzes data at different frequency and time scales. Through fractional wavelet transform, the multi-scale local features of the CSI feature vector are extracted to obtain a series of transformation sequences. The transformation sequence contains the spatiotemporal information of the channel and reveals the change pattern of the signal at different scales. The transformation sequence is subjected to nonlinear transformation to obtain nonlinear features. Nonlinear features can reflect the behavior of the signal at different levels and in more complex modes, which is helpful for signal recognition and classification. The nonlinear transformation process is implemented through mathematical functions (such as activation functions, nonlinear filtering, etc.), so that the system can capture the deeper structure of the signal and improve the accuracy of recognition and processing. Wavelet scattering decomposition is performed on the nonlinear features. Wavelet scattering transform is a technology that extracts high-order features of signals through multi-layer wavelet transform and nonlinear processing, which can effectively capture important information in the signal. The nonlinear features are decomposed into multiple components to form a scattering matrix, which can represent the characteristics of the signal at different scales, frequencies and levels. The scattering matrix is ​​processed in groups. Effectively aggregate the features in the scattering matrix to make the feature set more compact and well distributed. By grouping, the different parts of the signal features are classified into corresponding groups to form a set of RF features. The RF feature set is assigned to the federated learning node for distributed computing. Federated learning is a distributed machine learning method that allows multiple computing nodes to train models locally without sharing data, thereby protecting data privacy. By assigning the RF feature set to multiple federated learning nodes, parallel processing is performed at different locations to improve computing efficiency and reduce the bandwidth requirements for data transmission. During the local computing process, each federated node processes the RF features assigned to it and generates local features, which represent the key information extracted by different nodes under local signal processing. Encrypt the local features. Use symmetric encryption, asymmetric encryption, or homomorphic encryption to ensure that the results of local calculations are not tampered with or leaked during transmission. Transmit the encrypted local features to the central processing node. At the central node, the encrypted data from each federated learning node is aggregated, and the global features are obtained by merging and aggregating the local features of each node. The global features are decrypted and the encrypted data is restored to plaintext using the key to ensure that the data is always kept confidential during transmission. The decrypted data is processed and analyzed to form a decrypted data set. The decrypted data is input into the federated fractional scattering network for feature aggregation to form a complete feature representation containing global information to obtain the identity authentication result.

[0034] Symmetric decryption is performed on the global features. Symmetric encryption is a technology that uses the same key for encryption and decryption, which can effectively protect information security during data transmission. By symmetric decryption of the global features, the original plaintext data, i.e., the decrypted data, is restored. After the decrypted data is restored, the data will be integrity checked to generate a check matrix, which contains information about the integrity of the data for subsequent error detection and correction. The check matrix can help the system determine whether the data has encountered any unexpected changes during transmission, thereby improving the accuracy and security of the authentication process. The check matrix is ​​input into the feature extraction module in the federated fractional scattering network for processing. The feature extraction module uses three layers of convolution processing units, each of which includes a convolution operation layer, a batch normalization layer, and an activation function layer. The function of the convolution layer is to extract features from the input data through convolution operations, and the batch normalization layer is used to standardize the results after the convolution operation to ensure that the output of each layer meets certain scale requirements, thereby accelerating the training process and improving the stability of the model. The activation function layer enables the network to learn more complex features by introducing nonlinear transformations. In this process, the feature map generated by the convolution calculation contains the spatial and temporal features of the input data. The feature map is processed by dimensionality reduction. By reducing the dimension of the data, high-dimensional data is mapped to a low-dimensional space to remove redundant information and improve computational efficiency. After dimensionality reduction, a reduced-dimensional matrix is ​​obtained. The reduced-dimensional matrix is ​​input into the feature fusion module of the federated fractional scattering network. The feature fusion module combines features using the attention mechanism and residual structure. The attention mechanism can dynamically adjust the weights of different features according to the importance of the data, so that the model can focus on the most relevant features and improve classification performance. The residual structure introduces jump connections to make information easier to propagate in the network, avoiding the gradient vanishing or exploding problems encountered in deep network training. Through the attention mechanism and residual structure, the feature fusion module can effectively combine and optimize features from different sources to obtain a fusion vector. The fusion vector is input into the classification module, which contains four layers of fully connected units. Each layer of fully connected units performs weighted processing on the input and transforms the information from the previous layer into a new representation. Regularization and normalization are used in each layer to ensure the stability of model parameters during training and prevent overfitting. Regularization ensures that the model does not rely too much on the training data by imposing penalties on the parameters, while normalization ensures that the data has a uniform scale across layers. Through the forward propagation of the multi-layer fully connected network, a classification vector is obtained, which contains classification information about identity authentication and reflects the possibility that the input data belongs to a certain identity category. The classification vector is threshold filtered to remove some unimportant classification results and only retain the most representative parts. After threshold filtering, the remaining data is input into the multi-layer perception classification module. Multi-layer perception classification is a technology that further improves classification accuracy through neural networks. It can analyze and classify data more carefully through complex hierarchical structures to obtain the first authentication value.The confidence of the first authentication value is calculated to evaluate the reliability and accuracy of the first authentication value and generate a confidence vector. The confidence vector is then used to generate a confusion matrix. The confusion matrix is ​​a tool for evaluating classification performance, which is used to compare the relationship between the true category and the predicted category. The confusion matrix is ​​used to evaluate the accuracy of identity authentication, so as to make final adjustments to the authentication result. The first authentication value and the second authentication value are input into the decision module, and the decision module obtains a fusion value through weighted calculation. Adjustments are made according to the importance of different authentication results to ensure that the final authentication decision is more accurate. The fusion value is judged according to the preset threshold. When the fusion value is greater than the preset threshold, the identity authentication result is output to confirm the identity of the communication device.

[0035] S5, calculate the amplitude and phase weight coefficients of the antenna array according to the identity authentication result and the target data stream, generate beam parameters, and perform compensation matrix operation on the beam parameters according to the CSI eigenvector to obtain a precoded signal.

[0036] Among them, the identity authentication result and the target data stream are feature combined. The identity authentication result represents the communication system's confirmation of the device identity, while the target data stream contains specific communication information. Through feature combination, the data from these two different sources are integrated into a new data set to obtain combined data. In the process of feature combination, the data is adaptively calculated to generate a weight vector. The weight distribution is dynamically adjusted according to the characteristics of the data, so that the weight vector can most effectively reflect the combination of the target data stream and the identity authentication information. The antenna array response analysis is performed on the weight vector to determine the signal reception and transmission capabilities of each antenna unit in different directions, helping the system understand how the array responds to different signal sources. By analyzing the relationship between the weight vector and the array response, a response matrix is ​​obtained, which represents the response characteristics of different antenna units in various directions. In order to optimize the performance of the array, the response matrix is ​​orthogonally processed to obtain an orthogonal vector. By converting the original matrix into an orthogonal matrix, it is ensured that each signal path is independent of each other, and the efficiency and anti-interference ability of signal transmission are improved. The orthogonal vector is amplitude decomposed. The strength information of the signal is separated from its phase information so that the amplitude part of the signal can be processed separately. Through amplitude decomposition, amplitude data is obtained, which represents the strength information of the signal in different directions. According to the amplitude data, a weight matrix is ​​constructed to obtain the amplitude and phase weight coefficients of the antenna array. The amplitude and phase weight coefficients are the amplitude and phase weights of each antenna unit in the antenna array during signal transmission and reception. These weight coefficients directly determine how the array forms a beam, thereby affecting the transmission quality and coverage of the signal. The amplitude and phase weight coefficients of the antenna array are input into the beamforming module to generate initial parameters. Beamforming is the process of adjusting the signal phase and amplitude of each unit in the antenna array so that the antenna array can enhance the signal strength in a specific direction. At this stage, the initial parameters represent the beam direction and amplitude information generated based on the current weight coefficients. In order to optimize the directivity of the beam, the initial parameters are spatially scanned to explore different beam directions to find the optimal signal coverage direction. By scanning in different directions, the most appropriate beam parameters are determined to ensure that the signal can cover the required area and effectively avoid interference. After the beam parameters are determined, compensation analysis is performed. The beam parameters are adjusted according to the CSI eigenvector to compensate for problems such as signal attenuation and noise interference that occur during transmission. The CSI feature vector provides the spatiotemporal information of the channel, which can help the system analyze the state of the channel and generate a compensation sequence based on this information. The compensation sequence reflects the adjustment of the beam parameters that the system needs to make under the influence of the channel. Based on the compensation sequence, the compensation coefficient is calculated to determine the amount of correction required during the transmission of the signal. The beam parameters are compensated and calculated. Based on the compensation coefficient, the beam parameters are adjusted to improve the transmission quality of the signal.During the compensation calculation process, the amplitude, phase and other parameters of the beam are dynamically adjusted to adapt to the changes in the channel and optimize the propagation effect of the signal. Through this process, the system can effectively reduce the attenuation and interference of the signal and improve the transmission quality and reliability of the signal. After completing the compensation calculation, the compensation parameters are optimized to improve the accuracy and efficiency of beamforming so that the signal can be transmitted to the maximum in the target direction. By optimizing the compensation parameters, the optimal transmission scheme is found under the channel conditions to ensure the efficient propagation of the signal. The target parameters are encoded. The optimized beam parameters are converted into a data format suitable for transmission so that they can be transmitted through the communication link. Through encoding, the effective transmission of the beam parameters is ensured and they can be correctly decoded at the receiving end. Based on the coding sequence, precoding calculation is performed to obtain the precoded signal. The signal transmission mode is adjusted according to the encoded beam parameters to optimize the reception quality of the signal at the receiving end. Through precoding calculation, the signal is appropriately modulated and optimized during the transmission process to ensure the efficient propagation of the signal in the channel.

[0037] In one example, CSI data collection and spatiotemporal focus modulation are performed on a communication channel to obtain CSI spatiotemporal feature data, including:

[0038] Perform time-series sampling on the communication channel to obtain the CSI original sequence;

[0039] Perform amplitude extraction and phase extraction on the CSI original sequence to obtain CSI amplitude data and CSI phase data, and perform feature fusion on the CSI amplitude data and CSI phase data to obtain a CSI feature sequence;

[0040] Perform time domain feature decomposition on the CSI feature sequence to obtain time domain data, and perform space domain feature decomposition on the CSI feature sequence to obtain space domain data;

[0041] Performing focal modulation operation on the time domain data and the spatial domain data to obtain a modulation coefficient, and performing weighted combination on the time domain data and the spatial domain data according to the modulation coefficient to obtain a feature vector;

[0042] Perform an orthogonal transformation on the feature vector to obtain the orthogonal basis vector, and project the CSI feature sequence onto the orthogonal basis vector to obtain the projection matrix;

[0043] The feature is reconstructed according to the projection matrix to obtain a reconstructed sequence, the reconstructed sequence is nonlinearly mapped to obtain mapping data, and the mapping data is subjected to singular value decomposition and dimension transformation to obtain CSI spatiotemporal feature data.

[0044] In this example, the communication channel is sampled sequentially to obtain the original sequence of CSI (Channel State Information). In wireless communication systems, CSI is important information that describes channel characteristics, including parameters such as amplitude and phase. Through sequential sampling, this information is extracted from the channel. Assume that the system is at time point Collect channel data, expressed as:

[0045] ;

[0046] in is the amplitude of the channel, is the phase of the channel, Is an imaginary unit. Through sampling, the original data sequence of CSI is obtained , which is the original CSI sequence. Amplitude extraction and phase extraction are performed on the original CSI sequence. Amplitude extraction is the process of extracting the channel amplitude part. Modulo, the formula is:

[0047] ;

[0048] in, and Represent the real and imaginary parts of the complex channel information respectively. Through this process, the CSI amplitude data is obtained Phase extraction is to extract the phase part of the channel, using the polar coordinate representation of complex numbers, the formula is:

[0049] ;

[0050] Through this formula, the phase data is obtained . Perform feature fusion on the amplitude data and phase data of CSI to form a comprehensive feature sequence. Assume is the amplitude data, is the phase data, combining the two into a characteristic sequence :

[0051] ;

[0052] in, and is the fusion coefficient, which is used to control the contribution of amplitude and phase to the final feature. In this way, the CSI feature sequence is obtained , which contains both amplitude and phase information. The CSI feature sequence is decomposed in the time domain and in the space domain. Time domain feature decomposition is to analyze the feature sequence in the time domain and extract the time domain features through Fourier transform or wavelet transform. For example, the feature sequence is converted to the frequency domain through Fourier transform:

[0053] ;

[0054] in, is the time domain feature sequence, is the frequency domain feature sequence, is the frequency. Time domain data is extracted from frequency domain features to help analyze the time variation of the signal. Spatial domain feature decomposition is to analyze the spatial structure of the CSI feature sequence through principal component analysis or singular value decomposition. Assuming there is an array with multiple antennas, spatial domain feature decomposition is performed using the following formula:

[0055] ;

[0056] in, and are the left and right singular vectors, respectively, is a diagonal matrix containing singular values. The spatial domain data is extracted from the original feature sequence through singular value decomposition. . Perform focus modulation operations on the time domain data and the spatial domain data. Adjust the time domain and spatial domain distribution of the signal in some way to concentrate the energy or improve the directivity of the signal. Assume that the modulation coefficient For time domain data and airspace data Perform a weighted combination:

[0057] ;

[0058] in, and are modulation coefficients, and these coefficients are used to adjust the influence weights of the two data in the final combination. Perform an orthogonal transformation on the eigenvector to eliminate the correlation between the data, making the features more independent, and use an orthogonal matrix for transformation. Assume that the eigenvector is , the vector after orthogonal transformation Calculated by the following formula:

[0059] ;

[0060] in, is an orthogonal matrix that satisfies , that is, the product of the transpose of the matrix and the matrix itself is the unit matrix. After orthogonal transformation, the orthogonal basis vectors are obtained, and the original CSI feature sequence is projected onto these orthogonal basis vectors to obtain the projection matrix. According to the projection matrix, feature reconstruction is performed. The main information of the signal is recovered from the orthogonal basis vectors. The inverse transformation is used for reconstruction, and the formula is:

[0061] ;

[0062] Reconstructed feature sequence It is an approximate representation of the original CSI data, retaining its important features. The reconstructed sequence is nonlinearly mapped and mapped through nonlinear activation functions (such as ReLU, Sigmoid, etc.) so that the data can better adapt to subsequent processing steps. The mapped data is subjected to singular value decomposition and dimensionality transformation. Singular value decomposition is used to decompose the mapped data into components of multiple dimensions and extract the main components of the signal. Through dimensionality transformation, the redundancy of the data is reduced and important signal information is retained. The CSI spatiotemporal feature data is obtained.

[0063] In one example, a residual vector quantization operation is performed on the CSI spatiotemporal feature data to generate a quantized data stream, and the quantized data stream is recursively encoded to obtain a target data stream, including:

[0064] The CSI spatiotemporal feature data is divided into blocks to obtain a feature data group, and a normalization operation is performed on the feature data group to obtain a normalized vector;

[0065] Generate a codebook for the normalized vector to obtain an initial codebook, and set a quantization parameter threshold according to the initial codebook to obtain a quantization control matrix;

[0066] Input the normalized vector into the first-level quantization module for vector quantization to obtain first-level quantized data; perform difference operation on the first-level quantized data and the normalized vector to obtain target residual data;

[0067] Input the target residual data into the second-level quantization module for vector quantization to obtain second-level quantized data, and cascade the first-level quantized data and the second-level quantized data to obtain a quantized data stream;

[0068] Performing hierarchical coding on the quantized data stream to obtain a coding sequence, and performing data interleaving on the coding sequence to obtain an interleaving matrix;

[0069] Recursively processing the interleaving matrix according to the quantization control matrix to obtain coded data, and dynamically updating the coded data to obtain an updated codebook;

[0070] The coded data is reconstructed according to the updated codebook to obtain a reconstructed sequence, and the reconstructed sequence is verified and error-corrected to obtain a target data stream.

[0071] In this example, the CSI spatiotemporal feature data is divided into blocks, and the original spatiotemporal feature data Divide into several sizes sub-blocks so that each sub-block can be processed independently. Assume that the original data is ,in Indicates feature vectors, through data block, the feature data group B is expressed as:

[0072] ;

[0073] in, It is blocks, including In this process, the data is divided according to the time series or spatial structure. The feature data group is normalized to unify the scale of different blocks of data for subsequent quantization. The normalization process is performed using the following formula:

[0074] ;

[0075] in, represents the normalized eigenvalue, is the original data, is the mean of the feature, is the standard deviation. By normalization, the feature data set Each vector in becomes a standard normal distribution with zero mean and unit variance, which helps to improve the accuracy of subsequent quantization and encoding processes. After normalization, the initial codebook is generated, and the quantization parameter threshold is set according to the initial codebook. The initial codebook is a set of multiple possible feature vectors, generated by a clustering algorithm (such as K-means). Assume that the codebook C contains codewords, the codebook is represented as:

[0076] ;

[0077] Each codeword Corresponding to a possible eigenvector. In the quantization process, the feature data is approximated according to these codewords. The quantization control matrix is ​​used to represent the control parameters in the quantization process, such as quantization accuracy or threshold, and is generated through statistical analysis of the initial codebook. The specific form is:

[0078] ;

[0079] in, Representation and Codewords Related quantization control parameters. The normalized vector is input into the first-level quantization module for vector quantization to obtain the first-level quantization data. The feature vector is mapped to the discrete codeword space by minimizing the quantization error. The first-level quantization process is expressed by the following formula:

[0080] ;

[0081] in, represents the feature vector after the first level quantization, It is Normalized feature vectors, is a codeword in the codebook, Represents the Euclidean distance of the vector. Through this operation, the first-level quantized data is obtained The first-level quantization data and the normalized vector are interpolated to obtain the target residual data. The target residual data refers to the difference between the first-level quantization error and the original data. The specific calculation formula is:

[0082] ;

[0083] in, Indicates The residual data is input into the second-level quantization module for second-level quantization to obtain second-level quantized data. The second-level quantization process is similar to the first-level process. By selecting the optimal codeword To minimize the quantization error:

[0084] ;

[0085] Through this step, the second level of quantitative data is obtained , and then cascade the first-level quantized data and the second-level quantized data to obtain the quantized data stream:

[0086] ;

[0087] in, Represents the concatenation of two data sequences. Perform hierarchical coding on the quantized data stream to obtain a coding sequence. Encode according to different levels of quantized data, and use hierarchical coding technology to encode data at different levels separately. Suppose the coding sequence is , the encoding process is expressed as:

[0088] ;

[0089] The coding sequence is interleaved to obtain an interleaving matrix. The anti-interference ability of the coding is improved by changing the order of data. The interleaving process is implemented through permutation operations, through a permutation matrix To represent the interleaving matrix:

[0090] ;

[0091] Based on the quantization control matrix, the interleaving matrix is ​​recursively processed to obtain the encoded data. The recursive processing is achieved by cyclically updating and optimizing the data. The specific operation is represented by the recursive algorithm:

[0092] ;

[0093] Dynamically update the codebook for the coded data to adapt to different channel conditions. The updated codebook Adjustment is done through online learning or feedback mechanism. According to the updated codebook, the coded data is reconstructed to obtain a reconstructed sequence. The reconstruction process is achieved through inverse encoding or inverse quantization. The reconstructed sequence is checked and error-corrected to ensure the correctness and integrity of the data and obtain the target data stream.

[0094] In one example, the target data stream is mapped to a three-dimensional tensor space, a real-valued tensor model is constructed, and unitary parallel factorization and regularized least squares iterative operations are applied to the real-valued tensor model to obtain a CSI feature vector, including:

[0095] The target data stream is processed in the time dimension to obtain a time series, and the time series is mapped in the space dimension to obtain a space matrix;

[0096] The spatial matrix is ​​sliced ​​into blocks to obtain a slice set, and the slice set is reconstructed in three dimensions to obtain a three-dimensional tensor;

[0097] Perform data alignment and normalization on the three-dimensional tensor to obtain a standardized tensor, and then perform factorization based on the standardized tensor to construct a real-valued tensor model;

[0098] Perform a forward operation on the real-valued tensor model to obtain a forward tensor, and perform a backward operation on the real-valued tensor model to obtain a backward tensor;

[0099] Perform unitary decomposition on the forward tensor and the backward tensor to obtain a unitary characteristic matrix, and perform parallel factor decomposition based on the unitary characteristic matrix to obtain decomposition parameters;

[0100] Input the decomposition parameters into the iterative optimizer to obtain an iterative sequence, and perform a convergence test on the iterative sequence to obtain a convergence index;

[0101] The decomposition parameters are optimized and calculated according to the convergence index to obtain the CSI feature vector.

[0102] In this example, the target data stream is processed in the time dimension to obtain a time series. The target data stream is high-dimensional data, containing measurement data at multiple moments. These data are arranged in chronological order to form a time series, which is recorded as ,in Indicates The time dimension is regarded as a one-dimensional sequence, and the time series is extracted through sliding windows or sampling techniques. Time series processing includes operations such as denoising and filtering to reduce the noise impact of the data. The time series is mapped to the spatial dimension to obtain a spatial matrix. The one-dimensional time series is converted into a two-dimensional or higher-dimensional matrix to capture the spatial structure contained in the time series data. The time series is mapped to the spatial matrix through some transformation rules or models (such as Fourier transform, wavelet transform or linear transform), which is recorded as ,in represents the number of rows of the spatial dimension, Represents the number of columns of the spatial dimension. The spatial matrix contains the spatial features corresponding to each moment. Perform a block slice operation on the spatial matrix to obtain a slice set. Divide the spatial matrix into several small blocks for subsequent processing. Assume that the spatial matrix for The matrix is ​​divided into several matrices of size A small matrix of . Slice collection That is, a collection of these small matrices:

[0103] ;

[0104] in is the number of slices, Indicates Slices. Split the large matrix into several small blocks, each of which corresponds to a local area of ​​the data. Perform three-dimensional reconstruction on the set of slices. Reassemble the set of slices into a three-dimensional tensor. Assume that slices, the reconstructed 3D tensor It is expressed as:

[0105] ;

[0106] Among them, the reconstructed tensor It is a three-dimensional matrix that represents the characteristics of the data in three dimensions. The three-dimensional tensor is aligned and normalized to obtain a standardized tensor. The purpose of data alignment is to ensure that the data of each dimension is compared at the same scale, while the normalization operation is to eliminate the differences between different data dimensions so that the data is numerically consistent. The normalization formula is:

[0107] ;

[0108] in is the normalized tensor value, is a tensor exist The mean of the positions, is the standard deviation of the corresponding position. Through normalization, the numerical distribution of tensor data is ensured to have zero mean and unit variance, making subsequent processing more stable. Based on the standardized tensor, factorization is performed to construct a real-valued tensor model. Tensor factorization is a technique that decomposes a high-dimensional tensor into multiple low-dimensional tensors to extract the potential structure. Tensor factorization methods include CP decomposition and Tucker decomposition. Choose a suitable factorization method based on the standardized tensor Factorize and get a low-rank tensor model, expressed as:

[0109] ;

[0110] in are three factor matrices, is the factorized rank, symbol Represents the outer product of tensors. The factorized real-valued tensor model effectively represents the potential structure of the original data. After the factorization process is completed, the real-valued tensor model is subjected to forward and backward operations. The forward operation refers to reasoning through the model to obtain the prediction result; while the backward operation is to update the model parameters through optimization methods such as backpropagation or gradient descent. The forward operation is expressed by the following formula:

[0111] ;

[0112] in is the forward function of the model, are the parameters of the model, is the result of the forward operation. The backward operation minimizes the loss function by updating the parameters and uses the gradient descent algorithm to optimize the parameters:

[0113] ;

[0114] in is the loss function, is the gradient of the loss function, is the learning rate, is the updated parameter. Perform unitary decomposition on the forward tensor and the backward tensor. Unitary decomposition is a technique for extracting data features through orthogonal transformation, which is used in signal processing and data dimensionality reduction. Through unitary decomposition, the unitary feature matrix is ​​obtained , which contains the orthogonal features of the data. Based on the unitary decomposition, parallel factor decomposition is performed to obtain the decomposition parameters, which are expressed as:

[0115] ;

[0116] The decomposition parameters are input into the iterative optimizer for further optimization. The iterative optimizer optimizes the decomposition parameters through multiple iterations until convergence. Each iteration adjusts the parameters according to some optimization criterion (such as minimizing the loss function or maximizing the likelihood). The parameter update after each round of iteration is expressed by the following formula:

[0117] ;

[0118] The convergence of the iterative sequence is tested to determine whether it has converged, and the final optimization calculation is performed based on the convergence index. The convergence index is the rate of change of the loss function or the prediction accuracy. According to the convergence index, the decomposition parameter finally obtained is the required CSI feature vector.

[0119] In one example, a CSI feature vector is decomposed by performing a fractional wavelet transform to obtain a radio frequency feature set, and the radio frequency feature set is input into a federated fractional scattering network for feature processing to obtain an identity authentication result, including:

[0120] Performing fractional wavelet transform on CSI feature vector to obtain transformation sequence, and performing nonlinear transformation on the transformation sequence to obtain nonlinear features;

[0121] Perform wavelet scattering decomposition on the nonlinear features to obtain a scattering matrix, and perform grouping processing on the scattering matrix to obtain a set of radio frequency features;

[0122] Assign the RF feature set to the federated learning node to obtain a distributed feature sequence, and perform local calculations on the distributed feature sequence to obtain local features;

[0123] Perform encryption operations on local features to obtain encrypted data, and transmit the encrypted data to the central processing node to obtain global features;

[0124] The global features are decrypted to obtain decrypted data, and the decrypted data is input into the federated fractional scattering network for feature aggregation to obtain the identity authentication result.

[0125] In this example, a fractional wavelet transform is performed on the CSI feature vector. Compared with the classical wavelet transform, the fractional wavelet transform uses a fractional scale factor, denoted as By adjusting this parameter, the detail level of wavelet transform can be controlled. The transformation process is expressed as:

[0126] ;

[0127] in, represents the result of fractional wavelet transform, is the fractional wavelet basis function, parameter Controls the detail of the transformation. Through this transformation, the resulting transformation sequence Contains the characteristic information of the signal at different scales. Perform nonlinear transformation on the transformation sequence. Nonlinear transformation includes ReLU function, Sigmoid function or hyperbolic tangent function, etc. The formula is as follows:

[0128] ;

[0129] in, represents the output after nonlinear transformation, It is an element in the transformation sequence. Through nonlinear transformation, the nonlinear features in the transformation sequence are extracted. These features have better expression ability for the pattern, volatility and irregularity of the signal. The result is recorded as nonlinear feature vector . Perform wavelet scattering decomposition on nonlinear features. Wavelet scattering transform is a multi-level signal processing method that captures low-order and high-order features in the signal by successively passing the signal through low-pass filtering, high-pass filtering and scattering operations. Wavelet scattering decomposition is a recursive application of wavelet transform, and its goal is to extract layer-by-layer features of the signal. Assume that the nonlinear feature vector is , the scattering matrix is ​​obtained by wavelet scattering decomposition , which contains multi-level characteristic information of the signal. Perform preliminary filtering and nonlinear activation, then process the results at multiple levels, and finally output Contains the low-frequency and high-frequency characteristics of the signal. Perform grouping processing to group the scattered features according to different dimensions or categories. Through the grouping operation, the RF feature set is obtained. , which contains multiple grouped feature matrices:

[0130] ;

[0131] in, It is The feature matrix after grouping is is the number of groups. Each group contains the characteristics of the RF signal at different scales, which can effectively represent the characteristics of the communication channel. Assign to federated learning nodes for distributed computing. Federated learning is a distributed machine learning method that allows multiple devices to perform calculations locally and exchange only model parameters without sharing original data. Each federated node processes the feature set based on its local data to obtain local features. Assume that federated learning nodes, each node Processing feature sets , get local features , and its calculation process is expressed as:

[0132] ;

[0133] in, is the local computation function of each node, and the output is the local feature calculated by the node. The local feature is encrypted to ensure the privacy of the data. The encryption process uses a symmetric encryption algorithm, such as the AES algorithm, and the formula is as follows:

[0134] ;

[0135] in, is the encrypted local feature, is the encryption key. Encrypted data It will be transmitted to the central processing node, which aggregates the local features of each node to obtain the global features. The global features are decrypted to obtain the decrypted data. The decryption operation uses the same key and algorithm as the encryption. The decryption process is as follows:

[0136] ;

[0137] Decrypt the data Input the federated fractional scattering network for feature aggregation. The federated fractional scattering network is a deep learning network that uses fractional wavelet transform for feature extraction and aggregation. It can integrate feature information from different nodes to obtain a global feature representation. The feature aggregation operation is performed through weighted averaging, maximum pooling or other methods to effectively fuse the features of each node. Feature representation after aggregation It is expressed as:

[0138] ;

[0139] in, is the weight coefficient, which is set according to the amount of data or computing power of each node. Based on the aggregated global features, the federated fractional scattering network performs the identity authentication task. The result of identity authentication According to global characteristics For classification, use traditional classification algorithms (such as SVM, decision tree, etc.) for prediction:

[0140] ;

[0141] Get the identity authentication result.

[0142] In one example, the global features are decrypted to obtain decrypted data, and the decrypted data is input into the federated fractional scattering network for feature aggregation to obtain an identity authentication result, including:

[0143] Symmetrically decrypt the global features to obtain decrypted data; perform integrity check on the decrypted data to obtain a check matrix;

[0144] The check matrix is ​​input into the feature extraction module of the federated fractional scattering network. The feature extraction module includes three layers of convolution processing units. Each layer includes a convolution operation layer, a batch normalization layer, and an activation function layer. The feature map is obtained through convolution calculation.

[0145] Perform dimensionality reduction processing on the feature map to obtain a reduced-dimensional matrix, and input the reduced-dimensional matrix into the feature fusion module of the federated fractional scattering network. The feature fusion module uses the attention mechanism and residual structure to combine features to obtain a fusion vector.

[0146] The fusion vector is input into the classification module of the federated fractional scattering network. The classification module contains four layers of fully connected units. Each layer of fully connected units uses regularization and normalization to obtain the classification vector through forward propagation.

[0147] Perform threshold filtering on the classification vector to obtain a filtering result, perform multi-layer perception classification on the filtering result to obtain a first authentication value, perform confidence calculation on the first authentication value to obtain a confidence vector, generate a confusion matrix according to the confidence vector, and obtain a second authentication value;

[0148] The first authentication value and the second authentication value are input into the decision module, a fusion value is obtained through weighted calculation, and a threshold judgment is performed on the fusion value. When the fusion value is greater than a preset threshold, the identity authentication result is output.

[0149] In this example, the global signature is symmetrically decrypted. Assume that the encrypted data is , decrypted using a symmetric key encryption algorithm (such as AES), the decryption operation is performed by the key Execute to get the decrypted data The decryption process is expressed as:

[0150] ;

[0151] in, Indicates the encrypted data. is a symmetric key, and is the decrypted data. By decrypting, the original global feature data is restored. The decrypted data is integrity checked. The decrypted data is hashed using a hash algorithm (such as SHA-256) to generate a check matrix. This matrix is ​​used to verify the integrity of the data. Assume that the decrypted data is , integrity check through hash function To execute, get the check matrix :

[0152] ;

[0153] in, is a hash function, is the decrypted data, is the check matrix. The hash algorithm ensures the data cannot be tampered with and verifies whether the data has changed during the transmission process. Input the feature extraction module of the federated fractional scattering network. The feature extraction module consists of multiple convolutional layers. The function of the convolutional layer is to extract important features from the input data through filters. The feature extraction module contains three layers of convolutional processing units, each of which includes a convolutional layer, a batch normalization layer, and an activation function layer. Through the convolution operation, the feature map of the data is obtained. The layer convolution operation is calculated by the following formula:

[0154] ;

[0155] in, For the The convolution output of the layer, is the input data (i.e., the check matrix), is the convolution kernel, is the bias. After the convolution operation, the data is normalized and processed nonlinearly by the batch normalization layer and the activation function. The feature map is reduced in dimension to reduce the computational complexity and extract key features. Through principal component analysis, the high-dimensional feature map is converted into a low-dimensional representation to obtain a reduced-dimensional matrix. The reduced-dimensional matrix is ​​input into the feature fusion module. The feature fusion module combines the attention mechanism and the residual structure for feature fusion. The attention mechanism focuses on more important features by learning weighted coefficients, while the residual structure helps information flow in the network to avoid information loss. The feature fusion process is carried out by weighting and summing, and the fused features are represented as fusion vectors. The fused vector Input classification module, which is composed of multiple fully connected units (i.e. neural network layers). Each layer is regularized and normalized, and a classification vector is obtained through forward propagation calculation, which represents the probability distribution of input data in different categories. Assume that the classification module contains The output classification vector of the fully connected unit is , the calculation process is expressed as:

[0156] ;

[0157] in, and are the weights and biases of the fully connected layer, Is the classification output. For the classification vector Perform threshold filtering to remove unimportant features. , only the values ​​in the classification vector that are greater than the threshold are retained, and the filtering result is obtained:

[0158] ;

[0159] in, is the result after threshold filtering. is the filtering threshold. The filtered results are classified by multi-layer perception to obtain the first authentication value. Through the multi-layer perceptron network, its output is the probability of the authentication value, indicating whether the user's identity is successfully authenticated. Assume that the first authentication value is , and calculate the confidence. Use the softmax function to calculate the confidence and get the confidence vector , the calculation formula is as follows:

[0160] ;

[0161] Based on the confidence vector, a confusion matrix is ​​generated to evaluate the credibility of the authentication. Through the confusion matrix, the second authentication value is obtained :

[0162] ;

[0163] The first authentication value and the second authentication value Input to the decision module, which obtains the fusion value through weighted calculation :

[0164] ;

[0165] in, and is the weight coefficient, indicating the importance of each authentication value. Determine whether the identity authentication is successful:

[0166] ;

[0167] when When it is greater than the threshold, the authentication result is passed (identity authentication is successful); otherwise, the authentication fails.

[0168] In one example, the amplitude and phase weight coefficients of the antenna array are calculated according to the identity authentication result and the target data stream, the beam parameters are generated, and the compensation matrix operation is performed on the beam parameters according to the CSI eigenvector to obtain the precoded signal, including:

[0169] Performing feature combination on the identity authentication result and the target data stream to obtain combined data, and performing adaptive calculation on the combined data to obtain a weight vector;

[0170] Perform antenna array response analysis on the weight vector to obtain a response matrix, and perform orthogonal processing on the response matrix to obtain an orthogonal vector;

[0171] Perform amplitude decomposition on the orthogonal vector to obtain amplitude data, and construct a weight matrix based on the amplitude data to obtain the amplitude and phase weight coefficients of the antenna array;

[0172] Inputting amplitude and phase weight coefficients into a beamforming module to obtain initial parameters, and performing spatial scanning on the initial parameters to obtain beam parameters;

[0173] Perform compensation analysis on the CSI feature vector to obtain a compensation sequence, and calculate a compensation coefficient based on the compensation sequence;

[0174] Compensation calculation is performed on the beam parameters to obtain compensation parameters, and the compensation parameters are optimized according to the compensation coefficients to obtain target parameters;

[0175] The target parameters are encoded to obtain a coding sequence, and a precoding calculation is performed based on the coding sequence to obtain a precoded signal.

[0176] In this example, the identity authentication result and the self-labeled data stream are combined to obtain the combined data. The combined data is formed by weighted averaging or concatenation. For example, two data vectors and Splice to get the combined data vector :

[0177] ;

[0178] in, is the feature vector of the authentication result, is the feature vector of the target data stream. Perform adaptive calculations. Adjust the weights through the learning algorithm to obtain a weight vector suitable for the current task. For example, using gradient descent to minimize a loss function:

[0179] ;

[0180] in, is the expected output, is the input data, is the weight vector. Perform antenna array response analysis to evaluate how a given signal is distributed among the different antenna elements. Assume that the antenna array consists of antenna elements, using the response matrix To represent the response of the antenna array to the signal, each element Indicates Antenna pair The response strength of each signal source:

[0181] ;

[0182] The response matrix is ​​obtained by modeling the physical characteristics of the antenna array. Perform orthogonal processing to obtain orthogonal vectors The orthogonalization is achieved using QR decomposition, where ,in is an orthogonal matrix, is an upper triangular matrix, and the orthogonal vectors are Column vectors in . For orthogonal vectors Perform amplitude decomposition to obtain amplitude data. Extract the modulus of the vector, that is, calculate the amplitude of each orthogonal vector. The amplitude data is expressed as , obtained by taking the absolute value of each component and processing the result. On this basis, the weight matrix is ​​constructed according to the amplitude data , each column of the matrix represents the amplitude weight of different antennas. After constructing the amplitude and phase weight coefficients of the antenna array, the beamforming stage begins. Beamforming is to adjust the transmission direction of the signal according to the amplitude and phase weight coefficients of the antenna array so that the signal can be concentrated in the target direction. In this process, the initial parameters are calculated by the beamforming module, and the initial parameters are spatially scanned to obtain the beam parameters. Spatial scanning is achieved by changing the phase or weight of the array, and the beam parameters obtained are Including direction, gain and other information. When the beam parameters After determination, the CSI feature vector is compensated and analyzed. The transmission characteristics of the signal are adjusted according to the current state of the channel. Compensation sequence It is obtained by analyzing the channel state information CSI and calculated by an optimization algorithm, such as the minimum mean square error (MMSE). The transmission signal is adjusted according to the channel conditions so that the signal can be enhanced as much as possible at the receiving end. Calculate the compensation factor , these coefficients are used to adjust the strength or phase of the transmitted signal to optimize the quality of signal reception. Obtained by solving optimization problems such as:

[0183] ;

[0184] in, is the received signal, is the channel matrix, is the compensation coefficient. After optimizing the compensated beam parameters, the target parameters are obtained. . Target parameters Contains the final transmission characteristics of the signal, such as frequency, gain, etc. The encoding process converts the target parameters into a signal format suitable for transmission, such as quantizing or modulating it through a digital signal processing algorithm so that it can be sent in actual communication. and coding sequences , perform precoding calculation and obtain the precoded signal , the signal is properly modulated and encoded before it is finally transmitted. The purpose of the precoding process is to adjust the transmission signal according to the channel state information so that the signal quality can be maximized at the receiving end.

[0185] Reference Figure 2 , this embodiment provides an integrated airborne communication antenna device, including:

[0186] Acquisition module 1, used to perform CSI data acquisition and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data;

[0187] Processing module 2, used to perform residual vector quantization operation on CSI spatiotemporal feature data to generate a quantized data stream, and perform recursive encoding processing on the quantized data stream to obtain a target data stream;

[0188] Operation module 3 is used to map the target data stream to a three-dimensional tensor space, construct a real-valued tensor model, and apply unitary parallel factor decomposition and regularized least squares iterative operations to the real-valued tensor model to obtain a CSI feature vector;

[0189] Decomposition module 4 is used to perform fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and input the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result;

[0190] The generation module 5 is used to calculate the amplitude and phase weight coefficients of the antenna array according to the identity authentication result and the target data stream, generate beam parameters, and perform compensation matrix operations on the beam parameters according to the CSI eigenvector to obtain a precoded signal.

[0191] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0192] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0193] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0194] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0195] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0196] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0197] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A communication method for an integrated airborne communication antenna device, characterized in that: The following steps are involved: Perform CSI data collection and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data; Performing a residual vector quantization operation on the CSI spatiotemporal feature data to generate a quantized data stream, and performing recursive encoding processing on the quantized data stream to obtain a target data stream; Mapping the target data stream to a three-dimensional tensor space, constructing a real-valued tensor model, and applying unitary parallel factor decomposition and regularized least squares iterative operations to the real-valued tensor model to obtain a CSI feature vector; Performing fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and inputting the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result; The amplitude and phase weight coefficients of the antenna array are calculated according to the identity authentication result and the target data stream to generate beam parameters, and a compensation matrix operation is performed on the beam parameters according to the CSI eigenvector to obtain a precoded signal.

2. The communication method of the integrated airborne communication antenna device according to claim 1, characterized in that: The performing CSI data acquisition and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data includes: Perform time-series sampling on the communication channel to obtain the CSI original sequence; Performing amplitude extraction and phase extraction on the CSI original sequence to obtain CSI amplitude data and CSI phase data, and performing feature fusion on the CSI amplitude data and the CSI phase data to obtain a CSI feature sequence; Performing time domain feature decomposition on the CSI feature sequence to obtain time domain data, and performing space domain feature decomposition on the CSI feature sequence to obtain space domain data; Performing a focus modulation operation on the time domain data and the spatial domain data to obtain a modulation coefficient, and performing a weighted combination on the time domain data and the spatial domain data according to the modulation coefficient to obtain a feature vector; Performing an orthogonal transformation on the feature vector to obtain an orthogonal basis vector, and projecting the CSI feature sequence onto the orthogonal basis vector to obtain a projection matrix; Feature reconstruction is performed according to the projection matrix to obtain a reconstructed sequence, nonlinear mapping is performed on the reconstructed sequence to obtain mapping data, and singular value decomposition and dimensionality transformation are performed on the mapping data to obtain CSI spatiotemporal feature data.

3. The communication method of the integrated airborne communication antenna device according to claim 2, characterized in that: The performing a residual vector quantization operation on the CSI spatiotemporal feature data to generate a quantized data stream, and performing recursive encoding processing on the quantized data stream to obtain a target data stream includes: Blocking the CSI spatiotemporal feature data to obtain a feature data group, and performing a normalization operation on the feature data group to obtain a normalized vector; Performing codebook generation on the normalized vector to obtain an initial codebook, and setting a quantization parameter threshold according to the initial codebook to obtain a quantization control matrix; Inputting the normalized vector into a first-level quantization module for vector quantization to obtain first-level quantized data; performing a difference operation on the first-level quantized data and the normalized vector to obtain target residual data; Inputting the target residual data into a second-level quantization module for vector quantization to obtain second-level quantized data, and cascading the first-level quantized data and the second-level quantized data to obtain a quantized data stream; Performing hierarchical coding on the quantized data stream to obtain a coding sequence, and performing data interleaving on the coding sequence to obtain an interleaving matrix; Recursively processing the interleaving matrix according to the quantization control matrix to obtain coded data, and dynamically updating the coded data to obtain an updated codebook; The coded data is reconstructed according to the updated codebook to obtain a reconstructed sequence, and the reconstructed sequence is checked and error-corrected to obtain a target data stream.

4. The communication method of the integrated airborne communication antenna device according to claim 3, characterized in that: The target data stream is mapped to a three-dimensional tensor space, a real-valued tensor model is constructed, and unitary parallel factor decomposition and regularized least squares iterative operation are applied to the real-valued tensor model to obtain a CSI feature vector, including: Performing time dimension processing on the target data stream to obtain a time series, and performing space dimension mapping on the time series to obtain a space matrix; Slicing the spatial matrix into blocks to obtain a slice set, and performing three-dimensional reconstruction on the slice set to obtain a three-dimensional tensor; Performing data alignment and normalization on the three-dimensional tensor to obtain a standardized tensor, and performing factorization processing based on the standardized tensor to construct a real-valued tensor model; Performing a forward operation on the real-valued tensor model to obtain a forward tensor, and performing a backward operation on the real-valued tensor model to obtain a backward tensor; Performing unitary decomposition on the forward tensor and the backward tensor to obtain a unitary characteristic matrix, and performing parallel factor decomposition based on the unitary characteristic matrix to obtain decomposition parameters; Inputting the decomposition parameters into an iterative optimizer to obtain an iterative sequence, and performing a convergence test on the iterative sequence to obtain a convergence index; The decomposition parameters are optimized and calculated according to the convergence index to obtain a CSI feature vector.

5. The communication method of the integrated airborne communication antenna device according to claim 4, characterized in that: The performing fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and inputting the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result, including: Performing fractional wavelet transform on the CSI feature vector to obtain a transform sequence, and performing nonlinear transform on the transform sequence to obtain nonlinear features; Performing wavelet scattering decomposition on the nonlinear characteristics to obtain a scattering matrix, and performing grouping processing on the scattering matrix to obtain a set of radio frequency features; Distributing the radio frequency feature set to a federated learning node to obtain a distributed feature sequence, and performing local calculation on the distributed feature sequence to obtain a local feature; Performing encryption operation on the local features to obtain encrypted data, and transmitting the encrypted data to a central processing node to obtain global features; The global feature is decrypted to obtain decrypted data, and the decrypted data is input into a federated fractional scattering network for feature aggregation to obtain an identity authentication result.

6. The communication method of the integrated airborne communication antenna device according to claim 5, characterized in that: The decrypting process of the global feature to obtain decrypted data, and inputting the decrypted data into a federated fractional scattering network for feature aggregation to obtain an identity authentication result includes: Symmetrically decrypting the global feature to obtain decrypted data; performing integrity check on the decrypted data to obtain a check matrix; Inputting the check matrix into a feature extraction module of a federated fractional scattering network, wherein the feature extraction module comprises three layers of convolution processing units, each layer comprising a convolution operation layer, a batch normalization layer and an activation function layer, and obtaining a feature map through convolution calculation; Performing dimensionality reduction processing on the feature graph to obtain a dimensionality reduction matrix, and inputting the dimensionality reduction matrix into a feature fusion module of the federated fractional scattering network, wherein the feature fusion module combines features using an attention mechanism and a residual structure to obtain a fusion vector; Inputting the fusion vector into a classification module of the federated fractional scattering network, wherein the classification module comprises four layers of fully connected units, each layer of fully connected units uses regularization and normalization processing, and obtains a classification vector through forward propagation; Performing threshold filtering on the classification vector to obtain a filtering result, performing multi-layer perception classification on the filtering result to obtain a first authentication value, performing confidence calculation on the first authentication value to obtain a confidence vector, generating a confusion matrix according to the confidence vector to obtain a second authentication value; The first authentication value and the second authentication value are input into a decision module, a fusion value is obtained through weighted calculation, and a threshold judgment is performed on the fusion value. When the fusion value is greater than a preset threshold, an identity authentication result is output.

7. The communication method of the integrated airborne communication antenna device according to claim 6, characterized in that: The calculating the amplitude and phase weight coefficients of the antenna array according to the identity authentication result and the target data stream, generating beam parameters, and performing compensation matrix operation on the beam parameters according to the CSI eigenvector to obtain a precoded signal includes: Performing feature combination on the identity authentication result and the target data stream to obtain combined data, and performing adaptive calculation on the combined data to obtain a weight vector; Performing antenna array response analysis on the weight vector to obtain a response matrix, and performing orthogonal processing on the response matrix to obtain an orthogonal vector; Performing amplitude decomposition on the orthogonal vector to obtain amplitude data, and constructing a weight matrix according to the amplitude data to obtain amplitude and phase weight coefficients of the antenna array; Inputting the amplitude and phase weight coefficients into a beamforming module to obtain initial parameters, and performing spatial scanning on the initial parameters to obtain beam parameters; Performing compensation analysis on the CSI feature vector to obtain a compensation sequence, and calculating a compensation coefficient according to the compensation sequence; Performing compensation calculation on the beam parameters to obtain compensation parameters, and optimizing the compensation parameters according to the compensation coefficients to obtain target parameters; The target parameter is encoded to obtain a coding sequence, and a precoding calculation is performed based on the coding sequence to obtain a precoded signal.

8. An integrated airborne communication antenna device, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: An acquisition module is used to perform CSI data acquisition and spatiotemporal focus modulation on the communication channel to obtain CSI spatiotemporal feature data; A processing module, configured to perform a residual vector quantization operation on the CSI spatiotemporal feature data to generate a quantized data stream, and perform recursive encoding processing on the quantized data stream to obtain a target data stream; A computing module, used for mapping the target data stream into a three-dimensional tensor space, constructing a real-valued tensor model, and applying unitary parallel factor decomposition and regularized least squares iterative operations to the real-valued tensor model to obtain a CSI feature vector; A decomposition module, used for performing fractional wavelet transform decomposition on the CSI feature vector to obtain a radio frequency feature set, and inputting the radio frequency feature set into a federated fractional scattering network for feature processing to obtain an identity authentication result; A generation module is used to calculate the amplitude and phase weight coefficients of the antenna array according to the identity authentication result and the target data stream, generate beam parameters, and perform compensation matrix operations on the beam parameters according to the CSI eigenvector to obtain a precoded signal.

Images