A secure communication method and system based on wireless transmission

Abstract

This invention discloses a secure communication method and system based on wireless transmission, comprising the following steps: acquiring data to be transmitted and mapping it to generate a complex data sequence; inputting an improved CVNN, weighting and fusing it by phase and amplitude branches to generate a multi-branch feature sequence; performing multi-scale residual stream diffusion to construct a high-dimensional secure tensor in time, frequency, and space; extracting endogenous features and channel statistical features to form a secure feature matrix; generating an initial dynamic key and iteratively updating it to form a hierarchical dynamic key coupled with the channel; combining the key with the complex data sequence for encryption and redundant encoding to generate an encrypted data frame; the receiving end decrypts and verifies the data, and when the security level is lower than a threshold, triggering the sending end's DropConnect to randomly block and iteratively update the key and encryption strategy until transmission is complete. This invention achieves dynamic key iteration, physical layer protection, and low-error-rate reliable transmission for secure wireless data transmission.

Description

Technical Field

[0001] This invention relates to the field of communication security technology, and in particular to a secure communication method and system based on wireless transmission. Background Technology

[0002] Existing wireless communication systems typically rely on traditional encryption algorithms and fixed-key mechanisms to ensure data security during data transmission. Data packets are processed using XOR operations and simple checksums with a static key. After decryption and integrity verification, the receiving end restores the data to its original sequence. Traditional methods have limitations when dealing with dynamic wireless channel environments and complex hardware characteristics. Due to fixed-key or simple dynamic key update strategies, attackers can potentially predict or crack the encryption process by analyzing channel characteristics or repetitive data patterns, leading to insufficient physical layer data security.

[0003] Existing methods are typically based on real-number signal processing for feature extraction and encryption key generation, which do not fully utilize the higher-order correlations of phase and amplitude information, resulting in limited randomness and unpredictability of generated keys. The diffusion capabilities of multi-branch signal features in time, frequency and spatial dimensions are insufficient, making it difficult to form a high-dimensional security feature tensor that is tightly coupled with channel features, thereby limiting the complexity and resistance to attacks of the keys.

[0004] In existing data transmission processes, there is a lack of a closed-loop control mechanism for security feedback at the receiving end. It is impossible to dynamically adjust the encryption strategy and key generation based on the actual security score at the receiving end, and the security guarantee is not adaptive. At the same time, traditional systems do not make sufficient use of inter-channel correlation and hardware fingerprint features, and cannot achieve accurate feature weighting and iterative updates on multi-channel complex signals, resulting in the inability to fully enhance the security of encrypted data frames at the physical layer.

[0005] The aforementioned technical limitations restrict the security and reliability of wireless communication systems in high-risk environments. Therefore, how to provide a secure communication method and system based on wireless transmission is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a secure communication method and system based on wireless transmission. This invention fully utilizes complex data sequences, multi-branch feature extraction, dynamic key iteration, and DropConnect random masking mechanism, and describes in detail the operation steps of physical layer security fingerprint generation, key coupling, and encrypted data frame transmission. It has the advantages of strong anti-attack capability, unpredictable keys, and secure and reliable data transmission.

[0007] A secure communication method based on wireless transmission according to an embodiment of the present invention includes the following steps: Collect the data to be transmitted, map it into I / Q complex signals, standardize and frame the complex signals, and generate complex data sequences; The complex data sequence is input into the improved CVNN. A weighted adaptive fusion mechanism based on the complex covariance matrix is ​​established between the phase branch and the amplitude branch to extract the high-order covariance tensor between features and generate a multi-branch complex feature sequence. Multi-scale residual flow is applied to multi-branch complex feature sequences to diffuse the features to three scales: time, frequency, and space, thereby constructing a high-dimensional secure tensor with unidirectional irreversibility. Based on the high-dimensional security tensor, the intrinsic features of the wireless signal and the instantaneous statistical features of the channel are extracted to generate the physical layer security fingerprint, and a security feature matrix is ​​formed by dynamic weighting. An initial dynamic key is generated based on the security feature matrix, and iteratively updated in the local and global ranges through a nonlinear mapping algorithm to generate a hierarchical dynamic key coupled with the instantaneous characteristics of the channel. By combining the hierarchical dynamic key with the complex data sequence, block encryption, integrity verification and redundancy coding are performed to generate encrypted data frames, which are then sent to the receiving end via a wireless channel. The receiving end decrypts, verifies the integrity of, and assesses the security of the encrypted data frames. When the security level is lower than a preset threshold, it sends a trigger signal to the sending end through the feedback link to activate the DropConnect connection random masking mechanism of the improved CVNN hidden layer of the sending end. It iteratively adjusts the non-linear step of the key update frequency and encryption strategy until all data frames are securely transmitted and the final received data set is generated.

[0008] Optionally, the process of generating the multi-branch complex feature sequence includes: The complexized data sequence is input into the phase branch and the amplitude branch respectively, and a linear mapping operation is performed on the I component and the Q component to generate a preliminary complex feature vector. The complex covariance matrix is ​​calculated from the initial complex eigenvectors of the phase branch and the amplitude branch to obtain the complex domain correlation weight matrix; Based on the complex domain correlation weight matrix, the complex eigenvectors of the phase branch and the amplitude branch are weighted and linearly fused to generate a fused complex eigenvector. Normalize the fused complex feature vectors to generate normalized complex feature vectors; The normalized complex feature vectors are input into the next CVNN hidden layer to form a multi-branch complex feature sequence; In each training iteration, the fusion weight matrix is ​​updated based on the change in the covariance matrix of the complex feature vector output in the previous round. The complex feature vectors of the phase branch and the amplitude branch are weighted and fused to form the updated fused complex feature vector, which is then input into the next CVNN hidden layer to generate a multi-branch complex feature sequence.

[0009] Optionally, the construction process of the high-dimensional security tensor includes: The multi-branch complex feature sequence is divided into fixed-length subsequences according to the time dimension, and a one-dimensional convolution operation is performed on each subsequence to generate a time-scale feature sequence. Perform a short-time Fourier transform on the time-scale feature sequence in the frequency dimension to map the complex features to the spectral representation, and perform a two-dimensional convolution operation on the spectral features to generate a frequency-scale feature sequence. A channel topology mapping is established in the spatial dimension for the frequency scale feature sequence. The complex features of each channel are weighted and superimposed with the complex features of neighboring channels to form a spatial scale feature sequence. The feature sequences at the time, frequency, and spatial scales are respectively concatenated using residuals. Each scale feature is added to the original input feature to form the residual output at each scale. The residual outputs at each scale are then superimposed in the order of time, frequency, and space to generate a high-dimensional complex feature tensor with unidirectional irreversibility.

[0010] Optionally, the process of generating the security feature matrix includes: The high-dimensional complex feature tensor is divided into fixed-size sub-blocks according to the three dimensions of time, frequency and space. For each sub-block, the amplitude mean, phase mean and amplitude-phase covariance of the I / Q components are calculated to generate the complex feature vector of the sub-block. Construct a matrix by arranging the complex eigenvectors of all sub-blocks into a channel sequence, calculate the complex cross-correlation matrix between the eigenvectors of each channel and its neighboring channels, and obtain the statistical characteristic matrix between channels; The complex eigenvector of each channel is multiplied element-wise with the corresponding inter-channel statistical feature matrix to form an initial weighted eigenvector sequence; The initial weighted feature vector sequence is accumulated by channel to generate a preliminary security feature matrix; In the first iteration, the zero matrix is ​​used as the security feature matrix of the previous iteration. In subsequent iterations, the updated security feature matrix of the previous iteration is used to calculate the element-wise difference. The difference is multiplied by the channel weighted learning rate of the security feature matrix and accumulated to the corresponding channel weight of the original security feature matrix to form the updated security feature matrix. The updated security feature matrix is ​​normalized by channel, with the sum of the weighting coefficients of all channels kept to 1, and the normalized matrix is ​​output.

[0011] Optionally, the generation process of the hierarchical dynamic key includes: The complex eigenvalues ​​corresponding to each channel of the security feature matrix are divided into blocks of fixed length, and the amplitude and phase components are extracted sequentially. The amplitude and phase values ​​are converted into fixed index bits of the key binary sequence through a mapping function. The index bits are determined according to the channel order and the length of the subsequence. The eigenvalues ​​of each channel are mapped to predefined bit positions through modulo operation. All channel mapping results are concatenated to form the initial dynamic key. The initial dynamic key is input into a nonlinear mapping algorithm, which performs nonlinear permutations and XOR operations on subsequences of the key within a local range, and updates the position of the subsequences according to the channel index order to form a locally updated key. Globally, a non-linear cyclic shift operation is performed on the local update key to map the channel amplitude and phase values ​​of the security feature matrix to the corresponding bit positions in the key binary sequence according to the mapping function. The key segment order is adjusted according to the channel order to form the global update key. Repeat the local update and global update operations for several rounds, using the key generated in the previous round as input in each round, until the preset number of iterations is reached, and output the dynamic key after the iteration is completed; The iteratively completed dynamic key is mapped bit-level to the instantaneous characteristics of the channel using a mapping function. The channel amplitude, phase, and delay characteristics are mapped to the corresponding bit positions of the key. The key segmentation order is adjusted according to the channel order to form a hierarchical dynamic key coupled with the instantaneous characteristics of the channel.

[0012] Optionally, the process of performing block encryption, integrity verification, and redundancy coding includes: The complex data sequence is divided into fixed-length data blocks. Each data block is XORed bit by bit with the corresponding length of the hierarchical dynamic key to generate an initial encrypted block. Calculate the cyclic redundancy check code for the initial encrypted packets, and append the generated check code to the end of each packet to form a data packet that has been verified. For each data packet that has been verified, perform redundant encoding, copy the data packet or generate redundant bits according to a preset, insert the redundant bits in the order of the packets to generate an encrypted data frame, and each packet can be completely recovered at the corresponding position at the receiving end; The generated encrypted data frames are sent to the receiving end via a wireless channel, and the packet sequence number and timestamp are recorded during the transmission process.

[0013] Optionally, the process of decryption, integrity verification, and security assessment includes: Extract each data packet, its additional redundant bits, and CRC checksum from the received encrypted data frame according to the packet sequence number; Each data group is XORed bit by bit with the corresponding hierarchical dynamic key to generate a preliminary decryption bit sequence; The initial decrypted bit sequence is divided modulo-2 using a preset generator polynomial. The CRC remainder is calculated, and the remainder is compared bit by bit with the additional check code to generate an integrity verification matrix. For data groups that pass integrity verification, the amplitude and phase characteristics of the complex I / Q bit sequences are statistically analyzed. The amplitude and phase values ​​of each channel are compared element-by-element with the corresponding security feature matrix to generate a security scoring matrix.

[0014] Optionally, the DropConnect connection random masking mechanism specifically includes: The multi-branch complex feature sequence output channel of the improved CVNN hidden layer at the sending end is divided into channels, and each channel is assigned a random masking probability value. The random masking probability value is calculated by a mapping function. Based on the random masking probability, binary random sampling is performed on each channel to generate a masking matrix. The channel with a masking value of 0 has its corresponding output set to zero in this iteration, so that the channel does not participate in the calculation of dynamic key generation in the next iteration. The channel with a masking value of 1 retains its original output. The generated masking matrix is ​​applied to the output channel of the improved CVNN multi-branch complex feature sequence to obtain the hidden layer output after masking adjustment; After each iteration, the random shielding probability is updated based on the difference between the security feature matrix and the security score matrix fed back by the receiver. The channel shielding sampling and zeroing operation are then re-executed to form a new shielding adjustment output. The output of the masked and adjusted hidden layer is used to regenerate the hierarchical dynamic key. This process is repeated iteratively until all data frames are securely transmitted and the final received data set is output.

[0015] A secure communication system based on wireless transmission according to an embodiment of the present invention includes the following modules: The data acquisition module is used to acquire the data to be transmitted, map it into I / Q complex signals, complete standardization and frame division, and generate complex data sequences. The complex neural network feature extraction module is used to receive complex data sequences, establish covariance weighted fusion of phase and amplitude branches, and generate multi-branch complex feature sequences; The multi-scale residual flow processing module is used to perform time, frequency and spatial diffusion on multi-branch complex feature sequences to generate high-dimensional secure tensors with unidirectional irreversibility. The security feature generation module is used to extract endogenous features and channel statistical features based on a high-dimensional security tensor, and form a security feature matrix through dynamic weighting. The dynamic key generation module is used to generate an initial dynamic key based on the security feature matrix, and to form a hierarchical dynamic key through local and global nonlinear iterative updates; The encryption processing module is used to combine the hierarchical dynamic key with the complex data sequence, perform block encryption, integrity verification and redundancy coding, and generate encrypted data frames. The receiver security feedback module is used to decrypt encrypted data frames, verify their integrity, and assess their security. When the security level is below a threshold, it triggers the sender's DropConnect blocking mechanism through the feedback link.

[0016] The beneficial effects of this invention are: (1) By complexifying the data sequence, extracting features from multiple branches and processing residual streams at multiple scales, a high-dimensional security tensor is constructed, which enables the dynamic key generation to be tightly coupled with physical signal features, thereby improving the unpredictability and resistance to attacks of the key. (2) A hierarchical dynamic key is generated by dynamic weighting of the security feature matrix and nonlinear iterative mapping, combined with the DropConnect random masking mechanism, to realize dynamic adjustment of the hidden layer channel at the sending end, thereby enhancing the adaptability and security of the physical layer encryption. (3) By using packet encryption, CRC check and redundancy coding, and receiver security score feedback, an end-to-end closed-loop control is formed to ensure that data frames are reliably transmitted in the wireless channel, and to achieve data security, integrity and traceability. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a secure communication method based on wireless transmission proposed in this invention; Figure 2 This is a module connection diagram of a secure communication system based on wireless transmission proposed in this invention; Figure 3 This is a schematic diagram of the improved CVNN multi-branch feature extraction technique proposed in this invention. Figure 4 This is a schematic diagram of the encryption process and receiver feedback proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-4 A secure communication method based on wireless transmission includes the following steps: Collect the data to be transmitted, map it into I / Q complex signals, standardize and frame the complex signals, and generate complex data sequences; The complex data sequence is input into the improved CVNN. A weighted adaptive fusion mechanism based on the complex covariance matrix is ​​established between the phase branch and the amplitude branch to extract the high-order covariance tensor between features and generate a multi-branch complex feature sequence. Multi-scale residual flow is applied to multi-branch complex feature sequences to diffuse the features to three scales: time, frequency, and space, thereby constructing a high-dimensional secure tensor with unidirectional irreversibility. Based on the high-dimensional security tensor, the intrinsic features of the wireless signal and the instantaneous statistical features of the channel are extracted to generate the physical layer security fingerprint, and a security feature matrix is ​​formed by dynamic weighting. An initial dynamic key is generated based on the security feature matrix, and iteratively updated in the local and global ranges through a nonlinear mapping algorithm to generate a hierarchical dynamic key coupled with the instantaneous characteristics of the channel. By combining the hierarchical dynamic key with the complex data sequence, block encryption, integrity verification and redundancy coding are performed to generate encrypted data frames, which are then sent to the receiving end via a wireless channel. The receiving end decrypts, verifies the integrity of, and assesses the security of the encrypted data frames. When the security level is lower than a preset threshold, it sends a trigger signal to the sending end through the feedback link to activate the DropConnect connection random masking mechanism of the improved CVNN hidden layer of the sending end. It iteratively adjusts the non-linear step of the key update frequency and encryption strategy until all data frames are securely transmitted and the final received data set is generated.

[0020] In this embodiment, the process of generating a complex data sequence includes: The acquired data to be transmitted is first processed by baseband, which maps the original digital data into a complex domain representation, where the I component represents the in-phase component of the signal and the Q component represents the quadrature component. The mapping process can be completed by mapping each data bit to a set of complex symbols, for example, by generating the corresponding I / Q value through amplitude or phase modulation, so that each data unit has a unique representation in the complex plane. After mapping, the complex signal is standardized to unify the amplitude range of the I and Q components to a preset value range, thereby eliminating the influence of sampling offset and amplitude drift on signal characteristics and ensuring the stability of complex neural network processing. The standardized complex signal is further divided into frames. The continuous complex signal is divided into frame units according to a fixed time window or number of samples. Each frame contains a complete I / Q sample sequence. Timestamps or sequence number information are added to the frame boundaries to ensure that the feature sequences of multiple frames maintain sequential consistency and time dependence in the subsequent improved CVNN. The I / Q data sequences of each frame are integrated to generate complex data sequences, which serve as the basic data for the input of the multi-branch complex neural network of this invention. This ensures that each frame retains the complex domain information, amplitude and phase features, and time series structure of the original data, providing complete input for high-order covariance tensor calculation and physical layer security fingerprint extraction.

[0021] In this embodiment, the process of generating a multi-branch complex feature sequence includes: The complexized data sequence is input into the phase branch and the amplitude branch respectively, and a linear mapping operation is performed on the I component and the Q component to generate a preliminary complex feature vector. Each I / Q sample of the complexized data sequence is multiplied with the branch weight matrix, and the original complex signal is mapped to the hidden feature space through matrix multiplication. Each complex element participates in the mapping of the I component and the Q component at the same time, and a preliminary complex feature vector sequence is obtained. Calculate the complex covariance matrix for the initial complex eigenvectors of the phase branch and amplitude branch to obtain the complex domain correlation weight matrix; calculate the covariance of the complex elements between each channel for the initial complex eigenvectors of the phase branch and amplitude branch respectively, regard the I component and Q component as the real part and imaginary part of the complex vector, and obtain the covariance value of each pair of channels according to the complex covariance formula to form the complex domain correlation weight matrix. Based on the complex domain correlation weight matrix, the complex eigenvectors of the phase branch and the amplitude branch are weighted and linearly fused to generate a fused complex eigenvector; based on the complex domain correlation weight matrix, the complex eigenvectors of the corresponding channels of the phase branch and the amplitude branch are weighted and summed, and the weight of each channel is determined by the corresponding elements of the covariance matrix to obtain the fused complex eigenvector sequence. Normalize the fused complex feature vectors to generate normalized complex feature vectors; The normalized complex feature vectors are input into the next CVNN hidden layer to form a multi-branch complex feature sequence; In each training iteration, the fusion weight matrix is ​​updated according to the change in the covariance matrix of the complex feature vector output in the previous round. The complex feature vectors of the phase branch and the amplitude branch are weighted and fused to form the updated fused complex feature vector, which is then input into the next CVNN hidden layer to generate a multi-branch complex feature sequence. In each iteration cycle, the covariance matrix of the complex feature vector fused in the previous round is compared with the covariance matrix of the previous iteration element by element to obtain the covariance change of each channel. Based on the covariance change of each channel, the fusion weight of the corresponding channel is adjusted linearly or proportionally: if the covariance change is positive, the fusion weight of that channel is increased proportionally; if the covariance change is negative, it is decreased proportionally. The adjustment range can be controlled by a preset learning rate. In each iteration, the weight is limited to the range of 0 to 1. The fusion weights of all channels are normalized so that the sum of the weights is kept at 1.

[0022] In this embodiment, the process of constructing the high-dimensional security tensor includes: The multi-branch complex feature sequence is divided into fixed-length subsequences according to the time dimension, and a one-dimensional convolution operation is performed on each subsequence to generate a time-scale feature sequence. Perform a short-time Fourier transform on the time-scale feature sequence in the frequency dimension to map the complex features to the spectral representation, and perform a two-dimensional convolution operation on the spectral features to generate a frequency-scale feature sequence. A channel topology mapping is established in the spatial dimension for the frequency scale feature sequence. The complex features of each channel are weighted and superimposed with the complex features of neighboring channels to form a spatial scale feature sequence. The feature sequences at the time, frequency, and spatial scales are respectively concatenated using residuals. Each scale feature is added to the original input feature to form the residual output at each scale. The residual outputs at each scale are then superimposed in the order of time, frequency, and space to generate a high-dimensional complex feature tensor with unidirectional irreversibility.

[0023] In each iteration cycle, the element-wise difference between the previous high-dimensional complex feature tensor and the current convolution output is calculated. The difference is multiplied by the learning rate and accumulated into the residual flow convolution kernel weights. The covariance difference between the channel features and the neighboring channels is calculated, multiplied by the topological learning rate and accumulated into the channel topological weighting coefficients. The difference between the previous residual output and the current input features is multiplied by the residual connection learning rate and accumulated into the residual connection matrix. The next round of multi-scale residual flow operation is performed using the updated convolution kernel weights, channel topological weighting coefficients, and residual connection matrix.

[0024] In this embodiment, the process of generating the security feature matrix includes: The high-dimensional complex feature tensor is divided into fixed-size sub-blocks according to the three dimensions of time, frequency and space. For each sub-block, the amplitude mean, phase mean and amplitude-phase covariance of the I / Q components are calculated to generate the complex feature vector of the sub-block. Construct a matrix by arranging the complex eigenvectors of all sub-blocks into a channel sequence, calculate the complex cross-correlation matrix between the eigenvectors of each channel and its neighboring channels, and obtain the statistical characteristic matrix between channels; Specifically, the complex feature vectors of each channel in the multi-branch complex feature sequence are arranged in a fixed order to form a two-dimensional matrix. Each row corresponds to the complex feature vector of one channel, and each column corresponds to the complex elements of the same channel at different times, frequencies, or spatial locations. This matrix serves as the basic structure for statistical calculations between channels and subsequent weighting operations, ensuring that the feature order of each channel is traceable and consistent. For each channel in the two-dimensional channel matrix, adjacent channels are selected, such as adjacent channels before and after or topologically adjacent channels. The complex feature vector of the current channel is multiplied element-wise with the complex feature vectors of the adjacent channels and summed to obtain the complex cross-correlation value. This operation is repeated for all channels and their adjacent channels to generate a complex cross-correlation matrix, where each element of the matrix represents the degree of complex correlation between the corresponding channel pairs, which is used for dynamic weighting operations. The complex eigenvector of each channel is multiplied element-wise with the corresponding inter-channel statistical feature matrix to form an initial weighted eigenvector sequence; The initial weighted feature vector sequence is accumulated by channel to generate a preliminary security feature matrix; In the first iteration, the zero matrix is ​​used as the security feature matrix of the previous iteration. In subsequent iterations, the updated security feature matrix of the previous iteration is used to calculate the element-wise difference. The difference is multiplied by the channel weighted learning rate of the security feature matrix and accumulated to the corresponding channel weight of the original security feature matrix to form the updated security feature matrix. The channel weighted learning rate of the security feature matrix is ​​a scalar parameter used to iteratively update the weighted coefficients of each channel. In each iteration, the difference of the current channel weighted coefficients is multiplied by the element-wise difference of the security feature matrix output in the previous round, and then multiplied by the learning rate to obtain the weighted adjustment value for this round. By adjusting this learning rate, the magnitude and convergence speed of the channel weighted coefficient updates in each iteration can be controlled. The learning rate can be a fixed value or can be adaptively adjusted according to the iteration round or the channel feature variance to ensure that the weighted matrix remains numerically stable in each iteration. The updated security feature matrix is ​​normalized by channel, with the sum of the weighting coefficients of all channels kept to 1, and the normalized matrix is ​​output.

[0025] In this invention, the physical layer security fingerprint is generated based on a high-dimensional complex feature tensor. The complex feature vector of each channel is processed by dividing the amplitude and phase components into blocks, and the mean amplitude, mean phase, and amplitude-phase covariance of each block are calculated to form the channel's endogenous feature vector. The channel endogenous feature vectors are arranged in channel order to construct a two-dimensional matrix, and the complex cross-correlation matrix between each channel and its neighboring channel feature vectors is calculated to obtain the inter-channel statistical feature matrix. The physical layer security fingerprint is formed by multiplying the endogenous feature vector of each channel element-wise with the corresponding inter-channel statistical feature matrix, and accumulating them by channel to form a preliminary security feature matrix. During the iteration process, the element-wise difference between the preliminary security feature matrix and the previous round output matrix is ​​multiplied by the channel weighted learning rate and accumulated into the original channel weights. At the same time, the weighted matrix is ​​normalized so that the sum of all channel weighting coefficients is kept to 1. The updated security feature matrix is ​​used as the physical layer security fingerprint of this invention.

[0026] In this embodiment, the generation process of the hierarchical dynamic key includes: The complex eigenvalues ​​corresponding to each channel of the security feature matrix are divided into blocks of fixed length, and the amplitude and phase components are extracted sequentially. The amplitude and phase values ​​are converted into fixed index bits of the key binary sequence through a mapping function. The index bits are determined according to the channel order and the length of the subsequence. The eigenvalues ​​of each channel are mapped to predefined bit positions through modulo operation. All channel mapping results are concatenated to form the initial dynamic key. In this invention, the amplitude or phase value is first mapped to an integer value according to a preset quantization level, and then the integer value is combined with the channel number according to a weighted rule to obtain a preliminary index value. The preliminary index value represents the target position of the corresponding bit in the key sequence. The mapping function ensures that each channel feature value has a unique corresponding index position in the key sequence, forming a traceable mapping relationship from channel to key bit. Modulo operation is used to map the initial index value to a predefined bit position in the key binary sequence. The operation is as follows: take the modulo of the initial index value with the total bit length of the key, and use the remainder as the bit index of the key sequence. Write the amplitude or phase quantization value of the corresponding channel into the index bit. Through modulo operation, the feature value mapping of each channel can be evenly distributed within the total key length, ensuring that the mapping does not go out of bounds and keeping the index consistent during iteration and encryption. The initial dynamic key is input into a nonlinear mapping algorithm, which performs nonlinear permutations and XOR operations on subsequences of the key within a local range, and updates the position of the subsequences according to the channel index order to form a locally updated key. Specifically, the initial dynamic key is divided into several fixed-length subsequences. For each subsequence, a permutation index sequence is generated based on the value of each bit in the subsequence and the corresponding channel index. The bit positions of the subsequence are rearranged according to the index sequence to achieve non-linear permutation. Subsequently, the permuted subsequence is XORed bit by bit with the mapped bit sequence of the corresponding channel to generate the updated subsequence. For each subsequence, its position in the overall dynamic key is determined according to the channel index order. That is, the subsequence of the i-th channel is placed in the key sequence at bit positions i×L to (i+1)×L-1, where L is the length of the subsequence. In each iteration, the local update subsequence output from the previous round is used as input, the nonlinear permutation and XOR operation are repeated, and the key sequence is re-inserted according to the channel index order to form a new local update key.

[0027] Globally, a non-linear cyclic shift operation is performed on the local update key to map the channel amplitude and phase values ​​of the security feature matrix to the corresponding bit positions in the key binary sequence according to the mapping function. The key segment order is adjusted according to the channel order to form the global update key. The nonlinear cyclic shift operation is performed by dividing the local update key into multiple subsequences of fixed length. The bits in each subsequence generate a cyclic offset based on the subsequence value and the channel index. The offset can be calculated using a nonlinear function, such as by taking the weighted sum of the integer value of each bit in the subsequence and the channel index, and then taking the modulo of the subsequence length to obtain the cyclic shift step size. The bits in the subsequence are then shifted clockwise or counterclockwise according to the calculated cyclic offset to complete the nonlinear cyclic shift, resulting in the shifted subsequence. Specifically, mapping to the bit positions corresponding to the key binary sequence involves converting the amplitude and phase of the complex eigenvalues ​​corresponding to each channel of the security feature matrix into integer indices using a mapping function for the shifted subsequences. The integer indices are then moduloed by the total key length to obtain the bit positions in the key binary sequence. The amplitude and phase quantization values ​​of the corresponding channels are written into these bit positions, and each subsequence is arranged in channel index order to form a global update key, ensuring that the position of each channel feature in the key sequence is unique and traceable.

[0028] Repeat the local update and global update operations for several rounds, using the key generated in the previous round as input in each round, until the preset number of iterations is reached, and output the dynamic key after the iteration is completed; The iteratively completed dynamic key is mapped bit-level to the instantaneous channel features using a mapping function. The iteratively completed dynamic key is divided into fixed-length bit blocks, each corresponding to a channel. For each channel, the instantaneous amplitude, phase, and delay values ​​of the channel are quantized into integer indices using the mapping function. The integer indices are modulo the bit block length to obtain the target bit position. The amplitude or phase quantization value of the corresponding channel is written into the target bit position of the dynamic key, realizing the bit-level mapping of each channel feature in the key sequence. The channel amplitude, phase, and delay features are mapped to the corresponding bit positions of the key. The key segmentation order is adjusted according to the channel order to form a hierarchical dynamic key coupled with the instantaneous channel features.

[0029] In this embodiment, the process of performing block encryption, integrity verification, and redundancy coding includes: The complex data sequence is divided into fixed-length data blocks. Each data block is XORed bit by bit with the corresponding length of the hierarchical dynamic key. The complex I / Q bit sequence of each data block is aligned bit by bit with the binary sequence of the corresponding length of the hierarchical dynamic key, and XORed bit by bit is performed sequentially to obtain the preliminary encrypted bit sequence. The result of the operation at each bit position replaces the original data bit, keeping the block length unchanged; thus generating the preliminary encrypted block. Calculate the cyclic redundancy check code for the initial encrypted blocks. Use a preset generator polynomial to calculate the cyclic redundancy check code for the initial encrypted bit sequence according to the standard CRC algorithm. Append the generated check code to the end of each block to form a data block that has been verified. For each data packet that has been verified, perform redundant encoding, copy the data packet or generate redundant bits according to a preset, insert the redundant bits in the order of the packets to generate an encrypted data frame, and each packet can be completely recovered at the corresponding position at the receiving end; In this invention, after combining the complex data sequence with the hierarchical dynamic key, the data is divided into data blocks of fixed length. The bit sequence of each block is aligned bit by bit with the corresponding length of the dynamic key binary sequence, and an XOR operation is performed sequentially to obtain a preliminary encrypted bit sequence. The operation result at each bit position replaces the original data bit. A cyclic redundancy check (CRC) calculation is performed on the preliminary encrypted bit sequence using a preset generator polynomial to generate a CRC check code. This check code is then appended to the end of each data block to form an encrypted block with a check code. The preset generator polynomial can be a standard CRC-16 or CRC-32 polynomial, and the remainder generated by modulo 2 division is used as the check code. Based on this, a redundancy encoding operation is performed on the encrypted packets with check codes. The key bits of each packet are copied or generated by XOR, and then inserted into the packets according to the channel order to form a complete encrypted data frame. At the same time, the sequence number and timestamp of each packet are recorded to ensure that the receiving end can perform packet verification and data reassembly based on the redundancy bits and check codes.

[0030] The generated encrypted data frames are sent to the receiving end via a wireless channel, and the packet sequence number and timestamp are recorded during the transmission process.

[0031] In this embodiment, the process of decryption, integrity verification, and security assessment includes: Extract each data packet, its additional redundant bits, and CRC checksum from the received encrypted data frame according to the packet sequence number; Each data group is XORed bit by bit with the corresponding hierarchical dynamic key to generate a preliminary decryption bit sequence; The initial decrypted bit sequence is divided modulo-2 using a preset generator polynomial. The CRC remainder is calculated, and the remainder is compared bit by bit with the additional check code to generate an integrity verification matrix. For data groups that pass integrity verification, the amplitude and phase characteristics of the complex I / Q bit sequences are statistically analyzed. The amplitude and phase values ​​of each channel are compared element-by-element with the corresponding security feature matrix to generate a security scoring matrix.

[0032] In this invention, the security scoring matrix is ​​obtained by comparing the amplitude and phase characteristics of the complex data packets at the receiving end with the amplitude and phase characteristics of the corresponding channels in the preset security feature matrix element by element. Specifically, the cosine similarity between the complex feature vector of each channel and the channel vector of the security feature matrix is ​​calculated, and the cosine similarity result is used as the value of that channel in the security scoring matrix to form a scoring matrix containing all channels. In each iteration, the channel cosine similarity matrices of all packets are summarized into a complete security scoring matrix, which is used to determine whether any packet is below a preset threshold, thereby triggering the feedback signal at the transmitting end.

[0033] In this embodiment, the DropConnect connection random masking mechanism specifically includes: The multi-branch complex feature sequence output channel of the improved CVNN hidden layer at the transmitting end is divided into channels, and each channel is assigned a random masking probability value. The random masking probability value is calculated by a mapping function, which generates the probability value based on the difference between the security score matrix at the receiving end and the maximum score and the nonlinear exponent, so that the masking probability of the corresponding channel increases when the security score decreases. Based on the random masking probability, binary random sampling is performed on each channel to generate a masking matrix. The channel with a masking value of 0 has its corresponding output set to zero in this iteration, so that the channel does not participate in the calculation of dynamic key generation in the next iteration. The channel with a masking value of 1 retains its original output. The generated masking matrix is ​​applied to the output channel of the improved CVNN multi-branch complex feature sequence to obtain the hidden layer output after masking adjustment; After each iteration, the random shielding probability is updated based on the difference between the security feature matrix and the security score matrix fed back by the receiver. The channel shielding sampling and zeroing operation are then re-executed to form a new shielding adjustment output. The output of the masked and adjusted hidden layer is used to regenerate the hierarchical dynamic key. This process is repeated iteratively until all data frames are securely transmitted and the final received data set is output.

[0034] In this embodiment, the calculation and update process of the random shielding probability is as follows: After each iteration, perform the following operation on each channel: Calculate the safety score difference ;in, The current channel security score fed back by the receiving end. This represents the maximum security score. Calculate the probability of random shielding ;in, This is a non-linear exponential parameter that controls the increase in shielding probability as security decreases. The probability value is limited to between 0 and 1; Will Used for random channel masking sampling in the next round of iteration to generate a masking matrix.

[0035] A secure communication system based on wireless transmission according to an embodiment of the present invention includes the following modules: The data acquisition module is used to acquire the data to be transmitted, map it into I / Q complex signals, complete standardization and frame division, and generate complex data sequences. The complex neural network feature extraction module is used to receive complex data sequences, establish covariance weighted fusion of phase and amplitude branches, and generate multi-branch complex feature sequences; The multi-scale residual flow processing module is used to perform time, frequency and spatial diffusion on multi-branch complex feature sequences to generate high-dimensional secure tensors with unidirectional irreversibility. The security feature generation module is used to extract endogenous features and channel statistical features based on a high-dimensional security tensor, and form a security feature matrix through dynamic weighting. The dynamic key generation module is used to generate an initial dynamic key based on the security feature matrix, and to form a hierarchical dynamic key through local and global nonlinear iterative updates; The encryption processing module is used to combine the hierarchical dynamic key with the complex data sequence, perform block encryption, integrity verification and redundancy coding, and generate encrypted data frames. The receiver security feedback module is used to decrypt encrypted data frames, verify their integrity, and assess their security. When the security level is below a threshold, it triggers the sender's DropConnect blocking mechanism through the feedback link.

[0036] Example 1: To verify the feasibility of this invention in practice, it was applied to a high-security data transmission scenario in a wireless communication environment. In this scenario, it is necessary to securely transmit multi-channel digital sensor data to a remote processing unit while preventing external eavesdropping and tampering. Existing technologies, in dynamic channel environments, suffer from slow key update speeds and underutilization of physical layer security features, making encrypted data vulnerable to replay attacks or speculative attacks. This invention comprehensively solves these problems through complex neural networks, multi-branch feature extraction, multi-scale residual flow, high-dimensional security tensor construction, physical layer security fingerprint generation, dynamic key iterative update, and the DropConnect random masking mechanism.

[0037] In this scenario, the data acquisition module first acquires multi-channel sensor signals, mapping the raw digital data into I / Q complex signals. These signals are then standardized and framed to generate complex data sequences. Each frame contains a complete I / Q sample sequence, with sequence number information appended to the frame boundaries. The complex data sequences are then input into an improved complex neural network. The network has phase and amplitude branches, and the weights are adaptively fused using the complex covariance matrix to extract high-order covariance tensors between features, generating multi-branch complex feature sequences. In the multi-scale residual flow processing module, the feature sequences are diffused across time, frequency, and space. A high-dimensional secure tensor with unidirectional irreversibility is constructed using one-dimensional convolution, short-time Fourier transform, and spatial topological mapping.

[0038] The high-dimensional security tensor input security feature generation module extracts the channel's endogenous features and instantaneous statistical features to generate a physical layer security fingerprint, and generates a security feature matrix through channel weighting. In the dynamic key generation module, the amplitude and phase of the security feature matrix are processed by a mapping function and modulo operation to generate an initial dynamic key, which is then iteratively updated through nonlinear permutation, XOR operation, and cyclic shift to form a hierarchical dynamic key coupled with the instantaneous features of the channel. In the encryption processing module, the hierarchical dynamic key is combined with the complexized data sequence, and bit-by-bit XOR operation, CRC check, and redundancy coding are performed to generate an encrypted data frame and transmit it to the receiving end through a wireless channel.

[0039] The receiving end security feedback module performs decryption, CRC integrity verification, and security score matrix calculation on the encrypted data frames. The complex features of each channel are compared with the security feature matrix using cosine similarity. When the security is lower than the preset threshold, a trigger signal is sent to the sending end through the feedback link to start the CVNN hidden layer DropConnect connection random masking mechanism, which sets the output of the low-security channel to zero and re-iterates the generation of hierarchical dynamic keys and encrypted frames until all data frames are securely transmitted.

[0040] In the experiment, 10 channels of analog sensor signals (amplitude range 0~1, random noise amplitude ±0.05) were transmitted through the system of this invention, with 1024 I / Q samples per frame and 5 iterations. High-order covariance tensors were calculated in time, frequency, and spatial dimensions using multi-branch features extracted by a complex neural network. The channel weighted error of the generated security feature matrix was controlled within ±0.02. A dynamic key was generated using a mapping function, with the key entropy increasing by approximately 0.35 bits per bit after each iteration. The encrypted data frames, after CRC verification and redundant encoding, exhibited a bit error rate of less than 0.05% during transmission, with an average security score matrix of 0.92. Channels with a score below 0.85 triggered DropConnect to reset to zero and re-iterate key generation until the overall security score reached above 0.95.

[0041] Experimental results show that the system can maintain the integrity of the physical layer security fingerprint in high-noise, multi-channel environments. The dynamic key iteration and DropConnect mechanism effectively prevent leakage of low-security channel outputs, while redundant coding ensures reliable data recovery during wireless transmission.

[0042] Table 1: Comparison Data of Secure Communication Implementation Examples As shown in Table 1 above, in a ten-channel environment, each frame contains 1024 complex samples, and the initial key entropy reaches 7.2 bits / bit, indicating high randomness and complexity in key generation. The average security score is 0.92, indicating that the physical layer security fingerprint combined with a multi-branch complex neural network and dynamic key iteration method can effectively maintain signal security. The bit error rate is controlled at 0.05%, indicating that in a high-dimensional complex signal and multi-channel transmission environment, encrypted data frames generated through block encryption, integrity verification, and redundancy coding can be reliably transmitted without significant data loss or errors. The number of DropConnect triggers is 4, reflecting that the system can dynamically shield low-security channels during the iteration process, further ensuring data frame security.

[0043] Compared to the control group that did not use CVNN and DropConnect, the initial key entropy was only 6.5, the average security score decreased to 0.78, and the bit error rate increased to 0.23%. Furthermore, the DropConnect mechanism was not triggered, indicating that the traditional method is significantly inadequate in terms of security and data integrity under complex channels. The control group using only static keys had an initial key entropy of 6.7, a security score of 0.81, and a bit error rate of 0.18%. Similarly, the DropConnect mechanism was not triggered, showing that static keys cannot fully utilize physical layer features and multi-branch complex feature sequences, resulting in limited security and resistance to attacks.

[0044] This invention achieves a comprehensive effect of high randomness key generation, controllable iterative update, and low bit error rate transmission by using complex neural network feature extraction, multi-scale residual flow, high-dimensional security tensor construction, dynamic key iteration, and DropConnect random masking. This is superior to traditional methods that do not use the improved mechanism, and verifies the feasibility and effectiveness of this invention in multi-channel, high-noise wireless communication environments.

[0045] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A secure communication method based on wireless transmission, characterized in that, Includes the following steps: Collect the data to be transmitted, map it into I / Q complex signals, standardize and frame the complex signals, and generate complex data sequences; The complex data sequence is input into the improved CVNN. A weighted adaptive fusion mechanism based on the complex covariance matrix is ​​established between the phase branch and the amplitude branch to extract the high-order covariance tensor between features and generate a multi-branch complex feature sequence. Multi-scale residual flow is applied to multi-branch complex feature sequences to diffuse the features to three scales: time, frequency, and space, thereby constructing a high-dimensional secure tensor with unidirectional irreversibility. Based on the high-dimensional security tensor, the intrinsic features of the wireless signal and the instantaneous statistical features of the channel are extracted to generate the physical layer security fingerprint, and a security feature matrix is ​​formed by dynamic weighting. An initial dynamic key is generated based on the security feature matrix, and iteratively updated in the local and global ranges through a nonlinear mapping algorithm to generate a hierarchical dynamic key coupled with the instantaneous characteristics of the channel. By combining the hierarchical dynamic key with the complex data sequence, block encryption, integrity verification and redundancy coding are performed to generate encrypted data frames, which are then sent to the receiving end via a wireless channel. The receiving end decrypts, verifies the integrity of, and assesses the security of the encrypted data frames. When the security level is lower than a preset threshold, it sends a trigger signal to the sending end through the feedback link to activate the DropConnect connection random masking mechanism of the improved CVNN hidden layer of the sending end. It iteratively adjusts the non-linear step of the key update frequency and encryption strategy until all data frames are securely transmitted and the final received data set is generated.

2. The secure communication method based on wireless transmission according to claim 1, characterized in that, The process of generating the multi-branch complex feature sequence includes: The complexized data sequence is input into the phase branch and the amplitude branch respectively, and a linear mapping operation is performed on the I component and the Q component to generate a preliminary complex feature vector. The complex covariance matrix is ​​calculated from the initial complex eigenvectors of the phase branch and the amplitude branch to obtain the complex domain correlation weight matrix; Based on the complex domain correlation weight matrix, the complex eigenvectors of the phase branch and the amplitude branch are weighted and linearly fused to generate a fused complex eigenvector. Normalize the fused complex feature vectors to generate normalized complex feature vectors; The normalized complex feature vectors are input into the next CVNN hidden layer to form a multi-branch complex feature sequence; In each training iteration, the fusion weight matrix is ​​updated based on the change in the covariance matrix of the complex feature vector output in the previous round. The complex feature vectors of the phase branch and the amplitude branch are weighted and fused to form the updated fused complex feature vector, which is then input into the next CVNN hidden layer to generate a multi-branch complex feature sequence.

3. The secure communication method based on wireless transmission according to claim 1, characterized in that, The construction process of the high-dimensional security tensor includes: The multi-branch complex feature sequence is divided into fixed-length subsequences according to the time dimension, and a one-dimensional convolution operation is performed on each subsequence to generate a time-scale feature sequence. Perform a short-time Fourier transform on the time-scale feature sequence in the frequency dimension to map the complex features to the spectral representation, and perform a two-dimensional convolution operation on the spectral features to generate a frequency-scale feature sequence. A channel topology mapping is established in the spatial dimension for the frequency scale feature sequence. The complex features of each channel are weighted and superimposed with the complex features of neighboring channels to form a spatial scale feature sequence. The feature sequences at the time, frequency, and spatial scales are respectively concatenated using residuals. Each scale feature is added to the original input feature to form the residual output at each scale. The residual outputs at each scale are then superimposed in the order of time, frequency, and space to generate a high-dimensional complex feature tensor with unidirectional irreversibility.

4. The secure communication method based on wireless transmission according to claim 1, characterized in that, The process of generating the security feature matrix includes: The high-dimensional complex feature tensor is divided into fixed-size sub-blocks according to the three dimensions of time, frequency and space. For each sub-block, the amplitude mean, phase mean and amplitude-phase covariance of the I / Q components are calculated to generate the complex feature vector of the sub-block. Construct a matrix by arranging the complex eigenvectors of all sub-blocks into a channel sequence, calculate the complex cross-correlation matrix between the eigenvectors of each channel and its neighboring channels, and obtain the statistical characteristic matrix between channels; The complex eigenvector of each channel is multiplied element-wise with the corresponding inter-channel statistical feature matrix to form an initial weighted eigenvector sequence; The initial weighted feature vector sequence is accumulated by channel to generate a preliminary security feature matrix; In the first iteration, the zero matrix is ​​used as the security feature matrix of the previous iteration. In subsequent iterations, the updated security feature matrix of the previous iteration is used to calculate the element-wise difference. The difference is multiplied by the channel weighted learning rate of the security feature matrix and accumulated to the corresponding channel weight of the original security feature matrix to form the updated security feature matrix. The updated security feature matrix is ​​normalized by channel, with the sum of the weighting coefficients of all channels kept to 1, and the normalized matrix is ​​output.

5. A secure communication method based on wireless transmission according to claim 1, characterized in that, The generation process of the hierarchical dynamic key includes: The complex eigenvalues ​​corresponding to each channel of the security feature matrix are divided into blocks of fixed length, and the amplitude and phase components are extracted sequentially. The amplitude and phase values ​​are converted into fixed index bits of the key binary sequence through a mapping function. The index bits are determined according to the channel order and the length of the subsequence. The eigenvalues ​​of each channel are mapped to predefined bit positions through modulo operation. All channel mapping results are concatenated to form the initial dynamic key. The initial dynamic key is input into a nonlinear mapping algorithm, which performs nonlinear permutations and XOR operations on subsequences of the key within a local range, and updates the position of the subsequences according to the channel index order to form a locally updated key. Globally, a non-linear cyclic shift operation is performed on the local update key to map the channel amplitude and phase values ​​of the security feature matrix to the corresponding bit positions in the key binary sequence according to the mapping function. The key segment order is adjusted according to the channel order to form the global update key. Repeat the local update and global update operations for several rounds, using the key generated in the previous round as input in each round, until the preset number of iterations is reached, and output the dynamic key after the iteration is completed; The iteratively completed dynamic key is mapped bit-level to the instantaneous characteristics of the channel using a mapping function. The channel amplitude, phase, and delay characteristics are mapped to the corresponding bit positions of the key. The key segmentation order is adjusted according to the channel order to form a hierarchical dynamic key coupled with the instantaneous characteristics of the channel.

6. A secure communication method based on wireless transmission according to claim 1, characterized in that, The process of performing block encryption, integrity verification, and redundancy coding includes: The complex data sequence is divided into fixed-length data blocks. Each data block is XORed bit by bit with the corresponding length of the hierarchical dynamic key to generate an initial encrypted block. Calculate the cyclic redundancy check code for the initial encrypted packets, and append the generated check code to the end of each packet to form a data packet that has been verified. For each data packet that has been verified, perform redundant encoding, copy the data packet or generate redundant bits according to a preset, insert the redundant bits in the order of the packets to generate an encrypted data frame, and each packet can be completely recovered at the corresponding position at the receiving end; The generated encrypted data frames are sent to the receiving end via a wireless channel, and the packet sequence number and timestamp are recorded during the transmission process.

7. A secure communication method based on wireless transmission according to claim 1, characterized in that, The process of decryption, integrity verification, and security assessment includes: Extract each data packet, its additional redundant bits, and CRC checksum from the received encrypted data frame according to the packet sequence number; Each data group is XORed bit by bit with the corresponding hierarchical dynamic key to generate a preliminary decryption bit sequence; The initial decrypted bit sequence is divided modulo-2 using a preset generator polynomial. The CRC remainder is calculated, and the remainder is compared bit by bit with the additional check code to generate an integrity verification matrix. For data groups that pass integrity verification, the amplitude and phase characteristics of the complex I / Q bit sequences are statistically analyzed. The amplitude and phase values ​​of each channel are compared element-by-element with the corresponding security feature matrix to generate a security scoring matrix.

8. A secure communication method based on wireless transmission according to claim 1, characterized in that, The DropConnect connection random masking mechanism specifically includes: The multi-branch complex feature sequence output channel of the improved CVNN hidden layer at the sending end is divided into channels, and each channel is assigned a random masking probability value. The random masking probability value is calculated by a mapping function. Based on the random masking probability, binary random sampling is performed on each channel to generate a masking matrix. The channel with a masking value of 0 has its corresponding output set to zero in this iteration, so that the channel does not participate in the calculation of dynamic key generation in the next iteration. The channel with a masking value of 1 retains its original output. The generated masking matrix is ​​applied to the output channel of the improved CVNN multi-branch complex feature sequence to obtain the hidden layer output after masking adjustment; After each iteration, the random shielding probability is updated based on the difference between the security feature matrix and the security score matrix fed back by the receiver. The channel shielding sampling and zeroing operation are then re-executed to form a new shielding adjustment output. The output of the masked and adjusted hidden layer is used to regenerate the hierarchical dynamic key. This process is repeated iteratively until all data frames are securely transmitted and the final received data set is output.

9. A secure communication system based on wireless transmission, applied to the secure communication method based on wireless transmission as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The data acquisition module is used to acquire the data to be transmitted, map it into I / Q complex signals, complete standardization and frame division, and generate complex data sequences. The complex neural network feature extraction module is used to receive complex data sequences, establish covariance weighted fusion of phase and amplitude branches, and generate multi-branch complex feature sequences; The multi-scale residual flow processing module is used to perform time, frequency and spatial diffusion on multi-branch complex feature sequences to generate high-dimensional secure tensors with unidirectional irreversibility. The security feature generation module is used to extract endogenous features and channel statistical features based on a high-dimensional security tensor, and form a security feature matrix through dynamic weighting. The dynamic key generation module is used to generate an initial dynamic key based on the security feature matrix, and to form a hierarchical dynamic key through local and global nonlinear iterative updates; The encryption processing module is used to combine the hierarchical dynamic key with the complex data sequence, perform block encryption, integrity verification and redundancy coding, and generate encrypted data frames. The receiver security feedback module is used to decrypt encrypted data frames, verify their integrity, and assess their security. When the security level is below a threshold, it triggers the sender's DropConnect blocking mechanism through the feedback link.

Images